Episode Transcript
Transcripts are displayed as originally observed. Some content, including advertisements may have changed.
Use Ctrl + F to search
0:01
Again, if you're not keeping human in the loop, God
0:03
knows what can happen with your
0:05
algorithm. Human antelope? Human
0:07
in the loop. Human in the loop. In the
0:10
loop. I thought you were saying antelope. In
0:12
the loop. I was like, okay, I don't
0:14
understand this part. Okay, humans
0:17
in the loop.
0:19
Gotcha. It's
0:23
actually really hard to say because artificial
0:26
intelligence and neural networks are
0:28
quite often considered black boxes. You
0:31
can't really explain what's
0:33
happening, why they made this decision.
0:38
Because it's really good just to meet motivational
0:41
people. Sometimes you get such a great
0:43
energy boost just talking with someone
0:45
who's passionate about something. It doesn't
0:47
even have to be your field.
0:50
My name is Anastasia Georgievska. I'm
0:52
the CEO and co-founder of my
0:55
startup, which is called Hout.ai. Hout
0:57
is skin in German and hence
1:00
the name. My company is developing
1:03
software as a service for skincare
1:05
field. It's image-based system.
1:07
We think of it as a beauty intelligence system
1:09
that ultimately helps consumers
1:12
all around the world shop better and
1:14
eliminate unnecessary guesswork and anxiety
1:17
when it comes to picking up the right skincare and beauty
1:19
products. A little bit about myself. I'm
1:21
based in Tallinn in Estonia. I'm now 29
1:23
years old. I founded
1:25
this startup when I was 24 as
1:28
a fresh graduate. And yeah,
1:30
I'm super excited about artificial
1:32
intelligence, about the beauty field. I think
1:34
there's a lot to enhance and
1:37
a lot to change in this field. And
1:39
we are working every day to make it better and
1:41
more consumer-friendly.
1:43
And so I guess the website is Hout.ai,
1:47
if anyone wanted to check it out. Yes, that's
1:49
correct. Okay. Well, can you tell
1:51
us a little bit more about this? How you're using AI for
1:54
your company and the skincare? Sure.
1:56
I would say that my company is not only using artificial
1:59
intelligence. artificial intelligence as the only
2:01
tool in our toolbox.
2:04
Generally, the main capability of
2:06
our software is computer vision algorithms.
2:09
And computer vision doesn't necessarily
2:11
mean that algorithm has to be artificial
2:14
intelligence, which usually entails
2:16
using of some of the algorithms like
2:18
neural networks. We also use a lot of classical
2:21
computer vision models to identify
2:23
skin features. So right now we can track
2:26
around 150 facial
2:28
biomarkers to help personalize purchase
2:31
offerings. So the software by itself
2:33
can help
2:34
a consumer capture a picture
2:36
using just a simple smartphone camera.
2:39
You don't need any advanced high-end
2:41
medical devices to do that.
2:43
After the picture is captured, we're sending it to
2:45
our secure cloud where we run our computer
2:48
vision models in parallel and can
2:50
deliver results pretty quick in about
2:52
two, three seconds. And after the results
2:54
are complete, which means that we have built
2:57
your unique skin profile based on your facial
2:59
features, we will be matching skin
3:01
care products with your skin profile.
3:04
And how many people work in your company? And
3:07
can you give us a little bit of an idea as far as
3:09
revenue or amount of customers that you have?
3:11
So our company is a startup stage. At the
3:13
moment, we have 29 persons
3:16
in our team.
3:17
We work with multiple companies everywhere
3:20
around the world. We have around 90 clients,
3:23
which includes skin care manufacturers,
3:25
retailers like Ulta Beauty, and
3:27
also a number of smaller companies like
3:30
Longevity Clinics or Skin Clinics. Pretty
3:32
much businesses that produce services
3:35
or goods that have a factor in skin are
3:37
among our clients.
3:39
And I guess you gave us the number of clients, but
3:41
as far as where they're located, I know you said kind
3:43
of all around the world and you said you're in Estonia.
3:45
So I'm curious where most of those clients are.
3:47
They are US based or can you give us a
3:50
better idea of how you even pitch these clients,
3:52
what you're saying to them? Yeah,
3:53
US market is one of their... It's
3:55
not one of the biggest. It's actually the biggest market
3:58
for skin care products. So of course...
3:59
We're targeting US clients and US
4:02
is a big geography for us. Also Europe
4:04
and Asia Pacific regions, not
4:07
only limited to this, three locations,
4:09
we have clients in Australia, in Colombia,
4:12
in South Africa. So that's why I'm saying
4:14
everywhere around the world. When
4:15
we're pitching to them, I would say that
4:18
pitch is universal. Of course, for a different
4:20
alteration, depending, there
4:22
is a big difference on how we pitch in Asian
4:25
region and in the US. Use
4:27
cases sometimes are also different. But generally,
4:29
I think what makes our
4:31
software easy to understand is that people
4:34
can have very different opinions about beauty,
4:36
a beauty-beauty standard. But what
4:39
is universal is, and
4:41
when I'm saying universal, most population
4:44
will agree that having healthy skin, free
4:46
of imperfections, free of diseases, is
4:49
considered beautiful or at least
4:51
attractive, appealing.
4:52
And the fact that our software helps accurately
4:55
identify what are the skin concerns and
4:57
how to address them makes our pitch easier
4:59
because again, it's something universal, it's
5:02
something that every one of us would like to have.
5:04
And we are pitching to our clients that
5:07
if they start working with us, they
5:09
can personalize offering to
5:11
their clients much better, which obviously
5:14
increases the satisfaction of
5:17
your purchases. It helps you improve
5:19
some short-term metrics like I want
5:21
to sell more, right? But more and
5:24
more players in the beauty field are
5:26
saying that there is a better alternative to
5:28
just transactional relationships when
5:31
a customer comes to your platform and
5:33
you would never see them again, right? It's kind
5:35
of one-time purchase.
5:37
Brands, especially direct-to-consumer brands, whose
5:39
main touch point is online want to
5:41
build relationships with clients.
5:44
We want to build this trust. We want to
5:46
offer education. And this
5:48
is exactly what our software is doing. So it
5:50
helps explain what's wrong with your skin
5:53
and it also explains what products
5:55
it should be using and why.
5:57
And so are you more of like a B2B play
5:59
where you're a help?
5:59
I don't know if you can even release
6:02
some of the clients' names that you kind of give them
6:04
this AI. They can put it on their app
6:06
or whatever. And then if it's a certain brand,
6:09
then these people are like, hey, you should use
6:12
this brand to clear your acne or something
6:14
like that.
6:15
Exactly. So our software is primarily
6:17
B2B. We do, however, have plans
6:19
about launching our own B2C
6:22
app at some time point. I'm not sure this
6:24
is confirmed at the moment. To name
6:26
you some clients, our clients are Ulta
6:29
Beauty. I think if you are listening
6:31
to this podcast from the US, you are pretty familiar
6:33
with this beauty retailer.
6:36
Among manufacturers, one of our clients is Beyersdorf,
6:38
which is the owner of Nivea.
6:40
Yeah, so we help companies
6:42
target the right products to their customers.
6:44
If we're talking about retailer and manufacturers,
6:47
use cases will be slightly different because manufacturers
6:50
have usually particular brands, so
6:53
they would be matching products from
6:55
this brand's assortment to the skin concerns.
6:57
With a retailer, it can be more
7:00
sophisticated because retailers obviously have
7:02
bigger number of brands. Ulta
7:04
has, I think, thousands of brands
7:06
that they are offering. But
7:09
then the idea is that how you can offer
7:11
the right brand and the right products at
7:13
the same time, that also matches the budget, that
7:15
also matches the values of the consumer.
7:18
So do I want to show from the local
7:20
producers? Or maybe I want
7:22
to shop for K-pop skincare? Or
7:25
maybe I want to, I don't
7:27
know, only purchase cruelty-free products? So
7:30
that adds additional layers of complexity and
7:32
at the same time, it offers additional,
7:35
more deep personalization of your purchases.
7:38
That
7:38
makes sense. But how do they decide
7:40
it or you decide it? I guess that's what I'm trying
7:42
to figure out as well, as we kind of dive more
7:44
deep into this AI that you have.
7:46
So we let algorithms decide. We
7:49
use not only computer vision models
7:51
to recognize face features, but we
7:53
also use recommendation engines,
7:56
which utilize both decision
7:58
trees and some of the requirements.
7:59
neural networks, I will explain
8:02
what it basically means. So usually, neural
8:05
network is trained once,
8:07
like once per iteration, so you have
8:10
your training dataset, then
8:12
you have a set of answers
8:14
that you want the system to learn
8:17
how to reproduce. Let's say you have a
8:19
picture and you have an expert grading, maybe
8:21
a dermatologist, who would classify
8:24
what type of acne this is, what
8:26
type of wrinkles these are. So your
8:28
idea is to build a network that
8:31
will reproduce answers
8:33
as if it was a dermatologist. So you kind
8:36
of want to mimic a professional grading.
8:38
But this is not the only type
8:40
of
8:41
artificial intelligence algorithms. So the example
8:43
I just explained is called supervised learning,
8:46
because obviously, there's a supervisor, in this case,
8:48
it's dermatologist. Some networks
8:51
are trained and they function differently. So
8:54
the goal of this networks is not to only
8:56
learn from an expert, but also
8:58
learn from the consequences. So these
9:00
networks are called recurrent. So their
9:03
idea is that you will utilize
9:05
outcomes as input
9:08
data for improvement of their
9:10
accuracy of this algorithm. And we're
9:12
using these networks also to see what products
9:15
were performing better. So whenever
9:17
clients are using our system, they can
9:19
upload their inventory, we will be
9:21
looking at the ingredients, we will
9:23
be looking at our knowledge
9:26
of how different ingredients and combinations
9:28
perform on different persons to then
9:31
match with products. So decisions
9:33
are carried out by the algorithm. But we
9:35
also understand that brands might want
9:37
to shape the recommendations
9:40
as well. So this is where we're also offering
9:42
additional AI based tools to
9:46
make recommendations personalized and
9:48
aligned with what brands is trying
9:50
to achieve. Hope I didn't confuse you too much.
9:53
No,
9:53
that makes sense. I'm just trying to take it in and trying
9:55
to make sure everyone else does too. When you're talking
9:57
about the reoccurring network versus something learning.
9:59
What was the first one that you stated? Yes,
10:02
so it's just different approaches. So it's supervised
10:04
learning. Supervised learning. Okay. Yes.
10:07
Gotcha. That makes sense. I mean, anyone who's
10:09
context-closed, you're like, okay, they're learning from
10:11
someone who's supervising them, right? In this
10:13
case, a dermatologist versus their recurrent
10:16
network. Is that what you said?
10:17
Yes. So if we would compare, you know,
10:20
let's come up as an example, let's say treatment
10:23
of acne. So if we were to
10:25
only use a supervised learning,
10:28
we would want it to detect
10:30
different skin eruptions and maybe
10:33
classify them. So is it like blackhead
10:35
or is it papule or maybe it's post-acne
10:38
inflammation?
10:38
So the network can detect these regions and
10:41
classify them.
10:42
Then what if I want to solve
10:45
a different task? So I want to,
10:47
I know what acne a person has based on
10:49
their skin profile. And I want to
10:52
see how different treatments would
10:54
improve the condition of skin or maybe they
10:57
will not improve. Maybe even skin will get worse,
10:59
right? But I want to extract this learning.
11:02
So what product helped
11:04
to this consumer
11:05
or like didn't help, as I mentioned.
11:08
So if I'm also able
11:11
to fit
11:12
the outcome to the neural network,
11:14
maybe I can remeasure skin in
11:17
two weeks after I started using the product.
11:19
So then the neural network can also learn
11:22
what is the probability that this product
11:25
is suitable for other consumers
11:28
similar to consumer in my
11:30
training set. So in that sense,
11:32
you not only just say what's
11:35
the skin condition, you also try to
11:37
build correlations or
11:39
if you are very lucky causality because
11:42
correlations don't always mean
11:44
that,
11:44
some products will be helpful
11:47
for a particular group of users
11:49
or like consumers with a higher probability
11:51
than other products. So you're
11:53
saying
11:53
based on your acting discussion
11:56
and kind of having those different things, you're saying the
11:59
AI. you're doing the recurrent network,
12:01
is it automatically going to take into someone's age?
12:03
They're like, okay, this person's 80 versus person 16.
12:05
I'm going
12:07
to recommend this because I know they're 16
12:09
versus the 80-year-old might be dealing with wrinkles
12:12
instead. And so it's automatically
12:14
taking that into account as well and learning
12:16
from that.
12:17
It's actually really hard to say
12:20
because you maybe have heard that
12:22
artificial intelligence and neural networks
12:25
are quite often considered black boxes.
12:27
You
12:27
can't really explain in
12:30
most times what's happening,
12:32
why they made this decision. It's one
12:35
of the problems of general
12:37
AI and special AI. So general
12:39
AI, it's kind of a field overall
12:41
right and then special AI is application
12:44
of AI in the particular field.
12:46
But again, you can't always understand why
12:49
it's making the decisions. There are different ways
12:51
how you can interpret using heat
12:53
maps. For example, in computer vision,
12:55
you can try to build attention
12:57
maps. Attention map stands
12:59
for kind of heat maps that
13:02
try to describe what
13:04
pixels in the picture of an algorithm
13:06
considered to be most predictive. Again,
13:09
it's quite often speculative. So
13:11
you can then try to
13:13
kind of run additional tests and understand what
13:15
features made an impact. Making
13:17
algorithms more interpretable, you
13:20
actually always need to have a person in the loop
13:22
to understand if results make
13:25
sense. Depending on their field,
13:28
subject matter experts can tell if results
13:31
make sense, if they're aligned with the general knowledge.
13:33
It's actually one of the reasons why
13:35
here in Europe we have AI
13:38
act that will be adapted, I think,
13:40
in two years from now, that describes
13:43
the procedures of building
13:45
non-harmful AI, safe
13:47
AI. And one of the requirements
13:49
is at every stage of
13:52
development or using of artificial
13:54
intelligence algorithms, you will always have
13:57
a person in the loop, the one who is supervising
13:59
happening
14:01
and also checks that this
14:03
algorithm does make sense.
14:05
For most of us, learning a second language in
14:07
high school or college wasn't exactly
14:09
a high point in our academic careers.
14:12
Like back when I had to take Spanish in high school and
14:14
only remember Hola. But now, thanks
14:16
to Babbel, the language learning app that's sold
14:19
more than 10 million subscriptions, there's
14:21
an addictively fun and easy
14:23
way to learn a new language. Whether you'll be traveling
14:25
abroad, connecting in a deeper way
14:28
with your family, or you just have some
14:30
free time, Babbel teaches bite-sized
14:32
language lessons that you'll actually use
14:35
in the real world. And guess what? I chose
14:37
to brush up on my Spanish so not only could
14:39
I learn Hello, but I could also learn Adios.
14:42
See, with Babbel's 15-minute lessons, they
14:44
make it the perfect way to learn a new language
14:46
on the go. Babbel's expertly crafted
14:49
lessons are built around real life. You
14:51
learn how to have practical conversations about
14:53
travel, relationships, business, and
14:56
more. Most language learning apps use AI
14:58
for their lesson plans, but Babbel lessons
15:01
were created by over 100 language experts.
15:03
Their teaching method has been
15:05
scientifically proven to be
15:07
effective. With Babbel, you can choose from 14
15:10
different languages, including Spanish,
15:12
French, Italian, and German. Plus,
15:14
Babbel's speech recognition technology
15:17
helps you to improve your pronunciation
15:19
and accents. There are so many ways
15:21
to learn with Babbel. In addition to lessons, you
15:23
can access podcasts, games, videos,
15:26
stories, and even live classes.
15:29
Plus, it comes with a 20-day money-back
15:32
guarantee. Start your new language learning
15:35
journey today with Babbel. Right now,
15:37
get up to 55% off your subscription
15:39
when you go to babbel.com. That's
15:43
babbel.com. For
15:47
up to 55% off your subscription. Babbel.
15:50
Language for life.
15:53
I'm here with John Austinson. How are you doing
15:55
today, John? Hey, Austin. Doing great.
15:57
Thanks for having me. Absolutely. podcast
16:00
and I interviewed John on episode 250
16:03
of this very podcast. So you can hear more about John's
16:05
story and how he grew Fran Bridge consulting
16:07
right here. But in the meantime, would you mind reminding
16:10
our listeners what you do and what you could
16:12
potentially help them with? Yeah. You know, we work
16:14
with entrepreneurs and investors across the country,
16:16
helping them get into business ownership through franchising.
16:19
And when I say franchising, you likely think fast
16:21
food. And yet there's so many other industries out there
16:24
from home and property services to health
16:26
and wellness, from kids, pets, the aging
16:28
population, oil changes, all
16:29
of these understandable cash flowing businesses
16:32
that oftentimes are recession resistant. And 90%
16:35
of our clients end up purchasing an opportunity they
16:37
never thought about. We work with the largest brokerage
16:39
in the country, over 600 different franchise companies.
16:42
Having been a franchisor and a franchisee myself,
16:44
I'm very picky about which ones that we show to our clients.
16:47
Only the best of the best. The great thing, Austin
16:49
is it's entirely free to work with us. We're
16:51
funded by the companies very much like an executive
16:53
search type model. So our clients never pay us in nickel
16:56
and we do more deals for our clients than anybody
16:58
else in the country.
16:59
And what does a typical client look like for you?
17:01
Two thirds of our clients would be looking to keep
17:04
their day job. They're looking to get into business ownership,
17:06
maybe as a side hustle, or maybe they're already a business
17:08
owner and they can't get their full attention. We work with doctors,
17:11
lawyers, existing business owners, corporate
17:13
executives, really a wide array of
17:15
backgrounds all around the country.
17:17
As far as anyone who might be interested in your
17:19
service, is there a best way for them to reach you?
17:22
Yeah. Come out to our website, franbridgeconsulting.com.
17:24
That's F-R-A-N-Bridge-Consulting.com.
17:27
For all of your listeners, Austin, we'll also
17:29
send them a copy of our new book, either audio
17:31
or PDF version, or they can purchase it on Amazon.
17:34
But I would love to share that. Our book is called Non-Food
17:36
Franchise and we've gotten great feedback
17:38
since its release. If you're interested in taking
17:40
a next step, let my assistant Ashley know
17:43
and she'll schedule a call and we'll
17:45
discuss your situation and what could be a good fit.
17:47
Yeah. And I know you've already scheduled a few call with our
17:49
listeners. Could you just tell them what that typically
17:51
is like, like how long and if it's free for
17:53
them to do? Yeah, we've had a great response from your
17:55
listeners entirely free because
17:58
of the caliber of folks that we work with.
17:59
cut to the chase. We usually spend 20 to 30
18:02
minutes on that first call. And then as the next
18:04
step, that following week will come to them with
18:06
opportunities, usually around 10 or so in
18:08
their market. They're available to check all the
18:10
boxes. And we talk them through those and
18:13
then make introductions to the ones that seem
18:15
most intriguing to them.
18:16
Well, that sounds awesome. And again, if someone was interested
18:18
in scheduling a call, where's the best place for them to
18:21
go ahead and sign up? Yeah, come out to our website, franbridgeconsulting.com,
18:24
F-R-A-N bridgeconsulting.com, and
18:27
we would love to engage.
18:30
I mean, that makes perfect sense, it seems like
18:32
to me, especially mathematical
18:34
people. You're like, I need to know why it chose
18:36
whatever it chose, but you're saying that's not possible.
18:39
I mean, it's not always possible. Or
18:41
sometimes you can only understand half
18:43
of facts.
18:44
Primarily why you cannot
18:47
always understand how AI
18:49
is making its decisions is because
18:52
usually, like let's compare some previous
18:54
methods that existed before artificial intelligence.
18:57
And let's talk about image recognition.
19:00
So before artificial intelligence, you
19:02
usually would be crafting features
19:05
yourself, features that you want to
19:07
identify from the picture. And you say,
19:10
I
19:10
introduced this feature, this is how I describe
19:12
it, and this is how I want
19:15
to measure that.
19:16
And in that sense, you kind of give instructions
19:18
to algorithms which you implement in
19:20
a code,
19:21
and it works, hopefully, according
19:23
to your protocol, right? If you didn't make
19:25
any mistakes or it's just possible to
19:27
launch this algorithm. Then if I
19:29
think about artificial intelligence, quite
19:32
often they say,
19:33
okay, we should admit the fact that as
19:35
humans, our perception is also limited,
19:38
right? So what if we tell
19:40
a smart system that
19:42
is designed in a similar
19:44
way as our brains, and
19:47
we tell the system that the system
19:49
is not limited to choosing the
19:51
features, to choosing how it
19:54
will be solving the task. Let's kind of
19:56
give it full freedom.
19:58
So if you do that... algorithm
20:00
starts to learn dependencies in
20:03
the data to run its predictions.
20:05
And the fact why it's sometimes
20:08
so hard to discover what
20:11
features were used or like how
20:13
AI made this decision is because
20:15
we don't always understand the complex
20:18
dependencies in the data. This is why
20:20
they use this algorithms because we want to consolidate
20:23
knowledge
20:24
from the data because we cannot do it
20:26
ourselves because you need a lot more
20:28
computational resources.
20:30
And humans are very smart and this is why
20:32
we have this gut feeling, but just machines
20:34
can do it in a more structured way. So because
20:37
these dependencies are complex, again,
20:39
quite often it's hard to understand
20:42
why algorithm made one decision over another.
20:44
But maybe I'd like to talk more about how
20:47
we ensure that our AI makes
20:49
sense.
20:50
We're using the principles of human-in-the-loop.
20:53
So whenever we are training algorithms, it's supervised
20:55
by dermatologist or skin experts, and
20:57
then we're releasing our models. We're
21:00
testing them on retrospective data. We're
21:02
testing them on the real data coming
21:04
from consumers, and then we're ensuring that it works
21:06
correctly.
21:07
Again, if you're not keeping human-in-the-loop, God
21:09
knows what can happen with your algorithm.
21:12
Human-antelope? Human-in-the-loop.
21:14
Human-in-the-loop. In the loop. Oh, I thought you
21:16
were saying antelope. I
21:19
was like, okay, I don't understand this
21:21
part. Okay. Humans-in-the-loop. Gotcha.
21:24
Human-in-the-loop. Yes. Okay. So all
21:26
your AI has that right now, you're saying, right? Yes.
21:28
You're not doing the recurrent network. Where
21:30
do recurrent networks as well? Don't get me
21:33
wrong. We're using different techniques, and I
21:35
would say we're much more
21:37
existing and constantly developing, whereas
21:40
in your field of transformers, there
21:42
are generative adversarial networks.
21:45
I would say that one need to understand,
21:48
and again, I'm not a super-technical person,
21:50
write my backgrounds in biophysics, not in
21:52
software engineering. It's very important
21:54
to understand maths behind
21:57
AI, how it works, so that you
21:59
know what a limit is.
21:59
limitations, so you know what are the traps
22:02
and where you have to be extra conscious.
22:04
And then you just have a very rich toolbox
22:07
and you can choose the sharpest tool depending
22:09
on what task you need to solve.
22:11
But even if we went back to the, remember the two
22:13
things that you talked about, the supervised learning network versus
22:16
the recurrent network, you could run
22:18
AI on both of those and then kind
22:20
of decide, right? Someone can, I mean, I guess it is
22:22
supervised at the end. If you just look what the recurrent
22:25
network would have picked versus the supervised learning
22:27
network would have picked as far as like skin condition
22:29
and everything.
22:30
So technically recurrent network also
22:32
supervised because you have kind of a target.
22:35
It's just different approaches with recurrent networks
22:37
can also learn from experience. And
22:40
then, for example, a neural network,
22:42
which is trained for segmentation task to
22:45
segment objects in the picture, wouldn't
22:48
have kind of learning because just know the fact there
22:50
is a hyper-pigmentation spot and the
22:52
network detected that. But if you, for
22:55
example, apply some cream and then
22:57
you want to say, look, this is what happened after
23:00
a month of using this cream. This is how
23:02
my skin metrics looked before
23:04
a month ago. This is how they look right now.
23:07
So if you use recurrent networks, you can utilize
23:09
this experience as a learning to
23:12
then be able to predict
23:14
better recommendations in the future
23:17
using that learning from experience.
23:19
But what would you classify that as? Would
23:21
you still classify that as a recurrent network?
23:24
So recurrent network, it's kind of the network
23:27
architecture. It's an approach,
23:29
but you can use. You use primarily with networks
23:32
for learning from experience. I
23:34
guess we can dive more into this AI if that's
23:36
okay. I was just far just understanding the basics because
23:39
that was my thing when you said you didn't know why the
23:41
AI kind of chose it. That kind of boggled my mind
23:43
because the whole idea with even blockchain,
23:46
right? Like I'm not in the crypto or blockchain or anything
23:48
like that. But the whole idea with the blockchain is that you
23:50
can see every transaction that was made
23:52
and is tied to this basically for life,
23:54
right? So we know why this happened or
23:57
whatever occurred with this transaction from
23:59
my understanding. Again, I don't mess around with blockchain
24:01
or anything like that. But then with this recurrent
24:04
network, you didn't necessarily know, which is disturbing,
24:06
it seems like.
24:07
Yes, that's correct. You don't necessarily know how
24:09
it's making decisions. And it's
24:12
again, one of the big, I would say, community
24:14
concerns when it comes to artificial intelligence,
24:17
but it's not always interpretable.
24:20
And I would say, as far as I know,
24:22
it's never 100% interpretable.
24:25
So this, of course, raises concerns, right? It
24:27
offers lots of benefits,
24:29
but
24:30
we always need to make a step
24:32
back, maybe turn down our excitement
24:35
and think, you know, what's really happening?
24:37
So what are the benefits and what are the negatives?
24:39
So the benefits are the
24:42
fact that you can learn from massive
24:44
amounts of data. So you can, again,
24:47
uncover some dependencies that you wouldn't
24:49
do if you were not using this
24:51
big data approach, then a lot
24:54
of tasks can be automated. For
24:57
some specific fields, you can
24:59
achieve quite high accuracy very
25:02
fast. For example, in computer
25:04
vision, in natural language processing,
25:07
you can take some of the models open source
25:09
available on GitHub, you know, with their free
25:11
license, and you can train it on your
25:14
data and you can get more
25:16
than 80% accuracy in a
25:18
matter of like week just playing around.
25:20
So if you maybe spend additional month,
25:22
you can achieve accuracy for
25:25
specific tasks in a month.
25:27
Then what happens after that is achieving 96%
25:30
takes more time, but then going
25:32
from 96% to 99%
25:35
will be super, super hard,
25:37
right? So I would say that automation of some
25:39
stereo tasks can be done easily.
25:41
Then if you have been watching what's happening
25:43
in generative AI, now,
25:46
Dali, a project by opening AI can generate
25:48
images just from the textual input, which
25:51
is considered to be a
25:52
big breakthrough in the creativity
25:55
because we can generate some concepts
25:57
or some ideas because it's considered
25:59
to be.
25:59
very helpful for art
26:02
people, designers or artists
26:05
or
26:06
even social media managers. So you
26:08
can go to Dali and generate a picture
26:10
which will be unique and you can use it with a
26:13
copyright on your social platforms. So
26:15
then it opens up a lot of time
26:18
for being creative and then there's a tool
26:20
that can carry out the boring work,
26:22
if you will, or the hard work of actually
26:25
producing an art object.
26:27
Then OpenAI, Chagigipi
26:29
also generate a lot of interest around
26:31
that with being able to reproduce
26:34
the human conversation. So that
26:36
even save up time, you can use it for maybe
26:39
write some email with you didn't find
26:41
enough time to write in the whole week and
26:43
then within five minutes you can get your
26:46
email follow up.
26:47
So that frees time for
26:50
being creative, for carrying out
26:52
the tasks that artificial intelligence
26:54
cannot do, which requires human
26:57
involvement
26:58
and a lot of just daily things. We don't think
27:00
about that, but a lot of things are
27:02
AI based like our internet
27:05
searches, selection of the music,
27:07
maybe things we don't interact daily,
27:09
but in some warehouses you're
27:12
using KI for logistics, for
27:14
supply chain, for managing
27:16
the warehouses.
27:18
It's becoming more and more part
27:20
of our daily life. So these are obviously the benefits,
27:22
some tasks which took more time
27:25
before you can now do faster.
27:27
The disadvantages, I would say
27:29
that the only concern is for me as a
27:31
person who's working in the AI field is the
27:33
fact that you don't always
27:35
understand 100% why it made its
27:38
decisions. So this is why you need to ensure
27:40
you have the right system to monitor biases,
27:43
to detect any anomalies.
27:45
So you kind of have to be conscious about how you're
27:48
using that.
27:48
Other disadvantages that AI still
27:51
requires a lot of computational resources,
27:53
a lot of electricity, a lot of power,
27:55
a lot of GPUs to be trained,
27:57
which
27:58
I would say people
27:59
can argue. but that's not necessarily
28:01
really environmentally friendly because, you
28:03
know, electricity also means CO2
28:06
emission, right? You know, that's just what
28:08
comes up from the top of my head.
28:10
You're doing a good job. I mean, one thing I was writing
28:13
down, I was trying to figure out, do you consider Google
28:15
Maps AI? What do you mean Google
28:17
Maps AI? Yeah, yeah. Well, say if I want
28:19
to drive somewhere eight hours away
28:21
and I put it into my phone
28:23
and I put the address, is that considered artificial
28:26
intelligence or no?
28:27
So it's not necessarily what I was talking
28:29
is like Google Maps AI, but rather
28:32
when you're browsing the net, you have a very specific
28:34
query, you want to find
28:36
the hits and then what AI does, it filters
28:39
out the hits that you unlikely
28:42
need to make sure
28:44
that your search results are relevant
28:47
to what you were trying to achieve. So a
28:49
lot of these algorithms are also artificial intelligence
28:51
based.
28:52
Well, then I think Google Maps would have to be too. Just
28:54
thinking that way. Or I'm sure you've used
28:57
Apple Maps or whatever you use to getting where because it's
28:59
just taking how many people are put in that address,
29:01
right? Maybe at first when I don't even know
29:03
how long has AI been around?
29:05
So I would say that if you look
29:07
back into the history, AI as a
29:09
concept and neural networks has been around since
29:12
around 1980s, it was
29:14
already there.
29:15
So there is a very good book by Exeo
29:17
of Google China, which is called AI
29:19
Superpowers.
29:21
And this book,
29:22
the author talks about how AI
29:24
emerged and what enabled
29:26
it big breakthroughs.
29:28
When you said that Google search, so we're actually
29:30
all using AI, it's just the different versions
29:32
of it, kind of is what you were saying, like the
29:35
Superpowers or Nucleus Recurrent.
29:36
Yeah, I would say there is AI
29:38
that everyone thinks of AI when
29:40
we're talking right now, right? So
29:43
some modern techniques. And as I mentioned
29:45
this book, it's actually rolled by Kai-Fully, which
29:47
is a very famous thing tank in
29:50
AI space. And what
29:51
is it called again? Just so everyone can get that. So
29:53
the book is called AI Superpowers
29:55
and the author is Kai-Fully. So
29:58
there, let's say modern AI, you know.
29:59
true branding
30:01
happened in around 2015
30:04
or 2016 because GPUs became more available.
30:07
Also, a lot of developments from NVDA
30:09
came around and it became accessible
30:11
not only by super big corporations,
30:13
but for example, back in 2016,
30:16
we as a startup could use
30:18
artificial intelligence and GPUs to
30:20
train our first models. So I would say that
30:23
around 2016 was a time
30:25
when high-scale adoption
30:27
of AI began.
30:29
And that's when I started hearing, and again, just
30:31
so everyone knows that the author's name is K-A-I-F-U.
30:35
That's the first name and last name is L-E-E.
30:38
And we'll have it in the show notes if anyone's interested
30:40
in this book. But I think that's when I started hearing
30:42
about, well, I know it's funny that they're talking about AI
30:44
superpowers in China with this book, but that's
30:47
really when they started doing facial recognition. I feel
30:49
like I started hearing about it a few years ago, so obviously
30:51
they started doing it years before that. So it's
30:54
funny that you're saying AI 2015, 2016, when
30:56
it really started taking off, if you will.
30:58
Yeah, they obviously started doing
31:01
that earlier, maybe as R&D, a proof of concept.
31:03
From my experience, it really takes
31:06
time between R&D and leveraging
31:09
something to a real system. There's
31:11
a very big difference between just lab
31:14
prototype.
31:15
There's that joke around developers, which
31:17
is like, works on my machine, right? So how
31:19
to make sure that something you have developed
31:21
for R&D works in real life.
31:23
You're sometimes surprised. You think something will work
31:25
perfectly and then it fails. And this is same
31:28
as with AI, right? It's not super
31:30
resilient when it comes to change
31:33
in the input data. So sometimes real
31:35
life welcomes you as kind of, you understand
31:37
it's a failure, right? You have to start from scratch. It's
31:39
not working.
31:40
Thank you for giving me the history on it, at least
31:42
as simply as possibly as you could. But I was curious, going
31:45
back to your company, do you mind if we jump back to that?
31:47
Absolutely. I like talking about AI, and
31:49
I really like my startup. So I like talking about
31:51
that as well. So how do you actually make
31:54
money with these searches
31:56
that you're doing, or if you're making these app
31:58
for these brands or whatever? I don't know if they're
31:59
different types of revenue? Or can you tell us
32:02
how you figure out how to charge people for that
32:04
and actually make money doing this? Of
32:06
course, it's not something that we have developed from
32:08
the very beginning. It was a journey as
32:11
all the startups. Right now we're using
32:13
subscription model. So it's SaaS software, you can
32:16
subscribe to the software. We're charging
32:18
you for usage depending on how many
32:20
pictures you're processing.
32:22
At the beginning, I would say we
32:24
started as two scientists. So myself, a microfounder,
32:27
we have scientific background, myself
32:29
and biophysics and bioinformatics and microfounders
32:32
background is in theoretical physics.
32:34
Before we even started to sell something,
32:36
we wanted to build something that people want.
32:39
So that was the first thing. We launched two
32:41
beauty apps back in 2016
32:43
when we just wanted to understand if
32:45
someone will be interested in analyzing
32:48
their skin. With the launch of these apps,
32:50
we got first users, they were
32:52
really excited, they were waiting for the result
32:54
and we saw yes, there is something in that. It looks
32:56
like consumers want to understand what's happening with their
32:59
skin and they want to also
33:01
find a secret source how to make
33:03
their skin better. This also was a time when
33:05
we were approached by our first
33:08
partners in their beauty field skincare
33:10
companies
33:11
who wanted to bridge the gap
33:13
between marketing and R&D
33:14
because quite often marketing builds
33:17
up its messages based on
33:19
what R&D is able to achieve. So
33:22
whether these are better formulation or
33:24
maybe better packaging or recorded
33:26
results for particular demographics.
33:29
So all of that is used by marketing to
33:31
kind of address right consumers and
33:34
market their products to the right groups.
33:36
We realized that there is scarcity
33:40
of selfie based, like something simple
33:42
because skin is analyzed in very different
33:44
ways. Quite often it requires a lot of expensive
33:47
equipment, different type of microscopy
33:49
methods that we can cost up to a million.
33:52
You can't put it into every point of
33:54
sales, you can't leverage to every
33:56
consumer so you need something which is more affordable
33:59
and this is more of a business.
33:59
So I would say it was quite
34:02
right time for us in 2016 because AI
34:04
started to become more and more popular, but also
34:06
consumer devices as smart homes
34:09
became more affordable. So I like
34:11
to think about that right now when
34:13
our brands or like our clients want
34:15
to start using our system, they sometimes
34:18
don't understand that the most important hardware
34:20
investment has already been made
34:22
by their clients and it's in their pocket. And
34:24
this is where smartphone.
34:26
So we thought, yes, we need to go into
34:29
smartphone software.
34:31
We, of course, like developed multiple methods
34:33
for this different derma scans and we still wanted
34:35
to utilize our previous R&D.
34:38
And one of our biggest breakthroughs technically
34:40
was again to translate
34:43
clinical methods into the fall. It took
34:45
us a long time, around two and a half years,
34:47
but we finally got there. So we developed software
34:49
for selfie. We thought, OK,
34:52
you can buy that now. Let's go to the market
34:54
and let's offer it to the brands. Let's offer it to potential customers. So
34:58
it turned out that it's quite hard
35:00
to sell something
35:01
that you cannot test easily, but
35:03
you cannot start using easily because it takes
35:05
like especially bigger companies more time
35:08
just to get on track for
35:10
them to experiment. It requires, I
35:12
don't know, six people at least to
35:14
coordinate, especially if it's a big company. So
35:17
you really need to have a solution which is
35:19
easy to use, fast to use, and
35:22
something that they can play around.
35:23
So this is when we started to go into API business. So it says
35:25
business. We put our algorithms
35:27
into the cloud and every client could send
35:30
requests to our cloud, send their data and
35:32
get it back annotated with a skin profile
35:35
or maybe user clustering.
35:38
So something that's accessible everywhere
35:40
around the world. So we started to sell
35:42
the software to bigger corporations, more
35:45
kind of enterprise type of deals. But
35:48
then we saw that there is a big opportunity on
35:50
the market because beauty market is huge. Beauty
35:52
personal care, I think, is
35:54
now estimated around 580 billion US dollars.
35:58
It's growing quite fast. There
36:00
is additional market for
36:02
AI and beauty, which is
36:04
valid around $4 billion. So
36:08
the market is huge and there's a lot of players there.
36:10
But
36:11
you can't sell to enterprise
36:13
in the same way
36:14
as you would be selling to SMEs. So
36:17
we tried selling solution as API
36:19
to SMEs. It turned out they
36:21
can't use API.
36:22
Very few SMEs have technical
36:25
resources to take your
36:27
API, which still requires coding
36:30
converse side, developing backend, developing
36:32
frontend, connecting it
36:34
with your cloud. For most
36:37
SMEs, this is not achievable.
36:39
So then we thought, okay, it looks like
36:41
there is an opportunity, but our product does not fit
36:43
for this big opportunity
36:45
of clients. And we started to think how we can
36:48
do it better. And this is when we came up with
36:50
an idea of WordPress for
36:52
skincare. So it's a widget that
36:54
you can put on your website, whether it is Shopify
36:57
or WooCommerce or BigCommerce, and
36:59
start recommending products from this widget, which
37:01
you can customize using online portal.
37:04
You don't need to make any developments. It's
37:06
kind of zero code, zero
37:08
development product, but you can get all
37:10
of the benefits of our artificial intelligence system.
37:13
So this is when we came to the idea that
37:16
technology is very important. If
37:18
you have bad technology, no one will buy your solution.
37:20
But it's not the only thing which
37:23
your clients will base their decision on. You
37:25
need to have a technology that you can sell.
37:28
And you also need to have a technology that someone can
37:30
buy. It means that they actually can use it.
37:33
And you doing the ease of use thing where
37:35
you're saying you made a WordPress widget or whatever,
37:38
versus the AI. And when you're saying SME,
37:40
you're just talking about small medium enterprise in case
37:43
anyone didn't know. So that's most small
37:45
businesses. It makes sense they don't have the people
37:47
in the background to connect to your AI, but by
37:50
you making a widget or something simpler for them
37:52
that they can just put on their WordPress site, which I feel
37:54
like most small businesses use now.
37:56
That made it much more open, I guess, to you getting more clients,
37:59
huh?
37:59
It's actually enabled us quite good growth
38:02
last year. So we tripled our number of
38:04
clients during 2022.
38:06
Yeah, that makes total sense to me. Cause like beforehand,
38:08
when you're talking about, even I understand being
38:11
able to hook up to an API, but it's not
38:13
easy. Like if you use, for instance,
38:15
I use Cloudflare on my website,
38:17
or if you use Zapier or something,
38:20
it's like, if they're connecting it, it's because usually it has
38:22
an open API and then you have to get a code from there,
38:24
put it in there, make sure they connect and then
38:26
things can change on those platforms. So I
38:29
definitely see how that made it much easier for
38:31
people like me to be like, okay, I can use this if
38:33
it's an app versus me trying to get
38:35
someone to code it up through Upwork
38:36
and connect with your features and see
38:38
what I can do. Just to make it as basic as we
38:40
could. And it seems like that was the big play that helped
38:43
you out a lot.
38:44
Precisely. And also we need to think of support,
38:46
right? As you mentioned, APIs tend
38:48
to change from time to time because, you know, some
38:51
libraries of,
38:52
for example, Python or like Java languages
38:55
quite often used for backend development or like
38:57
for frontend development, different languages.
39:00
The code evolves. It's getting updated.
39:03
So if you didn't update
39:05
your code, it's likely that your API
39:07
can also break. And small companies
39:10
don't have those resources that maybe
39:12
can pay someone for one time set
39:14
up integration fee. But it's quite
39:16
hard for them to invest in
39:19
the support as well. So this is where I say
39:21
we suggest to use our widget for
39:23
your commerce. And we make
39:26
sure that we are responsible for supporting
39:28
that for you. And you just continue
39:31
using that and building better relationships
39:33
with your clients. Yeah.
39:34
I do want to come back to that. Let's come
39:36
back to how you decided to do that and
39:39
came up with that idea. But you kind of
39:41
briefly talked about it as far as your age
39:43
today, your 30, correct? Or 29?
39:45
Yes, almost 30, 29. Yeah.
39:48
So when you came out of college, you
39:50
were 23 when you were thinking about this, right? This
39:52
was your first thing that you started doing
39:54
right out of college. You didn't have another job or
39:56
anything like that. You just started working
39:58
with AI right away.
40:00
I actually did have a job as bioinformatician. I
40:02
was working as research associate
40:05
in a company but was performing drug
40:07
discovery using all sort of visual
40:09
intelligence, but my work was primarily
40:11
focused on analysis of different
40:13
biological samples data. So not
40:16
exactly what I'm doing right now. I got
40:18
this job as a part of, I think, some internship
40:21
from university and I didn't have any savings.
40:23
I didn't have,
40:24
honestly, any kind of plan for
40:26
my life. I was scientist by
40:29
training, so I was thinking maybe I should
40:31
be pursuing an academic career as
40:34
most of my friends at university
40:36
did, get a PhD and then after
40:38
postdoc position and continue being a scientist.
40:41
The problem that I had in
40:43
my university years is I honestly
40:45
didn't understand, you know, what the opportunities existed
40:48
except for
40:49
scientific career.
40:51
The education I had was great, it was fundamental,
40:54
but with nothing about entrepreneurship,
40:56
right? You knew nothing about how you
40:59
can utilize your skills. The best
41:01
opportunity that you could think outside
41:03
of academic career would be working for pharma,
41:06
big pharma because, you know, skills in bioinformatics
41:08
are kind of a good match for them, but that's
41:11
it. But at one of the competitions
41:13
for bioinformatics where I also met my co-founder,
41:16
Konstantin, I also met Alex
41:18
Zabrunkov, who is the CEO
41:20
of Insilica Medicine, which is a unicorn
41:22
drug discovery company,
41:24
who became my mentor.
41:25
So he did a lot for me, I
41:27
think he taught me everything about business. He
41:30
always believed in myself, he
41:32
was kind of promoting me
41:34
as a female founder and just telling me, you
41:36
know, what it takes to kind of
41:38
run a business and I should explore the idea
41:41
of building this skincare startup,
41:43
right? Then after met Alex, I also started
41:45
to get into that startup
41:48
life, right? You start to meet founders, you
41:50
discuss peer to peer, what's happening, how
41:52
you are trying to troubleshoot and
41:55
make sure, you know, your startup can
41:57
continue existing, right? Because you don't have any
41:59
money.
41:59
At the beginning, you don't have any investment,
42:02
you don't know if your idea will work or not, if someone
42:04
will buy that, how your life will go, or maybe
42:07
you should stop doing that and get back to this big pharma.
42:09
So I would say what was quite messy, but life,
42:12
especially of a startup life, it's kind of a marathon,
42:15
it's not a sprint. So you kind of just continue
42:17
doing everyday, evolving
42:20
your product, evolving your technology. And
42:22
eventually, if technology is good, you
42:25
start to understand how it can be applied, what
42:27
could be the use cases, you expand your network,
42:29
start to
42:29
meet not only kind of foreign people, but you start
42:32
to build business people. And if
42:34
you happen to find good ones, they will help
42:36
you find this direction. So that's why I
42:38
think it's so important if you are
42:41
coming from scientific backgrounds and
42:43
don't necessarily have training in business. It's
42:46
very important to find your team,
42:48
it's important to build a network, just finding
42:51
people with different opinions, with different skill sets,
42:53
discussing your ideas, getting
42:56
feedback, ideally finding a mentor.
42:58
If you can find a mentor who
42:59
is a stupid star in your field, that's amazing.
43:02
That's not always happening, right? Start
43:04
finding experts, start finding
43:07
people who have good reputation,
43:09
and you can figure out different ways
43:11
how to partner with them. But it's very important for
43:13
scientists not to be inside of their bubble,
43:15
right? It's very important to understand that technology
43:18
is great, but it's not the
43:20
only component of a startup, right? There
43:23
should be more to that, like
43:25
product strategy. And if
43:27
you don't feel like you can do it by yourself,
43:30
you should be looking for people who
43:32
can join your team and help you bring
43:35
your startup to where you see the
43:38
right fit for that and what's your vision.
43:40
But how about for your particular startup?
43:43
I mean, did you stop doing your job
43:45
that you said you had right out of college and then go
43:47
all in on this? Or was it kind of part-time as far
43:49
as trying to figure out your timeline of how you
43:51
grew your company to where it is
43:54
today?
43:54
So at the beginning, I had like my day job,
43:57
which was an office job, and then at night I would
43:59
be...
44:00
running my startup or on the weekends. So I kind
44:02
of first years, I think I had two jobs
44:04
because already graduated. So I had
44:06
to earn some money as well. The startup didn't
44:08
earn any money the first years. So then
44:11
we started to get first clients. It
44:13
became obvious, you know, but you have to choose. And
44:15
the choice was obvious for me, right?
44:17
I was super excited about what I'm doing. I
44:20
couldn't imagine my life in a different way.
44:22
And this is why I didn't hesitate to just
44:24
dedicate 100% of my time to my startup.
44:27
But when you decided to go 100% in,
44:29
were you making money on it? Yes, I
44:31
was making money, not a lot, but you
44:33
know, it was enough to cover my base
44:36
cost of living, which again, was not very high because
44:38
I was just a student.
44:39
So I was making money, not a lot, but
44:41
it was good enough for me to continue working
44:44
on buildings, the company. Well,
44:45
can you take us over the journey of the
44:48
first couple of years? I mean, cause I know you said today
44:50
you have 28 employees, but it's just curious of
44:52
the growth of your company. What other highlights
44:54
you've had over the last six or seven
44:56
years as far as building this thing?
44:58
Sure. So at first, where's
45:00
my son, when I make a founder working on
45:02
the startup, we started to get some
45:05
contracts from clients and the complexity
45:07
of his contracts.
45:09
It was getting higher. So we understood
45:11
we cannot do it ourselves. So my co-founder
45:13
said, I know a guy, I studied
45:15
with him also for critical physics and
45:18
he's great, but he works in a bank and
45:21
he's data science leader, right?
45:23
And we kind of have very little money.
45:25
I don't understand what we can offer to him, but
45:27
you know, we have very cool projects, have really cool
45:29
technology. We have interesting use case. So
45:32
he said, you know, I will probably go and have some
45:34
beers with him, you know, and just maybe ask
45:36
if he can maybe help us part-time. Maybe
45:39
we can have a small hackathon to make
45:41
a boost to our technology. So my co-founder
45:44
went to have some beers with that guy. And that
45:46
was one of our first team plays because he
45:48
was also working at his full-time job
45:51
for some time and then eventually in the year he joined
45:53
us. So it became free of us. We started
45:55
to look for people in our network, people
45:58
we studied this also like. by
46:00
physicists joined us, also some people
46:02
from physical faculty. Then we went
46:04
on hackathons to look for smart
46:07
people and motivate it because you have to be smart
46:09
and motivated to kind of dedicate your
46:11
weekends for hackathon. Hackathon
46:14
for those who have never visited
46:16
one is a programming competition. So it's
46:18
coming from towards hack and
46:21
marathon. So it's hackathon. So we
46:23
went to this programming competitions
46:25
to look for people. We were kind of pitching them what
46:27
we're doing. Everyone got excited.
46:30
What was AI? It's medical.
46:33
It's like health. It's wellness.
46:35
It has social impact. So
46:37
to some extent, when you are
46:39
looking for people, when you're trying to hire,
46:42
you are selling them the
46:44
idea of your company, especially when
46:47
you cannot offer them strong
46:49
incentives like high salary or fancy
46:52
laptops or a fancy office,
46:56
you have to be patient as a founder. At
46:58
least one of you has to be. And this
47:00
is how our first of our hiring happened. Maybe
47:02
until we were 10 people, when we were 10 people,
47:05
we already started to make some money,
47:07
but we could pay salaries and then we
47:09
could even find recruiters or
47:12
afford a recruiter who would be looking for
47:14
more team members to us. But
47:16
again, we still were a small startup. So
47:19
until now, we still
47:21
take a lot of efforts to hire people.
47:24
It's important to hire
47:26
people who
47:27
have the same attitude as you
47:29
are, who are passionate about what you're doing,
47:32
especially when you're a small company.
47:34
I don't mean like people have to be super
47:36
excited in a way that some people
47:38
just do their job right. Everyone has different motivation.
47:40
Sometimes it's what you're doing. Sometimes
47:43
it's the impact. Sometimes it's money.
47:45
But my personal thoughts is
47:47
if you're building a startup and money is the only
47:50
motivation, it's not good. I
47:52
don't know how it functions in bigger corporations.
47:55
I think when you have an already very
47:57
well-established managerial structure,
48:00
managers, KPIs, systematic
48:03
performance reviews, then maybe you
48:05
can have more people who are just motivated
48:07
by money.
48:08
But I think in a startup, it's not a way to
48:10
go.
48:11
It doesn't mean that you have to kind of, on the interview,
48:13
a person will only tell you how great
48:15
you are, how technology is great, but you need to
48:17
understand what is in it for them,
48:20
right? So why does he want to join your
48:22
company? That has to be authentic. You
48:24
don't have to hire people just because they told
48:26
you too many compliments as a founder.
48:28
Or you, especially when you're saying these hackathons.
48:31
So anyone who's listening now, let's say you're
48:33
not even in the technology game, like you're going
48:35
to build a technology type business. Well, you got
48:37
to think where your clients are going to be, right, and how
48:39
you make money from them as far as the
48:41
client base. But you also have to, if you want to
48:43
hire, you're like, okay, where would people be
48:45
that I want to hire that are specialized in
48:48
this? So maybe different types of
48:50
meetups, depending on what your specialty is, but
48:52
you'd want to go to those type of events to
48:54
try to find people, at least the first couple
48:56
of people, seems like to make sure that, yeah,
48:58
maybe you need a program or someone to help you even develop
49:01
an app. And you're like, okay, let me try one of these hackathons.
49:03
Maybe that's somewhere where I need to go because I know nothing about
49:05
it. So that was obviously a clever idea and
49:07
I think pretty smart. So kudos to
49:10
figuring that out.
49:11
Yes, everyone should now look up
49:13
some hackathon and sign up for that because it's really
49:15
good just to meet motivational people. Sometimes
49:17
you get such a great energy boost,
49:20
just talking with someone who's passionate
49:22
about something. It doesn't even has to be your
49:24
field. So that's really inspirational. I
49:27
think it's a really good place to kind of occasionally
49:29
visit.
49:30
Makes total sense because even when you're hiring, when
49:32
you're talking about hiring, I guess in general, some
49:34
people are like, you can tell it in their voice, are super
49:37
excited when they get up to work on something. But
49:39
maybe they're not as good proficiently
49:41
as someone who might be a little bit calmer.
49:44
Sometimes when you hire both those people, they kind of bring
49:46
out the best in everybody where you have this guy who has
49:48
good energy, makes everyone more positive versus
49:50
if you have hired someone who is negative and
49:53
might be good at their job, no one wants that
49:56
in their company because it's going to bring down everybody else. So
49:58
maybe they're personally really good at their job. but
50:00
the negative energy they're bringing to everybody else is bringing it
50:02
down. It makes it not worth it. So again, yeah,
50:05
I think that's a great idea. Trying to find a tech person,
50:07
look at packathons in your area, but there's different
50:09
ways you could brainstorm your next business idea
50:11
and think about, okay, where would be a perfect co-founder
50:13
be? Where would I find that person? Am I going to find
50:16
that person on the basketball court down
50:18
the street? More than likely not. But what happens
50:20
if I go to some type of event that's
50:22
in the industry that I want to be in? Hey, you
50:25
have a much higher percentage of trying to find someone
50:27
who will work with you and hopefully be an employee or
50:29
maybe a co-founder
50:29
or something of that nature.
50:31
I totally agree. And I even
50:33
want to add to what you said about how to
50:35
keep the moral in the team. Some
50:37
people are negative, some people are just
50:39
very excited about things and just
50:42
very temperate is different. But generally,
50:44
I think that skills are important, of course,
50:46
especially if you are a tech startup, especially
50:49
if you are building deep tech
50:51
product. You have to understand how
50:53
it works, you have to hire professionals.
50:56
But
50:57
I would say that you still can
50:59
teach skills if some of them are missing, but
51:02
it's very hard to teach someone an
51:04
attitude. I think it's impossible.
51:06
So you kind of have embraced people who are who they
51:09
are. If you think it will work
51:11
out,
51:12
yes, you need to hire this person and
51:14
move forward. If you are saying over time,
51:16
it's not working out anymore,
51:19
I think it's a good idea to just kind of be frank
51:21
about that, talk with that person, maybe
51:24
just should be parting your ways. Because if
51:26
they are negative, probably there's some reason behind
51:29
that. Maybe we don't believe in a product. Maybe
51:31
we don't like some things about the company
51:33
and they're like a way to improve this. But
51:36
attitude is very important. It's
51:38
as important as skills. I think quite
51:41
often, it's even more
51:42
important than skills. No, I agree.
51:44
Because even what I kind of said earlier,
51:46
I'm like, if you have a person who has positive
51:48
energy and they're uplifting all these other people, just
51:51
think about the percentage better that you're doing all over
51:53
in your company versus again, if you had a negative person
51:55
bringing everyone down, maybe 20%, 30%. So I think that makes
51:57
a huge difference.
51:59
difference. And so after you made
52:02
your first couple of hires, can you walk us through
52:04
maybe the next big step in
52:06
your journey of creating this company?
52:08
Yeah. So right now as a company,
52:10
we think that we have identified where
52:13
segment, we were doing things maybe
52:16
a little bit better than everyone else. So
52:18
this is where we kind of want to grow.
52:19
Yeah. But real quick before when
52:22
you're talking about doing AI, even in the beginning,
52:24
did you decide that it was going to be the beauty field right
52:26
away? Did you just look at stats and decide
52:28
you're going to get in that field? Like how did you decide that?
52:31
Initially, we wanted to go into health tech in
52:33
a medical software, but medical software
52:36
usually has very small data sets.
52:38
And it's kind of crucial part for AI
52:40
to be able to train well is to
52:42
have large data sets.
52:44
In 2016, structured
52:46
repositories of data or
52:48
data lakes for clinics
52:51
didn't exist. There was no idea
52:53
of how this medical data will be regulated.
52:55
Everyone was scared. Everyone was kind of frozen
52:58
and hesitant to act.
53:01
I think that this medical revolution is
53:03
only happening like really in
53:05
the last couple of years when also
53:08
medical regulation changed, for example,
53:10
in Europe and became more straightforward.
53:13
Still, it's kind of not perfect, but it's, you know,
53:15
there's immense work done there. Also,
53:17
it was FDA similar approach for
53:20
how we should be treating AI software. But
53:22
let's think of like back in 2016, there
53:25
was no clear legal framework and
53:27
we just thought we're like two
53:29
person company with no experience
53:32
in the medical field. So we
53:34
wanted to kind of experiment and like kind of move faster.
53:37
So this is why we started to look at
53:39
fields that are not necessarily medical,
53:42
but they're also wellness. So this is when
53:44
we started to think about skin. This is when
53:46
we started to think about beauty. And
53:49
one of the decisions was, you know, let's try something
53:51
in beauty because it looks like there is less
53:53
regulation there and it's also
53:56
a challenging task and it's a big market. So this
53:58
were some of the rationals behind the
53:59
choice to go into the beauty space.
54:02
Got to ask. Cause yeah, that way you were thinking,
54:04
you're like, it's going to be much easier to get a lot more data
54:06
to actually use in our AI versus if you're
54:09
only making sales on so many medical devices
54:11
or something like that, it's almost impossible to even
54:13
know how much data, but you know, the data is like so low
54:16
amount that AI is not really going to help as much.
54:18
Yeah. So quite often you will be only
54:21
developing a model that works on
54:23
a very limited business case.
54:25
So AI has an attribute
54:27
which is called generalization capability.
54:30
It means how well your
54:33
model or like algorithm can generalize.
54:36
So if you have particular data input
54:38
and then you change it slightly, will the system
54:40
now think it's completely different thing or
54:43
will it think, okay, that's something similar
54:45
to what I did before. So the problem is
54:47
when you have a very small dataset, your
54:50
generalization capability is off. So
54:53
you can slightly iterate an image and
54:55
then for an algorithm that will be completely
54:57
different thing, but the algorithm never encountered
55:00
before. And quite often we'll just produce
55:02
results that don't make any sense. It's kind of rubbish.
55:05
So after you made this move into
55:08
getting into skincare, did you specifically
55:10
decide skincare or I guess if you can tell us a little
55:12
bit more detail, how this has panned out.
55:15
The main idea for us was what's an
55:17
easy way to capture data.
55:19
So we thought of scan because
55:21
you can photograph your scan with a selfie
55:24
phone. So I think it was one of the main
55:26
rationals for us. It's quite hard to
55:29
make a photo of any other parts of your body,
55:31
right? If you want to go, I don't know, ear
55:33
throat, or I don't know, even if you want
55:36
to track a patient, take, you know, their
55:38
picture of the top of your head. You can try
55:40
that. It's very hard. We will also try it as
55:42
a part of one of our software offering for
55:44
here. So yeah, that happened
55:46
quite naturally. Everyone wants
55:48
to have healthy skin, very smart phone. You can make
55:51
pictures of your skin. Let's try going
55:53
into skin wellness. So that
55:55
was 2018, right? Because that's when you said you
55:57
kind of started that AI. Yes. Right.
56:00
Okay. And then can you take us over
56:02
those couple of years? Cause I know we kind of jumped
56:04
to, I think 2022, you said you three X
56:08
your company by doing the WordPress plugin, but could
56:10
you fill us in between 2018 and 2022, the lessons you learned
56:12
that hopefully can help
56:15
anyone listening right now?
56:17
Yes, sure. So 2018, we
56:19
just started the company. We incorporated,
56:21
but we already had some R&D we
56:24
did before
56:25
and it was time to sell.
56:27
We had to figure out how to sell
56:29
these to someone.
56:30
We had multiple conversations with beauty
56:33
brands, with companies. They honestly were not
56:35
buying it the first year.
56:36
No one could understand, okay, that's AI.
56:38
It's super hype. It's super attractive
56:41
to be associated with AI. What's
56:43
the actual use case? Why should we buy
56:45
in case software? How I will be buying a
56:47
software?
56:48
Do I buy it and I send you pictures
56:50
to send me back the report or how does it
56:52
work? What if I want to process 1000 pictures? How
56:55
fast can you send me results? What if I want to do
56:57
it in a retail kiosk? Do I have to wait
56:59
an hour for you to send me results? I
57:02
had conversations with people in their
57:04
field who back in 2014 were applying
57:07
virtual makeup in the pictures
57:09
and you would be coming into a store
57:12
making a picture, it would be sent overseas,
57:15
I don't know, to maybe a completely different location
57:17
and there an artist would be applying
57:19
some virtual makeup on you, like real
57:22
makeup, just painting that and sending back to
57:24
their grocery store where you maybe made this
57:26
picture, I don't know, like Walmart, and
57:28
then you will be coming back and seeing
57:30
your results, right? So there
57:32
was no clear understanding what would
57:34
be the use case around that. So we spent
57:36
a lot of time just, you know, talking to people, trying
57:39
to understand what we would buy, trying to understand
57:41
what's their problem. In 2019,
57:45
2020, we started to get first idea, you know, but it
57:47
has to be something that you can use
57:49
easily. So this is when this idea of API
57:52
business came to us. We
57:54
spent quite a lot of time on engineering
57:57
sites, right? Because you have this amazing
57:59
AI model. It runs in
58:01
the cloud, it can generate meaningful
58:04
good results, which are comparable
58:06
in accuracy with an expert grading. But
58:09
what if someone sends
58:12
us 1000 pictures in one minute? For
58:14
the test of reasons, maybe it's like very high
58:16
demand for this API or maybe
58:19
there is like some fraud or attack, like maybe
58:21
someone's trying to corrupt your system, what you're gonna
58:23
do? So then realize that if
58:25
we want to work with corporations, they
58:28
value reliability, they value
58:30
well engineered things that don't break. So
58:33
again, in 2020, we were really investing
58:35
in our infrastructure, thinking through
58:37
the use cases, how we can make it secure, how we can
58:39
make it fast. So after that, AP
58:41
business was more or less established. But
58:44
another question came, like how we can sell more now, right?
58:47
So there is only 200 big brands,
58:50
big companies on the market. So how
58:52
we can sell above this to 200 brands and companies,
58:56
like the largest ones. So this is when we
58:58
started to think what could be the delivery system.
59:00
The idea of e-commerce widget came to
59:02
us really spontaneous. I don't even
59:05
remember how. I think we just had some
59:07
daily stand up with everyone
59:09
just talking about what's we gonna do next
59:12
for our software and then the idea came
59:14
like, but can I buy it without API?
59:16
How can I do that? And I think
59:19
with everyone started to reflect on that idea and
59:21
the idea of widget came by because
59:24
you want to embed it into website, make
59:26
it seamless. We also knew that we are developing
59:29
software for beauty. We need to kind of
59:31
make a step back and think, what's the difference
59:33
between beauty and no manufacturing? Or
59:35
I don't know, maybe machines or manufacturing cars.
59:38
Or like, what's the difference between beauty and pharma, right? Or
59:41
like maybe medical dermatology. So beauty
59:44
is immersive. It's a way of self-expression.
59:47
And this is some of their points
59:50
we wanted to make address
59:53
in our software.
59:54
Beauty is about expressiveness
59:57
if beauty is to be immersive. Software
59:59
with
59:59
we will deliver to customers who are
1:00:02
businesses also should allow them to
1:00:04
express themselves, meaning that they
1:00:06
should be able to kind of build branding
1:00:08
on our platform or customize
1:00:11
the language or customize the looks.
1:00:13
They want it to be seamless because experience
1:00:16
is important. Sometimes you go to some
1:00:18
brick and mortar stores, you know, just to maybe
1:00:20
take the bottle, feel how heavy
1:00:22
it is, feel like what's the texture,
1:00:25
maybe smell the scent of a cream. So
1:00:27
how we can translate this immersiveness,
1:00:29
how we can build trust, right?
1:00:32
Because if I want to say, you know, it's a better way to shop,
1:00:34
like, why do you think so? Why
1:00:36
would you be saying that using Hout AI
1:00:39
software is better alternative than, you
1:00:41
know, maybe using a quiz or, you
1:00:43
know, maybe better alternative to just randomly
1:00:45
choosing products? So this is when we started to think,
1:00:47
okay, software should develop a
1:00:49
comprehensive report, which is not too
1:00:52
advanced in a way that it's advanced,
1:00:54
but it's not kind of overwhelming. So
1:00:57
this again, our ideas kind of run product should
1:00:59
be trustable. It should be easy to
1:01:01
understand, like, what predictions
1:01:03
were made, why products were recommended. It
1:01:05
should be nicely visually
1:01:07
designed. We should have an option to customize
1:01:10
that. So you start to think, you know, what's what
1:01:12
are the kind of special traits
1:01:14
of this field? You start
1:01:17
to think more about what's important for
1:01:19
my clients. You can ask them. Sometimes
1:01:21
they tell you straightforward what they want. They quite
1:01:23
often don't tell you what they want. You
1:01:25
need to observe. You need to
1:01:27
listen. And all of these conversations,
1:01:30
everything, all of your experience, even like unsuccessful,
1:01:33
will form your vision. This is, I think, what
1:01:35
we did with our unsuccessful attempts. We
1:01:37
solved things that were not successful. But
1:01:39
then we also talked with our clients. We grow a network.
1:01:42
And this is how we came up with the idea of this constructor,
1:01:45
e-commerce widget, but we're selling right now very
1:01:47
effectively. And yeah, that's how
1:01:49
our journey was.
1:01:50
Being gradually making connections,
1:01:53
exploring what's happening, exploring
1:01:55
the trends, and consolidating this knowledge
1:01:58
into our product.
1:01:59
What does your revenue look like today?
1:02:02
That's information is confidential, so I will not
1:02:04
be disclosing exact number, but it's still
1:02:06
under 10 million in annual recurring
1:02:08
revenue. But we're hoping in
1:02:10
a couple of years, maybe in three years, to
1:02:13
surpass that threshold.
1:02:14
And how many clients? Because I know you said you looked
1:02:17
at the top 200, that's when you had the AI,
1:02:19
right? I was curious, after you looked at that,
1:02:22
I imagine you reached out to all of them.
1:02:23
Yes, unfortunately, not everyone
1:02:25
bought them.
1:02:26
Well, you don't necessarily have to do it with
1:02:28
everybody, right? But yeah, I was curious, how many of those
1:02:30
big clients you landed first before you decided you
1:02:32
had to kind of do a widget in order to reach
1:02:34
different demographics
1:02:35
of companies? So I think we
1:02:38
had around 30 clients, maybe
1:02:40
like 25, 30 clients, when we started
1:02:42
to think of how we can sell more.
1:02:44
I mean, it's obviously not a very big number, but
1:02:46
you need to think back, we were a startup, we
1:02:49
are a startup now. Right. And yeah, zero
1:02:51
in the beginning. So I don't think it's small
1:02:53
or anything. Yeah. And especially if you're selling to big
1:02:56
companies, that's much more difficult.
1:02:58
Well, it's to some extent more
1:03:00
difficult. And I think to some extent, it
1:03:03
has some benefits. Because honestly,
1:03:05
big companies, they will take a good
1:03:07
technology in shape of a product
1:03:09
that's undercooked,
1:03:11
because they have their technical capacities
1:03:13
to finish a product, polish that, maybe
1:03:16
change some user interface, or maybe
1:03:18
build additional business logic. They do have the resources,
1:03:21
they have designers, they have product
1:03:23
people to help them shape
1:03:25
your technology into the product and
1:03:27
consumer facing experience they want. In
1:03:29
that sense, yes, they can do that. Small companies
1:03:32
will not do that, right? They need to have something finished
1:03:34
ready, but they can just take and use.
1:03:36
I'm here with Megan Bennett. How's it going, Megan?
1:03:39
It is going great. How's my favorite
1:03:42
podcast host and the most handsome young man?
1:03:44
I'm doing fantastic. Thanks for stating
1:03:47
the obvious, Megan. But we're here to talk about you
1:03:49
and your company light years ahead.
1:03:51
I interviewed Megan on episode 177 of this very podcast. And
1:03:56
she helped all of our Patreon members on
1:03:58
group call three. So you can hear more. about Megan
1:04:00
and how she helped our Patreon members there as
1:04:02
well. So would you mind telling us what you
1:04:04
do and how you could help our listeners, Megan?
1:04:07
Yes. So my agency
1:04:09
is Light Years Ahead and we're boutique,
1:04:11
but we're a national PR firm. We're
1:04:14
women owned and we focus on emerging
1:04:16
brands, experts and services in the consumer
1:04:18
lifestyle space. We're based throughout the
1:04:21
US. We're in New York, Kansas City,
1:04:23
LA and Dallas. And we really specialize
1:04:25
in maximizing media exposure
1:04:27
for brands and experts, which can then create
1:04:30
more sales and brand awareness and
1:04:32
influence buying decisions. Our clients
1:04:34
range everything from small startups looking to
1:04:36
make a name for themselves to large brands
1:04:38
that are trying to become relevant again. My agency
1:04:41
Light Years Ahead, we target the very top editors,
1:04:43
writers and producers across all different
1:04:45
media outlets. And we've been doing this for over 20
1:04:48
years, which has earned us a very strong
1:04:50
reputation with the top media, with
1:04:52
outlets like Buzzfeed, Today Show,
1:04:54
Good Morning America, Refinery 29, PopSugar,
1:04:57
Forbes, and many more. We can
1:04:59
help you grow your brand into a household
1:05:02
name.
1:05:03
That sounds awesome. So if someone
1:05:05
might be interested in your service, what's the
1:05:07
best way for them to reach you? The
1:05:09
best way to reach out is to email me at megan,
1:05:11
M-E-G-A-N, at lightyearsahead.com.
1:05:14
That's megan at lightyearsahead.com. Or
1:05:16
you can check out our services and capabilities
1:05:19
at lightyearsahead.com, our website.
1:05:21
And I know you've helped a few of our past guests
1:05:23
as well with their PR and they do sing your
1:05:25
praises. So hopefully it can help some of our
1:05:27
listeners as well.
1:05:28
Absolutely. And we love working with
1:05:31
your listeners and entrepreneurs who are really
1:05:33
passionate about what they're doing. And
1:05:35
this is what we want to offer your listeners. The
1:05:37
first five listeners that schedule
1:05:39
a call with us to develop a PR
1:05:41
campaign will receive
1:05:43
$500 off their first month of services
1:05:45
with us. It's a great deal. Awesome. And
1:05:48
one more time, what's the best place for them to reach you to
1:05:50
pick you up on that offer?
1:05:52
You can reach me at megan at lightyearsahead.com
1:05:55
or check out our website at lightyearsahead.com
1:05:58
or you can go to our Instagram page.
1:05:59
at LYA PR.
1:06:04
Want to expand your customer base and boost
1:06:06
your revenue? Brevo's comprehensive
1:06:09
CRM suite is tailor-made for
1:06:11
businesses like yours. Brevo,
1:06:13
formerly known as SendinBlue, is
1:06:16
the leading customer relationship management
1:06:18
suite designed to fully cultivate long-term
1:06:20
customer relationships and empower
1:06:23
businesses to expand in a fast-changing
1:06:25
digital world. With Brevo, businesses
1:06:28
have unified view of the customer
1:06:30
journey in one easy-to-use
1:06:32
platform to grow meaningful
1:06:34
relationships. Brevo makes the simple
1:06:36
and accessible with intuitive and scalable
1:06:39
marketing and CRM tools such as email,
1:06:42
SMS, WhatsApp, chat, marketing
1:06:44
automation, meetings, and much more. In
1:06:46
a world where marketing encompasses the
1:06:48
entire customer experience, Brevo
1:06:51
puts the focus on lasting relationships
1:06:53
with real people. Brevo gives
1:06:56
you the tools to attract, engage, and
1:06:58
nurture customer relationships. You can build
1:07:00
automated customer experiences, email
1:07:02
marketing workflows,
1:07:04
and landing pages that guide
1:07:06
your customer to your main message. They're
1:07:08
there to support businesses successfully
1:07:10
navigating their digital presence in order
1:07:13
to strengthen their customer relationships.
1:07:15
It's the perfect business growth tool for marketers,
1:07:18
SMBs, and sales teams
1:07:20
looking to build with a consolidated marketing
1:07:22
and CRM toolbox. Brevo's pricing
1:07:24
structure is based on the number of emails
1:07:26
sent, not the number of contacts
1:07:29
stored.
1:07:29
Get started with Brevo for free by
1:07:32
clicking on their link below or
1:07:35
going to brevo.com
1:07:37
forward slash entrepreneur and
1:07:39
use the promo code entrepreneur to
1:07:41
save 50% on your first three
1:07:44
months of the starter and business plan.
1:07:46
Again, that's B-R-E-V-O.com
1:07:50
forward slash entrepreneur and sign
1:07:52
up for free.
1:07:54
Well, I guess it's understanding, knowing who's
1:07:57
gonna be your clients first. You knew it had to be bigger
1:07:59
companies. because you need the data, right?
1:08:02
But
1:08:02
I'm sure you're probably underselling yourself right now
1:08:04
as far as how difficult it is to get your
1:08:06
foot in the door in the beginning. And then after you get in,
1:08:09
it's probably easier and it makes sense. And
1:08:11
like you said, they have these tech teams or whatever, but right
1:08:13
when you started, they didn't know who you were. I
1:08:15
guess you're in Estonia, right?
1:08:17
So it's like, who's this girl in Estonia emailing
1:08:20
me about AI, right?
1:08:22
Yes. Whenever we were connecting
1:08:24
with clients back then, first time we would say, you know,
1:08:26
Estonia, it's the company where Skype came from.
1:08:29
And everyone would say, oh, Skype.
1:08:30
Is that actually true or no? Yes, it's
1:08:33
true. So Skype is an Estonian company and
1:08:35
we still have a lot of Skype
1:08:37
first team players who are now also building where
1:08:40
maybe like second, third startup.
1:08:42
Skype co-founder is also like
1:08:44
the founder of Atomica VC firm,
1:08:46
which is based in UK. But I know
1:08:48
that all of their Skype X founders
1:08:51
are promoting entrepreneurship and
1:08:53
startup ecosystem. This is why Estonian startup
1:08:55
ecosystem is so powerful and so friendly.
1:08:58
Yes, we have amazing companies here. Do you know
1:09:00
CRM software, Pipedrive?
1:09:02
Yes. It's also Estonian company. Okay,
1:09:04
yeah, I use it. Both systems, which
1:09:07
like kind of Uber competitor is also Estonian company
1:09:09
initially. Transfer wise, initiative Estonian
1:09:11
companies. We have a lot of amazing companies that came from
1:09:13
Estonia.
1:09:14
Right, and specifically your own talent
1:09:16
Estonia, right? Which is the capital? Yes,
1:09:18
that's correct. Is that where all the startup people
1:09:21
are? Cause I'm just curious of case anyone who's listening,
1:09:23
maybe they're in Estonia, didn't even know that
1:09:25
or whatever, but I'm just curious. After you
1:09:27
listed off all these companies or most of the people
1:09:29
just in the capital there.
1:09:30
Well, we have also like many
1:09:32
people are based in here in the capital city
1:09:34
in Tallinn. We also have Tarteau, which is
1:09:36
very famous for its university.
1:09:39
It's amazing natural sciences university
1:09:41
and also like has offers degree in software
1:09:44
and maths.
1:09:45
We also have a big bio bank in
1:09:47
Tarteau, which is very famous in the
1:09:49
world, especially for longevity
1:09:51
and bioinformatics companies because it has an amazing
1:09:53
deposit of data.
1:09:55
And Tarteau university offers also like
1:09:57
different types of partnerships. But yeah, getting
1:09:59
back to the... that thing, but we didn't know who we are.
1:10:02
Again, people sell to people. So
1:10:05
you try to explain what the
1:10:07
technology does. You collect
1:10:10
proofs because you also, we had
1:10:12
research papers where we explained what we're doing
1:10:14
and how. So you have to build this trust.
1:10:16
You have to, it was not an easy thing,
1:10:19
right? And this is why you build relationships
1:10:21
with corporations for a long time because
1:10:23
they kind of evaluate. Is there any
1:10:25
reputational risks for me working with this company? Is
1:10:28
it worth my time or like
1:10:30
my team time on working with this company? So
1:10:33
I would say you have to be specific
1:10:35
about what you're offering. You have to
1:10:37
be knowledgeable of your product. You have to be knowledgeable
1:10:39
of your technology. But yeah,
1:10:41
as soon as you land a couple of big clients,
1:10:44
you start to have a reputational, they work with some
1:10:46
other big companies. Maybe I should kind of
1:10:48
be less skeptical about having a call with a super
1:10:50
small startup from Estonia, not even
1:10:53
all people know where Estonia is.
1:10:54
Right. I guess if anyone's wondering, kind
1:10:56
of by Sweden and Finland, they can think like that
1:10:59
or right by Russia. Yeah. It's a smaller
1:11:01
country, but maybe not a lot of people have heard about,
1:11:03
but back to even landing that first client,
1:11:06
right? Did you just keep emailing them out of
1:11:08
these top 200 companies that you're talking about and
1:11:10
maybe it took a year or two for you just giving
1:11:12
them updates every once in a while? I'm just curious if anyone's
1:11:15
listening now and they, they're
1:11:16
trying to sell to a big company, what
1:11:18
it took for you and what thoughts you might have
1:11:20
for them.
1:11:21
Yeah, I would say that email marketing
1:11:23
is still a thing. Social media marketing
1:11:25
when you're reaching out to people on LinkedIn is
1:11:28
also important, telling, like sharing
1:11:30
more about what you are doing, whether it is LinkedIn
1:11:33
blog or company blog, just, you know, whenever people
1:11:35
who are interested, because if people are interested,
1:11:38
if they're looking for something similar
1:11:40
to what you're doing, it's more
1:11:42
likely that they will notice
1:11:45
you. It's more likely that they will be
1:11:47
open to have a conversation with you.
1:11:48
So that's why email marketing is important. But
1:11:51
let's say you send, you know, maybe 10 emails
1:11:53
and a person already starts to think that you're
1:11:55
like stocking them, right? They finally
1:11:57
give you a chance and they think, okay, let's maybe connect.
1:11:59
a call, but they for sure will look
1:12:02
you up on our internet, right?
1:12:04
They type in your name and your company.
1:12:06
If your website is kind of doesn't even
1:12:08
explain what you're doing or you know, at least kind
1:12:10
of high level, if they don't see any
1:12:12
kind of social media where you maybe explain
1:12:15
more about your product or technology, however you're doing
1:12:17
that, they feel very skeptical,
1:12:19
right? So, I think like having a good website,
1:12:22
it doesn't have to be kind of a masterpiece
1:12:24
of arts, right?
1:12:25
But at least explaining your product,
1:12:28
explaining your technology, what it does,
1:12:30
you know, how it works is super helpful.
1:12:32
We were sending our clients, we were building
1:12:34
up with credibility over time, we realized that
1:12:37
we will publish our research work,
1:12:39
right? You quite often don't publish
1:12:41
your research work, but we thought
1:12:43
we want to explain how technology works. People
1:12:46
are still not understanding fully AI, that's normal,
1:12:48
you know, you can't be kind of expert in everything that's
1:12:50
happening. If you would expect with everyone that
1:12:52
understands AI or whatever
1:12:55
specific field you're working, it's like, do
1:12:57
you understand, you know, some of the very advanced chemistry,
1:12:59
but maybe it was developed in the last two, three
1:13:01
years, right? Likely not. So,
1:13:04
you always have to think if you're
1:13:06
in the shoes of another person, maybe a client with whom
1:13:08
you try to land, until people can
1:13:11
understand, you know, at least how it works,
1:13:13
high level, what you're trying to do, they will not
1:13:15
be convinced, right? So, we were continuing
1:13:18
also like sending emails or offering
1:13:20
to have an update and maybe sending a quick
1:13:22
blurb of what we have done
1:13:24
and how we have done that.
1:13:26
Even if I don't
1:13:27
respond to you, you know, maybe you send
1:13:29
them another email in a year, I think, okay, I
1:13:31
maybe remember these guys, it looks like they made
1:13:33
some progress. They already made it
1:13:35
through the year, which is great for a startup, some startups
1:13:38
die, you know, in less than a year, like in less than two
1:13:40
years. So, you need to kind of think
1:13:42
of how you can build credibility and
1:13:44
you build this credibility for explaining how your product works,
1:13:47
demonstrating that idea is legit,
1:13:49
that technology is legit or like product
1:13:51
is helping, you know, maybe collecting some metrics
1:13:53
and reaching out to your clients from time to time. Of course,
1:13:56
not abusing them,
1:13:57
but building your social media, building your
1:13:59
website is an important component of marketing.
1:14:03
I think for many people who can do marketing
1:14:05
in a more creative way, I'd like to meet them, but
1:14:07
this is what we were doing back then.
1:14:09
That's smart because at first you're probably just sending
1:14:11
the emails without like a great website
1:14:13
or anything like that. And
1:14:14
again, if anyone wants to check yours out
1:14:16
is haut.ai. But
1:14:19
it's very simple. It's like basically a one pager where
1:14:21
you have all the information there. And let's
1:14:23
say you didn't have this beforehand and
1:14:25
you didn't have on LinkedIn activities
1:14:28
where you're writing or doing interviews or
1:14:30
something like that. Like you said, you have to put yourself
1:14:33
in their shoes as a client. And like kind we
1:14:35
said earlier, finding your first employee or
1:14:37
whatever you got to think of where would those people
1:14:39
be? So especially today, people, if
1:14:41
they're not sure and eventually they want to get on a call with
1:14:43
you, they're going to do some Googling. So just Google
1:14:45
yourself or Google your company, see what's
1:14:48
coming up. If nothing's coming up, then you're thinking,
1:14:50
okay, maybe I need to improve
1:14:52
my SEO or at least get your social
1:14:54
media kind of thing started. It's not like something you have
1:14:56
to update all the time. But I think the more touches where
1:14:58
they can see that you're a legit company
1:15:00
or feel more confident, they might
1:15:03
not even tell you that's what they finally took them
1:15:05
to reply to the email. But in their
1:15:07
head, that's probably what it did before
1:15:09
if you didn't have all those social accounts or proof
1:15:11
of, okay, this person knows what they're talking about. You
1:15:14
don't always know why they got back to you, but guaranteed
1:15:16
that's one of the reasons.
1:15:18
Can't agree more. You need to have your social
1:15:21
presence. So eventually, like
1:15:23
you said, you did the WordPress plugin, which kind of
1:15:25
really catapulted things. And then I guess
1:15:27
what's gone on over the past year since
1:15:30
kind of done
1:15:30
that. So we started to think we
1:15:32
made some, you know, product changes to make
1:15:34
our product easy to use.
1:15:37
As we were growing, a number of clients realized
1:15:39
that some of their use cases, some
1:15:41
of the new use cases were developed by our
1:15:43
clients. You know, we were complaining that some
1:15:46
things are kind of buggy
1:15:47
or, you know, it's not very convenient to use some
1:15:49
particular features. It turned out that
1:15:52
customers have developed new features,
1:15:54
new kind of use cases, without us
1:15:56
understanding that.
1:15:58
So making sure, you know, that we...
1:15:59
now kind of streamline our user experience,
1:16:02
but this product is easy to use and it's evolving.
1:16:04
It's something that we're working on at the moment.
1:16:07
Then we also like adding additional
1:16:10
capabilities to help our
1:16:12
clients extract more value from our
1:16:14
software, including
1:16:16
automated inside generation or
1:16:19
suggestions for their product inventory. So
1:16:21
whatever makes their life easier. In fact,
1:16:23
what we're doing right now, we are continuing
1:16:25
building up the value on the initial
1:16:28
offering of computer vision-based
1:16:30
skin analysis. So we're evolving
1:16:33
constantly the feature set or
1:16:35
what you can do with our software. That's
1:16:38
what is keeping us busy. So we
1:16:40
want to understand how we can serve better to our clients.
1:16:43
We do it through collecting their feedback. We do
1:16:45
it through analyzing the product
1:16:47
data and also constantly keeping
1:16:49
an eye on what's happening in the industry, what
1:16:52
consumers want, conducting interviews
1:16:54
with real people, especially consumers
1:16:57
who are not tech-savvy. It's so
1:17:00
seductive to build a
1:17:02
bias unconsciously in your product, especially
1:17:04
if you kind of have a lot of technical team
1:17:07
members. I will give you an example. We
1:17:09
recently conducted a research
1:17:11
with users. So we recruited some users
1:17:14
through special websites where you can test
1:17:16
your products. And again, if you think that you need
1:17:19
to test something, there is so much
1:17:21
software. You can do it. You don't have to have a big
1:17:23
budget to start testing if your product
1:17:25
makes sense. For example, in this particular case,
1:17:27
I was using usability hub.
1:17:29
And it may be costing me $100 to run this test.
1:17:33
But getting back to the experiment,
1:17:35
we recruited some respondents and we asked
1:17:37
them what we want to have in the software. So
1:17:41
they told that they want to have skin analysis.
1:17:43
They want to have recommendations. And that
1:17:45
they also want to have an
1:17:47
educational component. So these three things.
1:17:50
So then we also invited our friends to participate
1:17:52
and give their feedback.
1:17:53
And results were different. So they
1:17:55
said that we want face scan, that we want to have recommendations,
1:17:58
and then that we also want to.
1:17:59
track skin dynamics, right?
1:18:02
So then we started to analyze this data
1:18:04
and understand, you know, why we have different
1:18:07
results from our network and like from our
1:18:09
initial users who were like randomly recruited.
1:18:12
So the thing is that tracking over time is technically
1:18:15
very advanced.
1:18:16
So
1:18:17
our assumption is that because we recruited from, you know,
1:18:19
our kind of network and a lot of developers or
1:18:21
like AI people, computer vision engineers,
1:18:24
we just think that we started to prioritize this feature
1:18:26
because it's technically very advanced
1:18:28
technique. You know, it takes a lot to
1:18:30
do that. Making sure the team
1:18:32
is growing, as a product is evolving, making
1:18:35
sure you know that use cases and
1:18:37
usage is relevant to the consumer,
1:18:40
not only kind of to the development team, but
1:18:42
I think that's very important. And this is kind of what we're
1:18:44
now reiterating in our product. And
1:18:47
this is how we want to grow our customer
1:18:49
base through serving better to their needs rather
1:18:51
than only doing something which is very technically
1:18:54
advanced just for the sake of that.
1:18:56
As you're in it right now, you know, I think that we
1:18:58
all learn in business, we try to reiterate or try
1:19:00
to make things better as we learn and get more information.
1:19:03
And obviously you're kind of doing that with your company
1:19:05
and it
1:19:06
seems like that's what your company is all about. So
1:19:08
just want to say thank you for coming on the show and spending
1:19:11
the time of helping me understand AI
1:19:13
a little bit and
1:19:14
telling us a little bit about your company and how everything's
1:19:17
going. I guess kind of in closing, I don't know if you
1:19:19
have any last
1:19:20
words of wisdom for all the entrepreneurs
1:19:22
out there or you can help anybody else,
1:19:24
what's kind of the best way for them to reach you?
1:19:27
So first of all, I want to say Austin, thank
1:19:29
you so much for inviting me. I was really excited
1:19:31
to share our journey and I
1:19:33
really hope that someone will listen to that
1:19:35
and think, okay, maybe you have this crazy idea, I never
1:19:38
thought it can come to life, but you
1:19:40
should try that. The worst thing that will happen,
1:19:43
it will not fly and maybe you can try some other
1:19:45
things. It's a good thing you can find. I think
1:19:47
that you will be passionate and will motivate you
1:19:49
and like improve a lot of your areas of your life. So
1:19:52
believe in yourself, writing is the right
1:19:54
people, it's very important and I think
1:19:56
it can lead to great success if you
1:19:59
find this special...
1:19:59
and find the right people to join your
1:20:02
team.
1:20:02
And if you'd like to get in touch with
1:20:04
me, just connect with me on LinkedIn. I'm really
1:20:06
happy to help whatever I can
1:20:08
do.
1:20:09
So anyone who's like listening now, as
1:20:11
far as maybe they could be a client
1:20:13
for yours, are they only kind of skin companies?
1:20:16
Who should reach out if you can actually see
1:20:18
if they can implement your AI here? Who are
1:20:20
the best people to reach out to you on your website?
1:20:23
I would say that we're a beauty intelligence company.
1:20:25
So we're not only limited to skin, we
1:20:27
have software for hair analysis, for
1:20:29
body analysis. We now entered oral
1:20:32
care and animal erosion. So
1:20:34
if you are in the beauty field, in aesthetics,
1:20:37
in longevity, and skin science, we
1:20:39
will highly appreciate if you reach out.
1:20:42
We'd like to connect with you, we'd like to talk with you
1:20:44
what's happening in the field and how we can help you.
1:20:46
Should they just reach out on hot.ai?
1:20:50
Yes, we have a contact form and it's the best way
1:20:52
to reach out to us. All right, well, thank you again
1:20:54
for coming on and sharing your story. Thank you
1:20:56
so much, Austin, for having me.
1:20:58
On the line, on the line. But
1:21:02
it's bad when you do it to your wife, though, because
1:21:04
then you have to crash on the couch. See,
1:21:07
I have to sleep on the couch every night too, man. See, we're
1:21:09
the same. Was
1:21:11
that helpful at all, Gary? Say no. Worst
1:21:15
experience of my life, one star review.
1:21:18
Yeah, thank you. I'm used to this. Wish I could
1:21:20
leave no stars. Oh yeah, hell yeah. No,
1:21:22
thanks guys. It was a really great experience.
1:21:26
Feel like there's a lot to reflect on. So
1:21:28
yeah, thank you.
1:21:30
And I can connect you with somebody too. Okay.
1:21:33
I have connections on that so I can help you get it.
1:21:35
Custom made, dirt cheap. I'll
1:21:37
share that with you. Look at that Patreon membership
1:21:40
already paying off.
1:21:42
No, look at that.
1:21:45
Thanks for coming, member. Oh, well,
1:21:47
I gotta thank my business partner. She's
1:21:49
signing me up because I've been talking about you. Well,
1:21:51
awesome business partner. I'm gonna have to use that as a plug
1:21:53
to tell people to do the same thing. Yeah,
1:21:55
yeah, it's really cool. But anyway, yeah, thanks for
1:21:58
setting this up. Get kind of the...
1:21:59
VIP treatment, if you like. Well,
1:22:03
I thought it was a lot more intimate than I thought it was going to be.
1:22:05
Like anyone who's thinking about doing it, you'll
1:22:07
be able to get involved, ask a question,
1:22:10
you know, which I don't have a lot of experience
1:22:12
with other group calls, but I would assume
1:22:14
that there's kind of a hierarchy to it. But this one,
1:22:17
if you're in there, you're going to get your shot to ask
1:22:19
an expert a question. So I
1:22:21
tried to compare my group calls. I started joining
1:22:24
random entrepreneur groups and just joining
1:22:26
their group calls and try to see what they're like. Dude,
1:22:29
the one you were on
1:22:29
and all of them have kind of gone that way. They're
1:22:32
all 10 X better than any other group I've been
1:22:34
in because become
1:22:36
a member to find out. So
1:22:39
with Patreon, I heard it many times
1:22:42
because you have that many episodes of signup,
1:22:44
so that's always in the back of mind, but then I checked
1:22:47
it out a few times and I was like, do I
1:22:49
really want to do this? So I'll push it off a
1:22:51
little bit. And then you posted your
1:22:54
goal achievement of 69
1:22:56
Patreon members. And I was like, you know what, what
1:22:58
better time than now? Originally,
1:23:00
I was going to go for the lower one than $9 a month. But
1:23:03
one, I want to have the conversation with you. But
1:23:06
two, I always find that anytime I cheap
1:23:08
out, I always find that I want to
1:23:10
return it and upgrade to what I really,
1:23:12
really wanted. So that's
1:23:14
why I'm paying the higher one, if
1:23:17
that makes sense. But it was just constantly
1:23:19
pushing it off, pushing it off. And then I
1:23:22
would just say, fuck it. I already listened to
1:23:24
all of them, so why not?
Podchaser is the ultimate destination for podcast data, search, and discovery. Learn More