Name: Intrepid Fox at FinovateEurope 2024 - When KYB meets GenAI
Uploaded: 2024-03-07T15:39:15.380Z
Duration: 859.96
Description: One of the leading topics at this year's FinovateEurope orientated around how Know Your Business (KYB) processes - the bane of many client onboarding endeavours for fintechs and banks alike - and more specifically, what happens when such processes intersect with the latest and brightest technologies.

FF.INT.001.mp4

FF.INT.001_1.vtt

ENGLISH

{{00:00:04}}
Hello everybody and welcome to Fintech
Futures at Finovate Europe here in London.

{{00:00:11}}
My name is Tyler Pathe.

{{00:00:13}}
I am a reporter for Fintech Futures and
today I'm very happy to be joined by

{{00:00:19}}
Roman Zilber, the founder of IntrepidFox.

{{00:00:22}}
Who is fresh off the demo stage,
literally within the next half an hour.

{{00:00:27}}
How did it how did it go?

{{00:00:29}}
It felt great.

{{00:00:30}}
A lot of people, a lot of energy.

{{00:00:34}}
I was happy to talk and
I think it went well.

{{00:00:37}}
Thank you.

{{00:00:37}}
Yes I I was in the audience when you
When you were making your speech and

{{00:00:42}}
I thought you touched upon some really
interesting points that we're going to

{{00:00:46}}
be exploring here in our interview today.

{{00:00:50}}
One of the really interesting
things that you did say is the

{{00:00:56}}
Sort of delays and the slowness of
the business onboarding process.

{{00:01:02}}
Why is it in 2024 that such a
process incurs so many delays?

{{00:01:08}}
Yeah, absolutely.

{{00:01:09}}
Great question.

{{00:01:10}}
In fact there are a lot
of research on this.

{{00:01:14}}
Subject.

{{00:01:14}}
And they say that it is several
days, seven days on average,

{{00:01:18}}
and more than a thousand dollars
the cost of such an onboarding.

{{00:01:24}}
It's amazing because again, we are living
in the age of automation, et cetera.

{{00:01:31}}
So it is strange.

{{00:01:32}}
The only answer I have.

{{00:01:34}}
As a banker with 30 years of experience
is that bankers are lazy and don't care.

{{00:01:38}}
No, but that's not true.

{{00:01:41}}
Actually, bankers are not
lazy and they do care.

{{00:01:43}}
But the problem is that during, the
process of onboarding of customers,

{{00:01:49}}
some documents and some information
you can retrieve from public sources,

{{00:01:54}}
from APIs, from registers, whatever.

{{00:01:58}}
But some documents like invoices,
contracts with suppliers.

{{00:02:03}}
Rental agreements, you can only
take from customer themselves.

{{00:02:08}}
And this causes a lot of friction.

{{00:02:10}}
Because you ask customers ask
you in return what exactly

{{00:02:16}}
do you mean by this document?

{{00:02:18}}
Then you respond to them, then they
say the wrong document, then you have

{{00:02:22}}
follow up questions on the document.

{{00:02:24}}
One of CEOs of FinTech I interviewed
he told me, Roman, quote me, any

{{00:02:31}}
request for Any additional document,
it is always two days, two days.

{{00:02:37}}
So this is, why it is so
long and so expensive.

{{00:02:42}}
So this is the cost.

{{00:02:44}}
And we have this problem at hand, which
is across all corners of the industry.

{{00:02:48}}
Intrepid Fox, where does that sort of
fit into making this process faster?

{{00:02:53}}
How does your solution meet that demand?

{{00:02:55}}
Yeah, actually, what we do, we single
out this hard part and we automate it.

{{00:03:03}}
How do we automate it?

{{00:03:04}}
We use large language models.

{{00:03:06}}
So something which is only came out
and only available very few months

{{00:03:13}}
now, to make this process not hours
or days, but seconds or minutes.

{{00:03:21}}
What it does, it actually, it
we made software which reads.

{{00:03:28}}
Customer documents, understands customer
documents, follow up questions according

{{00:03:34}}
to rules, which we predefined before,
according to criteria, which we gave it.

{{00:03:39}}
And this criteria is part
of our underwriting process,

{{00:03:43}}
which we anyway follow.

{{00:03:47}}
So it both understands the document,
retrieves information from the

{{00:03:52}}
document, and converts by the document.

{{00:03:54}}
This helps to eliminate.

{{00:03:56}}
Unnecessary human work and concentrate
our qualified people on really

{{00:04:02}}
important things like making final
decision whether to onboard the

{{00:04:06}}
company or not to onboard the company.

{{00:04:08}}
And you can take it from a wide range
of documents, I believe, as you said.

{{00:04:13}}
Yes, this is this is something we aim at
basically all the documents available,

{{00:04:21}}
and the technology can do this.

{{00:04:22}}
And this is amazing.

{{00:04:24}}
So it is amazing how fast we can read and
understand even complex documents and that

{{00:04:30}}
the accuracy there is much better than
accuracy which human, employees deliver.

{{00:04:36}}
That's a really interesting intersection
to exist in this, technology

{{00:04:43}}
meets data extraction harvesting.

{{00:04:47}}
What's really critical in, in
speed is also being accurate.

{{00:04:51}}
And you've mentioned that you're
using large language models.

{{00:04:54}}
You're also playing with different
forms of AI in your solution as well.

{{00:04:59}}
How can you ensure the accuracy of the
outcomes that the solution produces?

{{00:05:04}}
Good question.

{{00:05:05}}
When we talk about accuracy, it is
important to distinguish between

{{00:05:11}}
two cases there, there, could be
accuracy of, Extracting or retrieving

{{00:05:17}}
information from documents.

{{00:05:20}}
And here, first of all, we achieve
accuracy better than humans do.

{{00:05:26}}
But still, there are some
mistakes, can be mistakes.

{{00:05:29}}
To minimize mistakes, we split
the process as much as possible.

{{00:05:33}}
We call models only for small questions,
only for parts, not for the whole request.

{{00:05:42}}
And we use multi layered architecture,
so we have some algorithmic checks, we

{{00:05:48}}
have cross checks of the information.

{{00:05:52}}
We can sometimes pull
models on on the question.

{{00:05:56}}
You have a lot of techniques to
improve and to get to the point when

{{00:06:00}}
you don't have hallucinations and
you actually don't have mistakes.

{{00:06:04}}
This is technical problem.

{{00:06:06}}
We are able to handle it.

{{00:06:08}}
But another part of accuracy is the
accuracy of final decision here.

{{00:06:14}}
So basically you got all the
information and then you make

{{00:06:18}}
decision and this decision is wrong.

{{00:06:22}}
And here what we do the smartest things.

{{00:06:24}}
We just don't do this.

{{00:06:26}}
We believe that not regulators, not
Companies and banks, not customers, nobody

{{00:06:34}}
is prepared that artificial intelligence
takes decisions on onboarding.

{{00:06:40}}
Prepares information, yes.

{{00:06:41}}
Takes decision, no.

{{00:06:43}}
So the human in the loop who takes this
decision, but only this human is ten

{{00:06:47}}
times more efficient than they are now.

{{00:06:51}}
But in terms of the human that the
information is passed to, how important

{{00:06:57}}
is it from Intrepid Fox's standpoint
that they know How the outcome was

{{00:07:03}}
reached, how the process is explainable.

{{00:07:06}}
Yes, absolutely.

{{00:07:07}}
Transparency is extremely important.

{{00:07:09}}
And, compliance officers, they get
not only original documents from

{{00:07:16}}
the customers, but also information
about all the discrepancies or

{{00:07:22}}
issues we found in the documents
and follow up with the customers.

{{00:07:27}}
If you want, you can actually
look at the original documents.

{{00:07:31}}
But in many cases, you have to, but you
already know what you're looking for.

{{00:07:36}}
You already know the context,
et cetera, et cetera.

{{00:07:39}}
So it really speeds up, but we
are not making this a black box.

{{00:07:44}}
We are actually showing to our customers
how it is done, the whole process.

{{00:07:49}}
We are trying to be as
transparent as possible.

{{00:07:51}}
That's really interesting.

{{00:07:53}}
And one of the things that
you mentioned there is this

{{00:07:55}}
sort of element of compliance.

{{00:07:57}}
Yes.

{{00:07:58}}
As we know in the financial
industry, the regulations never

{{00:08:02}}
stand still for that long.

{{00:08:04}}
They're always constantly changing,
especially when we look at your use

{{00:08:10}}
of artificial intelligence and large
language models in that instance.

{{00:08:16}}
When we see the regulations around
that we have the, off the top of

{{00:08:20}}
my head, we have the EU AI Act,
for example, elements of that.

{{00:08:24}}
How is Intrepid Fox's solution
able to keep abreast with the

{{00:08:29}}
rate of change in regulation?

{{00:08:32}}
First of all, regulations, they
never change, they only improve.

{{00:08:36}}
So that's, jokes aside, we believe
that it's very good that we have

{{00:08:43}}
now emerging regulations on the AI.

{{00:08:47}}
Because AI at the moment
is in a gray zone.

{{00:08:49}}
It is always very, uncomfortable
for everyone to develop.

{{00:08:53}}
So it will help even if
regulation is not perfect.

{{00:08:56}}
Now it is good first step
and it will never be perfect.

{{00:09:00}}
So we have to start somewhere.

{{00:09:02}}
Secondly, we looked at these regulations.

{{00:09:06}}
They are not yet in place.

{{00:09:09}}
So let's see, we made some analysis.

{{00:09:13}}
We believe that.

{{00:09:14}}
What we do is what they call minimum risk,
because again, we don't take decisions

{{00:09:21}}
and it is not a decision on driving
car or things which or medical things.

{{00:09:28}}
So it is, in the end, it is just
collecting information, but making it

{{00:09:33}}
much, much faster and much, much more
accurate and efficient than before.

{{00:09:40}}
Hopefully.

{{00:09:41}}
We are good with this regulation.

{{00:09:44}}
Yes, absolutely.

{{00:09:46}}
I think from one thing that I took
from your demo this afternoon, is

{{00:09:51}}
that you have a very strong stance
on KYB and being a KYB solution.

{{00:09:57}}
In terms of your technology
and your software, where might

{{00:10:01}}
we see that extend beyond KYB?

{{00:10:05}}
Yeah, KYB.

{{00:10:06}}
It is a huge topic, it is a huge pain and
we are solving this pain and I believe

{{00:10:11}}
that we are able to help to cut for,
banks and for Fintechs cut billions of

{{00:10:19}}
dollars of cost and for customers save
millions of hours of lost time here.

{{00:10:27}}
So it's already a big one.

{{00:10:30}}
But another one which is also we have
built we are building solution for it.

{{00:10:37}}
It is around transaction monitoring,
because if you are a company which serves

{{00:10:45}}
business customers and you have 10, 000
business customers, it probably means

{{00:10:50}}
that you onboard, I don't know, let's say
500 business customers a month, maybe.

{{00:10:57}}
So 500 onboarding cases,
but you also process.

{{00:11:02}}
There are hundreds, if not millions
of payments, out of which, I don't

{{00:11:06}}
know, maybe 10, 000 payments you need
to check in Transaction Monitoring.

{{00:11:12}}
This is the same.

{{00:11:13}}
In Transaction Monitoring, you
ask customers for documents.

{{00:11:17}}
You made this payment, please
give us invoice or contract

{{00:11:20}}
which stands behind this.

{{00:11:23}}
And this is again the same friction,
same back and forth, and we

{{00:11:27}}
are able to solve this as well.

{{00:11:31}}
I'm afraid there are huge number of
scenarios in financial industry where

{{00:11:35}}
you can be useful with this solution.

{{00:11:38}}
We'll keep working.

{{00:11:39}}
At the moment, KYB is just an
amazing big opportunity for everyone.

{{00:11:43}}
Yeah, it sounds like KYB is
really your bread and butter and

{{00:11:47}}
transaction monitoring is really
coming up fast in your product suite.

{{00:11:52}}
When we look at the notion of AI
and Gen AI, some of the speakers on

{{00:11:58}}
the stage this morning were talking
about how fast it was progressing.

{{00:12:03}}
In terms of Intrepid Fox's own progression
in the fields of AI, in the technology.

{{00:12:08}}
Where might we see you expand to next?

{{00:12:14}}
I think in general, what I'd
like to say, I believe that we

{{00:12:19}}
are, what we see now is amazing.

{{00:12:23}}
It's absolutely the whole change
to our business processes.

{{00:12:28}}
Very soon, people will be able to run huge
Projects by very small teams, thanks to

{{00:12:36}}
this AI, big companies will be able not
to pursue, I don't know, three direction

{{00:12:42}}
of development, but maybe 20 or 30 or 500.

{{00:12:46}}
So it enables so much more things
and there will be a lot of disruption

{{00:12:53}}
and disruption mostly in my,
from my perspective will come.

{{00:12:58}}
If you talk about finance industry, not
to banks or fintechs, but mostly for their

{{00:13:04}}
software providers because the present
software built for institutions is made,

{{00:13:12}}
on different technologies, different
assumptions and different principles.

{{00:13:18}}
So this all will be redone in
the next, I don't know, 20,

{{00:13:24}}
30, 50 years, maybe faster.

{{00:13:29}}
And this is huge opportunity for people
and for companies to participate in this

{{00:13:36}}
thing and to help to move it forward
and to participate in this disruption.

{{00:13:42}}
So basically the project I am running.

{{00:13:46}}
I hope we'll be on the
forefront of this advancement.

{{00:13:50}}
It certainly sounds like it, and
you're, definitely placing yourself

{{00:13:53}}
in the right position by attending
events like Finovate today.

{{00:13:58}}
Roman Zilber, it's been absolutely
fantastic to speak to you.

{{00:14:01}}
Thank you so much.

{{00:14:03}}
And thank you for everybody who's joined
us for this interview with IntrepidFox

{{00:14:06}}
here at Finovate Europe this afternoon.

{{00:14:09}}
If you'd like to find out more,
please head to fintechfutures.

{{00:14:11}}
com and I look forward to seeing you soon.

{{00:14:15}}
Thank you.

Hello everybody and welcome to Fintech
Futures at Finovate Europe here in London.

I am a reporter for Fintech Futures and
today I'm very happy to be joined by

Roman Zilber, the founder of IntrepidFox.

Who is fresh off the demo stage,
literally within the next half an hour.

I was happy to talk and
I think it went well.

Yes I I was in the audience when you
When you were making your speech and

I thought you touched upon some really
interesting points that we're going to

be exploring here in our interview today.

One of the really interesting
things that you did say is the

Sort of delays and the slowness of
the business onboarding process.

Why is it in 2024 that such a
process incurs so many delays?

In fact there are a lot
of research on this.

And they say that it is several
days, seven days on average,

and more than a thousand dollars
the cost of such an onboarding.

It's amazing because again, we are living
in the age of automation, et cetera.

As a banker with 30 years of experience
is that bankers are lazy and don't care.

Actually, bankers are not
lazy and they do care.

But the problem is that during, the
process of onboarding of customers,

some documents and some information
you can retrieve from public sources,

But some documents like invoices,
contracts with suppliers.

Rental agreements, you can only
take from customer themselves.

Because you ask customers ask
you in return what exactly

Then you respond to them, then they
say the wrong document, then you have

One of CEOs of FinTech I interviewed
he told me, Roman, quote me, any

request for Any additional document,
it is always two days, two days.

So this is, why it is so
long and so expensive.

And we have this problem at hand, which
is across all corners of the industry.

Intrepid Fox, where does that sort of
fit into making this process faster?

Yeah, actually, what we do, we single
out this hard part and we automate it.

So something which is only came out
and only available very few months

now, to make this process not hours
or days, but seconds or minutes.

What it does, it actually, it
we made software which reads.

Customer documents, understands customer
documents, follow up questions according

to rules, which we predefined before,
according to criteria, which we gave it.

And this criteria is part
of our underwriting process,

So it both understands the document,
retrieves information from the

Unnecessary human work and concentrate
our qualified people on really

important things like making final
decision whether to onboard the

And you can take it from a wide range
of documents, I believe, as you said.

Yes, this is this is something we aim at
basically all the documents available,

So it is amazing how fast we can read and
understand even complex documents and that

the accuracy there is much better than
accuracy which human, employees deliver.

That's a really interesting intersection
to exist in this, technology

What's really critical in, in
speed is also being accurate.

And you've mentioned that you're
using large language models.

You're also playing with different
forms of AI in your solution as well.

How can you ensure the accuracy of the
outcomes that the solution produces?

When we talk about accuracy, it is
important to distinguish between

two cases there, there, could be
accuracy of, Extracting or retrieving

And here, first of all, we achieve
accuracy better than humans do.

But still, there are some
mistakes, can be mistakes.

To minimize mistakes, we split
the process as much as possible.

We call models only for small questions,
only for parts, not for the whole request.

And we use multi layered architecture,
so we have some algorithmic checks, we

We can sometimes pull
models on on the question.

You have a lot of techniques to
improve and to get to the point when

you don't have hallucinations and
you actually don't have mistakes.

But another part of accuracy is the
accuracy of final decision here.

So basically you got all the
information and then you make

We believe that not regulators, not
Companies and banks, not customers, nobody

is prepared that artificial intelligence
takes decisions on onboarding.

So the human in the loop who takes this
decision, but only this human is ten

But in terms of the human that the
information is passed to, how important

FinTech Futures

Expert Insight

One of the leading topics at this year's FinovateEurope orientated around how Know Your Business (KYB) processes - the bane of many client onboarding endeavours for fintechs and banks alike - and more specifically, what happens when such processes intersect with the latest and brightest technologies.