Name: Back and front office innovations: How is Zurich Insurance Group adopting AI solutions?
Uploaded: 2024-03-11T17:07:08.297Z
Duration: 556.88
Description: Indrek Vainu, Head of Conversational AI at Zurich Insurance Group, shares his experience in AI adoption within insurance, innovating in the back office as well as in customer facing processes. How can one succeed and what are the real challenges that innovators will face on the way? Find out in this interview.

FEU.INT.003.mp4

Back

Or filter by your interest

Browse all our content

Finovate

Unlock all the content with only {{price}}

Indrek Vainu.vtt

English

{{00:00:00}}
David Penn: Hi, I'm David Penn,

{{00:00:06}}
I'm here with Indrek Vainu.

{{00:00:07}}
He is the head of conversational
AI at Zurich Insurance Company.

{{00:00:10}}
Nice to meet you, Indrek.

{{00:00:11}}
Nice to meet you as well.

{{00:00:12}}
I thought we'd start off with just
a little bit about yourself and

{{00:00:15}}
your journey in financial services
that led you to Zurich Insurance.

{{00:00:17}}
Indrek Vainu: Yeah, sure thing.

{{00:00:18}}
So as a background, I'm
actually an entrepreneur.

{{00:00:21}}
I founded a startup company called Alpha
Chat, co founded it about eight years ago.

{{00:00:27}}
And then we built chatbots for customer
support for banks and telecoms.

{{00:00:32}}
We did that for a few years.

{{00:00:33}}
And then in 2021, our company got
acquired by Zurich Insurance Group.

{{00:00:37}}
And since then, I've been now head
of conversational AI at Zurich.

{{00:00:41}}
So I have a entrepreneurial background
now working in a large company.

{{00:00:46}}
David Penn: Interesting.

{{00:00:46}}
Interesting.

{{00:00:47}}
And the transition from chatbot
to AI is a It's a metaphoric

{{00:00:50}}
one, in a lot of respects,
technologically as well as personally.

{{00:00:53}}
On that note, can you talk to
us a little bit about some of

{{00:00:55}}
the generative AI projects that
you're working on, for example?

{{00:00:58}}
Indrek Vainu: Yeah, sure.

{{00:00:58}}
So what we do is we focus a lot on,
customer support automation with chatbots

{{00:01:04}}
the way how to have customers easier
way to interact with us and the whole

{{00:01:10}}
company and get information in terms
of how we work with AI machine learning

{{00:01:16}}
has been around for a while, right?

{{00:01:17}}
And I think the transformer models that
came about with ChatGPT a year and a

{{00:01:21}}
half ago have really changed the way the
technology, what it's capable of doing.

{{00:01:27}}
The thing that at Zurich, what
we're doing is we are taking...

{{00:01:32}}
we're experimenting a lot, but we're
taking a cautious approach toward

{{00:01:34}}
generative AI in the way that all of
our customer facing solutions have a

{{00:01:39}}
human in the loop, so we don't allow
generative AI to produce answers for

{{00:01:43}}
external customers while we do build
solutions, of course, internally for

{{00:01:47}}
our employees to augment their work.

{{00:01:51}}
There we use a generative AI, but for
outside customers, we focus more on

{{00:01:56}}
providing people with the solutions
that the AI doesn't generate on its own.

{{00:02:02}}
So there's no autonomous decision making.

{{00:02:05}}
David Penn: It's interesting.

{{00:02:05}}
And I appreciate the idea that you talked
about both the applications for AI.

{{00:02:08}}
Sort of customer facing
as well as internal.

{{00:02:10}}
I think sometimes we forget that
there's really something for both.

{{00:02:13}}
I'm wondering if maybe you could
tell us a little bit about some of

{{00:02:15}}
the pain points solved by AI in both
instances, some of the ways that it's

{{00:02:18}}
able to, again, with this human in the
loop way, help with customer facing

{{00:02:21}}
and also some of the things that's
helping your own employees on the back

{{00:02:23}}
office do some of their jobs better.

{{00:02:25}}
Indrek Vainu: Yeah, absolutely.

{{00:02:26}}
So on the employee side, I think
what generative AI is really

{{00:02:29}}
good at is research, right?

{{00:02:31}}
So finding information and you can
imagine the insurance industry is heavy

{{00:02:35}}
on all kinds of text ready documentation.

{{00:02:38}}
When you're looking for something,
it's needle in a haystack.

{{00:02:40}}
Now, generative AI is really good
at finding those needles, right?

{{00:02:44}}
So what took people a long time and pretty
significant budgets to do with IT a few

{{00:02:49}}
years back, is now a lot easier, right?

{{00:02:52}}
So it's a way to augment human
work and give them superpowers.

{{00:02:56}}
Now when we look at external customer
facing applications especially

{{00:03:00}}
with chatbots there's a way how
you can understand customers

{{00:03:03}}
better when they write things.

{{00:03:05}}
But still the answers you give them
are pre written because you need to be

{{00:03:09}}
consistent in your communication towards
customers and at Zurich we also have our

{{00:03:13}}
own AI framework that we have adopted
where basically we put a lot of emphasis

{{00:03:18}}
into what type of AI solutions we we build
and how they work, making sure that they

{{00:03:24}}
are transparent and  and don't have bias.

{{00:03:26}}
David Penn: Good.

{{00:03:27}}
I wanted to ask you something else,
given that your background in chatbots,

{{00:03:30}}
is there anything particularly with
that background and experience that

{{00:03:32}}
you're really able to bring to bear
currently with the work that you're

{{00:03:36}}
doing with customer facing applications?

{{00:03:37}}
Indrek Vainu: Yes.

{{00:03:37}}
So when we started in 2017, there
was also a hype cycle with chatbots.

{{00:03:42}}
Many don't remember that, but there was.

{{00:03:44}}
And then in 2019, the hype cycle
died and we went on a downslope.

{{00:03:48}}
Right now in 2022, we went up again.

{{00:03:51}}
So the perspective that I
have is technology goes up

{{00:03:54}}
and down, most likely forever.

{{00:03:56}}
Sure.

{{00:03:56}}
So I think now we have passed the peak.

{{00:03:58}}
Maybe we're a little bit coming down
on the hype cycle, which is good.

{{00:04:02}}
And I think the key thing I've
learned over the years working with

{{00:04:05}}
chatbots is that really that you
have to start with the business case.

{{00:04:09}}
So you have to start,
what is the business case?

{{00:04:11}}
And then look at the technology because
the technology is cool and exciting,

{{00:04:16}}
but if it doesn't really apply well to
your business case, then you can also

{{00:04:19}}
use some of the previous generation
technology really well as well.

{{00:04:23}}
So that's why we look at where it
makes sense to apply this, essentially.

{{00:04:28}}
David Penn: It's a
really interesting point.

{{00:04:29}}
We had a presentation earlier,
a very interesting one about AI.

{{00:04:32}}
And one of the things a lot of
the AI presentations talk about

{{00:04:35}}
is that fast adoption rate.

{{00:04:36}}
They compare it to all these
other sorts of technologies that

{{00:04:39}}
also had fast adoption rates,
but nothing comparing to the AI.

{{00:04:41}}
And I think it's an important point
that you make, that simply because

{{00:04:44}}
you have this rapid adoption in the
populace, doesn't mean that this rapid

{{00:04:48}}
practical applications going to work for
businesses that you do still need to make

{{00:04:51}}
a business case for AI, just like you
would any other technology, it seems.

{{00:04:55}}
Indrek Vainu: Yeah, exactly.

{{00:04:55}}
So I think it really comes
down to the control, right?

{{00:04:58}}
So if you have a tool for
internal employees to do search.

{{00:05:01}}
And you give people the bot says here
you have five options and then the

{{00:05:05}}
human decides what to do with that.

{{00:05:06}}
That's amazing, right?

{{00:05:07}}
Because otherwise you would have
to come up with those five options.

{{00:05:09}}
Now there's a machine that
does it in two seconds.

{{00:05:12}}
However, when you present that to
a customer it has to be a solid

{{00:05:16}}
answer that you can stand by, right?

{{00:05:17}}
And you cannot just give answers
that are somewhat correct, which is

{{00:05:21}}
where all the hallucinations come in.

{{00:05:23}}
So that's why you differentiate between
internal and external solutions.

{{00:05:27}}
David Penn: Really interesting.

{{00:05:27}}
I wonder if you tell us maybe some of
the success stories and things that

{{00:05:30}}
you've put together, things that you
maybe we're setting out and weren't

{{00:05:34}}
sure that we're going to come past,
but you managed to implement them and

{{00:05:36}}
really achieve some of your goals.

{{00:05:38}}
Indrek Vainu: Yeah.

{{00:05:38}}
So clearly we have a lot of cases
internally that we're working with and to

{{00:05:44}}
say that we've been doing it successfully
for three years, I would be lying because

{{00:05:47}}
the technology is a year and a half old.

{{00:05:48}}
So it's like the playing field is open.

{{00:05:51}}
Everybody's dabbling around and
figuring out what to do with it.

{{00:05:54}}
But some of the cases we've been doing
is really for our internal staff to go

{{00:05:58}}
through a lot of documents internally
and find information about, Some, let's

{{00:06:02}}
say a loss adjuster reports or things
that come from previous cases, and

{{00:06:07}}
then how you can use that data easily
in your future businesses that you do.

{{00:06:12}}
That's one way.

{{00:06:12}}
Or we look at ways how we can compare
sort of policy worded wordings from one

{{00:06:17}}
language to another, and then make sure
that the terms are consistent across

{{00:06:21}}
languages on the same policy, for example.

{{00:06:24}}
David Penn: I really appreciate
the points you're making,

{{00:06:25}}
too, about this consistency.

{{00:06:26}}
I think so often, to the extent
that we approach AI individually,

{{00:06:29}}
we don't think about these sort of
things again, that business case.

{{00:06:32}}
But you certainly can't have, just like
you couldn't have a human representative

{{00:06:34}}
giving one set of advice to one person
and something else to the other one.

{{00:06:37}}
It would just make no sense.

{{00:06:38}}
And that the AI could do just
the same thing and need to

{{00:06:40}}
make sure that doesn't happen.

{{00:06:42}}
Can you talk about some of the
other challenges that you face

{{00:06:44}}
implementing the technology that
you had to work your way through?

{{00:06:47}}
Indrek Vainu: Yeah, sure.

{{00:06:47}}
So the challenge is, of
course, there are many.

{{00:06:50}}
I think I was giving the talk
yesterday and I said the best thing

{{00:06:55}}
about generative AI is that it's so
easy to build very impressive demos.

{{00:07:01}}
And the worst thing about gen
AI is that it's so easy to

{{00:07:03}}
build impressive demos, right?

{{00:07:05}}
So the idea is that you can
build something really quickly

{{00:07:07}}
that's powerful, it's great.

{{00:07:09}}
But I think one thing that shouldn't
underestimate the intricacies

{{00:07:14}}
that go into building a demo.

{{00:07:15}}
Building something out at scale
and dealing with the edge cases.

{{00:07:19}}
I think with a lot of also big tech
companies see how they're releasing

{{00:07:22}}
products and then pulling them from
the market, or even before they

{{00:07:25}}
release products, they actually take
a lot of time internally to test

{{00:07:28}}
them because they've learned that
once you release it out to the wild.

{{00:07:33}}
It is called generative for a reason.

{{00:07:35}}
It just does things on its own, which is
really hard to explain what it does and

{{00:07:38}}
why, and you have no control over it.

{{00:07:40}}
But once you unleash it to the wild.

{{00:07:43}}
You need to have a certain
amount of control, right?

{{00:07:44}}
So I think that's the
learnings  that we've had.

{{00:07:49}}
And, but it's just it's, a
very, it's a great technology.

{{00:07:52}}
And of course I encourage everyone to
play around with it and work it, look

{{00:07:55}}
at your business cases and then just
apply it and take it step by step.

{{00:07:59}}
Because as I said,  the space
is like year, and a half old.

{{00:08:04}}
So everybody's a novice
in this area in general.

{{00:08:07}}
David Penn: Yeah.

{{00:08:07}}
That's a good way of thinking about it.

{{00:08:08}}
I was thinking maybe by way of wrapping
up, you just provided some suggestions,

{{00:08:11}}
so to speak for, companies institutions
that are looking to get involved  in AI.

{{00:08:16}}
I don't know which ones aren't.

{{00:08:17}}
Are there any other pieces of advice,
whether they're things that you've

{{00:08:20}}
observed, things you've experienced that
you would give to companies that are

{{00:08:23}}
looking to go down that road, which is,
Hey, here's something we've discovered.

{{00:08:26}}
Here's something you
might want to consider.

{{00:08:27}}
Indrek Vainu: I think in general,
I would say  look at your business

{{00:08:32}}
problems and then look at those and
see if gen AI is good at solving

{{00:08:38}}
those and then apply it there.

{{00:08:40}}
In many cases, all the technology works
perfectly fine, but you have now this

{{00:08:45}}
sort of it is, a game changing technology
compared to when I started with chatbots.

{{00:08:50}}
So definitely play around
with it, see how to apply it.

{{00:08:53}}
And and then just try it out.

{{00:08:55}}
I think that's the only
way to try things out.

{{00:08:57}}
And I think once we will sit here a
year from now, we'll be much smarter

{{00:09:01}}
and the year from now much smarter.

{{00:09:02}}
So the learning curve has just begun.

{{00:09:04}}
David Penn: No, that's another reason
to have you back a year from now.

{{00:09:05}}
We can talk about it some more.

{{00:09:08}}
Wonderful to meet you.

{{00:09:08}}
Wonderful to talk about what's going
on with you and Zurich Insurance.

{{00:09:11}}
Thank you very much.

{{00:09:11}}
Thank you.

{{00:09:12}}
And thank you very much as well.

He is the head of conversational
AI at Zurich Insurance Company.

I thought we'd start off with just
a little bit about yourself and

your journey in financial services
that led you to Zurich Insurance.

So as a background, I'm
actually an entrepreneur.

I founded a startup company called Alpha
Chat, co founded it about eight years ago.

And then we built chatbots for customer
support for banks and telecoms.

And then in 2021, our company got
acquired by Zurich Insurance Group.

And since then, I've been now head
of conversational AI at Zurich.

So I have a entrepreneurial background
now working in a large company.

And the transition from chatbot
to AI is a It's a metaphoric

one, in a lot of respects,
technologically as well as personally.

On that note, can you talk to
us a little bit about some of

the generative AI projects that
you're working on, for example?

So what we do is we focus a lot on,
customer support automation with chatbots

the way how to have customers easier
way to interact with us and the whole

company and get information in terms
of how we work with AI machine learning

And I think the transformer models that
came about with ChatGPT a year and a

half ago have really changed the way the
technology, what it's capable of doing.

The thing that at Zurich, what
we're doing is we are taking...

we're experimenting a lot, but we're
taking a cautious approach toward

generative AI in the way that all of
our customer facing solutions have a

human in the loop, so we don't allow
generative AI to produce answers for

external customers while we do build
solutions, of course, internally for

There we use a generative AI, but for
outside customers, we focus more on

providing people with the solutions
that the AI doesn't generate on its own.

So there's no autonomous decision making.

And I appreciate the idea that you talked
about both the applications for AI.

Sort of customer facing
as well as internal.

I think sometimes we forget that
there's really something for both.

I'm wondering if maybe you could
tell us a little bit about some of

the pain points solved by AI in both
instances, some of the ways that it's

able to, again, with this human in the
loop way, help with customer facing

and also some of the things that's
helping your own employees on the back

So on the employee side, I think
what generative AI is really

So finding information and you can
imagine the insurance industry is heavy

on all kinds of text ready documentation.

When you're looking for something,
it's needle in a haystack.

Now, generative AI is really good
at finding those needles, right?

So what took people a long time and pretty
significant budgets to do with IT a few

So it's a way to augment human
work and give them superpowers.

Now when we look at external customer
facing applications especially

with chatbots there's a way how
you can understand customers

But still the answers you give them
are pre written because you need to be

consistent in your communication towards
customers and at Zurich we also have our

own AI framework that we have adopted
where basically we put a lot of emphasis

into what type of AI solutions we we build
and how they work, making sure that they

are transparent and  and don't have bias.

I wanted to ask you something else,
given that your background in chatbots,

is there anything particularly with
that background and experience that

you're really able to bring to bear
currently with the work that you're

So when we started in 2017, there
was also a hype cycle with chatbots.

And then in 2019, the hype cycle
died and we went on a downslope.

So the perspective that I
have is technology goes up

Maybe we're a little bit coming down
on the hype cycle, which is good.

And I think the key thing I've
learned over the years working with

chatbots is that really that you
have to start with the business case.

So you have to start,
what is the business case?

And then look at the technology because
the technology is cool and exciting,

but if it doesn't really apply well to
your business case, then you can also

use some of the previous generation
technology really well as well.

So that's why we look at where it
makes sense to apply this, essentially.

David Penn: It's a
really interesting point.

We had a presentation earlier,
a very interesting one about AI.

And one of the things a lot of
the AI presentations talk about

They compare it to all these
other sorts of technologies that

also had fast adoption rates,
but nothing comparing to the AI.

And I think it's an important point
that you make, that simply because

you have this rapid adoption in the
populace, doesn't mean that this rapid

practical applications going to work for
businesses that you do still need to make

a business case for AI, just like you
would any other technology, it seems.

So I think it really comes
down to the control, right?

So if you have a tool for
internal employees to do search.

And you give people the bot says here
you have five options and then the

FinovateEurope

Case Study

Best Practice

Indrek Vainu, Head of Conversational AI at Zurich Insurance Group, shares his experience in AI adoption within insurance, innovating in the back office as well as in customer facing processes. How can one succeed and what are the real challenges that innovators will face on the way? Find out in this interview.