Modernization Hub

Modernization and Improvement
AI-first Insurance: Opportunities and Challenges

AI-first Insurance: Opportunities and Challenges


well good morning everyone thank you so
much for coming I’ll be talking about AI first insurance and opportunities and
challenges it introduces to the industry of insurance in an interview at Davos
last year Sergey Brin co-founder of Google mentioned that the the fast
developments in deep neural Nets has been surprising him despite him having a
front-row seat to the whole show and and he’s not alone many scientists in the
field who were actively doing research in that field couldn’t see it coming at
that pace and it’s evident at the excitement about the the development is
evident in the number of times terms related to AI is getting mentioned in
corporate earning calls like for example here you see terms such as machine
learning showing exponential growth in terms of how often they get mentioned in
corporate earning calls but of course that creates a lot of room for many
predictions about what a I can and cannot do in different industry
verticals including insurance but of course in my talk I try to steer away
from predicting the future of AI because even people who were more genius and
smart in the field made some of the mistakes in terms of their prediction
the famous one is Alan Turing one of the brilliant minds of AI history in his
famous 1950 paper talks about how he thinks the field of AI developing and in
part of the paper he talks about if I actually read from the paper said I
should be surprised if more than 10 to the power of nine bits was required for
satisfactory playing of the imitation game so this is the paper actually where
Alan Turing introduces the imitation game and he predicts 10 megabytes is
what we need in terms of memory for computers to play the imitation game and
have a satisfactory performance there and the number comes from the size of
Encyclopedia Britannica volume 11 so so basically prediction is a tough game in
terms of future of industry in light of AI so we steer away from it and most of
the stuff we talk about is in light of what AI can do today and is almost
predictable for the next few years so one framework that might help
talk about the development of AI in various industry verticals is the one
that comes from the book prediction machines by Ajay Agrawal and co-authors
what they argue is that the cost of prediction has dropped significantly in
the past few decades so basically today we see prediction happening very cheaply
and very very at a very low cost and their argue that as a result of this
what we are gonna see is that many industries will try to rethink and
reimagine their industry to benefit from this cheap economy so for instance he
argued that similar thing happened in in photography so photography was a
chemistry problem and computing became so available so cheap then that turned
for the photography from a chemistry problem into a computing problem now
we’ve got photography devices on our mobile phones and and and and various
computing devices so he says I’m the same type of trend you’re gonna see for
example HR was a human problem now it’s becoming an AI prediction problem
driving was a human expertise and experience problem is becoming a
prediction problem and the list continues as you saw from the previous
panel cybersecurity was a heavy computer science computer skill set problem
dropped race basically saying they made it a prediction problem and fraud as was
saying from on PETA was saying used to be a huge amount of human
looking at people in the bank kind of problem and now has become an AI on
prediction problem so then what does that mean for insurance well the good
news for the industry is that insurance was a prediction problem centuries ago
in 1744 to ministers from Church of Scotland observed the problem they had
which was after the clergyman died their family was left alone mean in need of
some financial support so they thought it would be great if they create a fund
where different clergymen contribute a small amount to the fund and as a result
of that that provides the money to support the families that are left
behind after their death and that became the foundation of what we know today as
life insurance and in order to solve for that problem they approached a professor
at the University of Edinburgh and the three of them together started
thinking about how to solve this problem and they use what is known as
Bernoulli’s law of large numbers to build a model of of how mortality is
going to develop over time in a portfolio of people who might or might
not die at some point in soon future and and if you think about it like if you
use a simple example of a coin toss what Bernoulli’s law says is that for
each toss of a coin it’s very difficult to predict is it going to be a head or
tail but if you have hundred coin tosses you can estimate how many of them are
going to be had how many of them are gonna be tail but it’s very difficult to
say exactly which one of them is head and tail so the old paradigm of
insurance was that I have a portfolio of rest at a portfolio but I want to know
what the risk is like so therefore insurance started to become a prediction
problem but still doesn’t mean it cannot benefit from it if there is no
opportunity for rethinking and reimagining what AI can do to make to
explain it a little bit clearly one way to think about insurance today is that
of every dollar an insurance company makes like if you look at a recent
report by Swiss Re in commercial insurance somewhere around 50 cents of
that premium dollar they make is paid back as loss ratio basically to pay for
people who make things slightly more than 20 cents of that dollar is spent on
acquiring customers and whatever is left somewhere around 20 cents is going to be
split between internal operation cost and underwriting profit so today the
part of this value chain that is using is becoming has been historically using
prediction is mostly around estimating what that loss ratio is a huge part of
it is there and of course that related to the reserving and other concepts as
well there is a huge amount of prediction going on and I think if you
see at the top as well another thing insurance companies do is
that they also add more to the dollar through investing that asset they earn
through premiums in in different asset classes so so that’s the kind of
framework that insurance is built on today and and then if you go back to the
coin toss example what happened in light of all these prediction became cheaper
if you other things also happen which is
prediction predictions are broadly available so in a way things that we
could not predict in the past seem to be predictable now so in a way it’s almost
like not only we can predict on a hundred coin toss how many of them are
going to be head or tail but also we almost can predict right now for each
coin toss is it a head or tail not only that but in a lot of Sciences that is
related to the risk that we are ensuring like in medicine in cyber sciences and
so on we also have the science that could be paid with a precise prediction
to lead to prevention so all these things together I think are still new
opportunities for insurers to start to benefit from and in light of that I
think we could envision a future in insurance where through better use of AI
you could drive the loss ratio down you could improve the acquisition ratio cost
an internal operation similarly is likely to go down and that leads to a
higher margin an insurance profit and also similarly the world of investment
that insurance are operating in is full of complex data and and complex
decision-making which again could benefit from this new economy of
prediction so therefore the future world of insurance could lead to a higher
margin which could lead to more innovation and that could result in
magnificent impact in the society so let me give a couple of examples you saw
doctors earlier so imagine if somebody is insuring for cybersecurity a
portfolio of clients historically it’s been about can be estimate the amount of
claim that is going to come from that portfolio now in light of technologies
like the one that dr. is introduced you can actually mitigate some of these
risks you through the prediction machines that you have you can predict
that this pattern of behavior in the network is risky therefore we can stop
it from happening and so on so I think these are the kind of stuff that could
drive the loss ratio down so that’s what I think is an opportunity that
traditionally when people think about prediction and insurance they think
about actuarial science and reserving and pricing where I think that’s a very
interesting emerging opportunity in light of the new economy of prediction
but of course it’s not simple it’s a very challenging process because a lot
of the time you hear companies that say are the old good old-fashioned company
that used to be we have got a few apps we’ve
got a few neural nets therefore we are AI first but it’s not really that it’s
that a lot more complex than that it requires a fundamental rethink and
reimagine ation of the business to to truly benefit from AI one way to explain
it maybe is the autonomous vehicles usually you see the pilot faces of
autonomous vehicles where there is a car like this red one here that has been
built pre AI the car looks just like a normal car that we are used to in the
past few decades and then they thought Oh autonomy is an interesting thing why
don’t we put a bunch of cameras around the car on top of the car and so on and
so on so we can eventually see if this car can drive autonomously the problem
is imagine if autonomy works what’s the implication for that company they need
to go through a fundamental rebuild of their manufacturing line because now it
means redo the process of manufacturing and everything because you need to build
a car that this camera and these sensors are embedded within it so it’s almost
like multiplying the cost of the two phases of a cost to get to the ultimate
autonomy goal whereas what Tesla did is a slightly
different so in Tesla when you look at the Tesla built in 2000 14 15 16 and so
on you don’t see any sort of cameras hanging around the car or something
because day one they thought autonomy is inevitable it’s more a question of if
it’s more a question of when rather than a question of if so therefore from day
one they built the car with that in mind so the car comes equipped with all the
sensors and so on that you might possibly need to achieve all the way to
level five autonomy and and then later recently you saw the Tesla release a
software update that made the car autonomous and that’s the essence of
designing based on AI if you have the good imagination of where AI can take
your industry you can build the car from day one that way so so that brings us to
the world of AR first digital transformation it term digital
transformation is a term that many industries are talking about today but I
think AR first digital transformation is fundamentally about the result of
reimagining every touch point and hence the overall business model in light of
machine learnings ability to inform and assist the users and the customers
employees of a business I think that’s where the true north is so if your
company is going through a digital transformation it’s a good time to think
about it do you want to have a no AI digital transformation and put those
cameras on top of it later and create a second cost when that shows some level
of success or you want to start from now rethinking and reimagining that process
I think that’s a good question for insurance industry as a whole to think
about but again as I said it’s not easy it’s gonna create a few challenges for
the industry the first one is talent I mean I like this famous coat that I save
here and there every now and then his talent can hit a target no one can head
and genius it’s a target no one can see and I think it’s very important for
insurers as well as many other industries to invest in talent because
without talent everything else that stays Assyria and you just become good
at buying services from vendors the second thing is a strategic data
acquisition I mean Google is a fascinating story in a sense that for a
decade they’ve been collecting so many data sets from so many users that
eventually through a huge amount of diverse data we started to see that the
old algorithms that the theory of it exists at decades ago is started to work
and that shows the power of data without data how can you build machine
intelligence and I think that’s a very important thing for insurers as well to
be mindful of and the third one is of course the concept of a bee testing and
fast decision cycles when your ecosystem goes digital you need to be able to test
a lot of things quite fast and that’s where it becomes very important to have
clear KPIs of what better version of this feature means and through that KPI
you could quickly test an Oreo a sand or a bee and pick your winner one and
continue to improve the system incrementally that way and also given
that you’re gonna have so many of these tests you cannot ask the CEO every time
for permission for oh we want to move the logo of the app here can we get
approval for this and that’s the reason you see in a lot of internet companies
there’s a huge amount of push towards power going down to the PM’s and
engineers because instead of a business with so many strategic big decisions you
end up having a digital business with many small decisions happening very fast
very frequently that requires more trust more power shift to engineers NPM and
and these are the four things that I think the industry needs to invest
maybe I should have started with this slide but actually now’s a good time to
maybe describe what insurance is so insurance in essence is the business of
transferring risk this optimal owner now in light of AI and and various
scientific developments that we are seeing how can we define the optimal
owner of the rest maybe I should elaborate a little bit more in the world
of general insurance right now there are estimates that are saying there is
somewhere around ten trillion dollar of risk out there and if you look at the
size of premium an insurance industry we covered so far around 1 trillion of it
so there is 9 trillion of risk out there that exists that people are experiencing
but there is no formal insurance coverage of that so the question is
who’s off to my owner of that risk can a I help the current insurers to become
better at owning these risk and and and be happy to have the reference first to
them another area which is very interesting now I’m kind of going a
little bit to my Oxford roll but again with an insurance in mind today if you
look at the global spend on health care it’s somewhere around 10 trillion this
is not estimated this is like really what the money we are spending today in
health care and that’s almost 10 percent of the global GDP and various players in
the healthcare ecosystem are built based on certain view of the world and and and
I’m going to just share a couple of examples of how these views are being
challenged and but the list can continue it’s a much bigger list one example is
we are fundamentally seeing different patients if if you look at the trend in
medicine right now in the past few decades now we have more people living
into their 60s and older than those who died before their 60s and that’s a good
news I mean that’s a moment that medical science should feel proud but at the
same time it introduces a fundamentally different challenge to to the world of
medicine so here is a paper that came out of our lab last year which shows
basically just look at the red line with 5 plus so what it said we looked at 5
and a half million people patients in NHS and we looked at the time at which
they have their fur incidents of a cardiovascular disease
like a stroke or something like that and at that point we counted other diseases
they had if you look at the x-axis is showing between 2000 all the way to 2014
so in the past 15 years almost what we see is that there is a 4x increase in
the number of other diseases people have when they get their first cardiovascular
disease is it the same patient if your underwriting manual is from 15 years ago
are you really talking about the same patients anyway it’s different
demographics people are living older when people are older they are
fundamentally a different patient and that’s very important for medical
science as well as insurance industries or other players in the industry to be
mindful of the second thing is this notion of alternative data in in
healthcare that is becoming more and more successful in showing pathways for
impact there is a recent paper out of Imperial that have been basically saying
if you tell me roughly your postcode I could go look at the images of your
streets through Google map for example and try to see if I can predict things
about your house as well as other things and this shows the amount of correlation
you see almost 80% correlation in the graph that is trying to look at the
pictures of your neighborhood not even in front of your window just roughly
your neighborhood pictures and can that picture be predictive of your health
deprivation and disability it’s a it’s a shocking result and and that’s actually
a London example here but they tested the results in Birmingham and Manchester
and a few other cities and it actually holds up quite nicely so what it means
is that your medical data is not just about your MRI images or it’s not just
about your genetics data or it’s just enough not about your blood test it’s
much more complex than that and that’s where I think a lot of the under
existing assumptions about medicine will be challenged and I’m not sure how
visible this chart is but basically what it shows here is a list of FDA approved
AI in medicine and the list couldn’t I mean did you see from here almost you’re
in 2018 all the way to almost now and the pace of number of AI fda-approved
ai’s for various parts of medical value change is increasing dramatically
and and it seems to be working quite nicely that FDA is happy for these
things to be released into practice and they cover ranges of medical practices
psychiatry of the ophthalmology and the radiology neurology everything there’s a
huge range of disciplines that are being impacted by these AI and I think that’s
where fundamentally you’re seeing that in medicine there is a high chance that
we are moving from like if you think about what insurance in medicine was it
was about predictive likelihood of a risk and then pay for the claim when it
comes in life of a new prediction economy it seems to be moving more and
more towards predict the likelihood of a risk try to prevent and at make that
prediction personal and try to prevent at a personalized level and that’s I
think the the transformation we seem to see and in light of that I hope that you
leave this talk with one question that you tried to answer which is after all
we see and a lot more that is out there that I didn’t get a chance to talk about
who is optimal owner of the medical risk and and of course the same questions
apply to many other domains of insurance with that I thank you for your time and
coming to the stop yep thank you thanks very much Reza so
um hello everyone oh that’s very interesting isn’t it
to get that kind of very specialist and analysis of business models and also
that health insurance angle we’re now with this brilliantly
mixed and diverse panel in terms of their experience and what what their
history is and what their jobs are today you can see I’m not going to waste any
time by saying what you can actually read we’re gonna now dig into the
nitty-gritty of what it is to actually think about and implement the right
kinds of changes using AI to AI technologies and actual insurance
companies with actual customers and actually the customer is going to join
this discussion as well because that’s going to be one of the focuses of our of
our discussion so very briefly then Orlando to my far right chief data
scientist Aviva ok I’ve just broken my promise
about not saying what you can read Orlando got some really interesting
experience actually which is going to come out in this panel shortly but I was
really impressed by the fact he was in Kings Cross 25 years ago enjoying the
social scene when I was hurrying from point A to B with my head down next to
him is Ashish now I I actually just referred to Ashish to his face as the
che guevara of AI in insurance and he seems to be quite happy with that
actually so you can make up your own mind when you hear him next to him is
Emma and Emma’s looking at me like that because she knows I’m going to say she
was really impressed that Orlando Bloom was here last night and actually I was
quite impressed as well I didn’t see him though and right next to me is Electra all abour a big focus on privacy bias and
the health of the whole ecosystem especially the end customer now what
you’ve got here is a panel of people who are actually although two of them work
for have big jobs in big insurance companies or actually Outsiders working
from the outside in that’s a really important perspective and actually with
Emma and Electra you also have some people who are deeply interested in
insurance but who are also looking at us and you from the outside so what the
guys are going to do is they’re gonna take a couple of minutes first what I
asked them to do was to pick one thing of the many things that they could talk
about that they’re doing now to change the status quo and why they think it
matters in relation to this bigger topic okay so we’re going to start with
Orlando so she should you give okay yeah Orlando what everyone how we’re doing
today good yes so one thing that’s difficult so and some of you might know
Aviva you’re probably familiar with the brand you might not know that we were
founded in 1696 so we’re an extremely old company and so Reza was talking
about prediction and so in the 1700s it becoming a big problem we were there at
the very start so prediction has always been part of the Aviva business for the
for the last few years Aviva within the industry has become known for again some
of the things that Reza has been talking about digital transformation disruption
the fact that we’ve shifted loads of people to Hoxton we only lots of Hopsin
square in church and we’ve also invested in talent these are kind of new skills
digital marketing data science machine learning at scale but really the
important thing has been about customers for us because customers we know don’t
trust insurers if insurance has become a it has a very corporate perception
amongst customers and people think their customers aren’t loyal to insurers and
customers think that we rip them off and we’ve tried to actually fall in love
with our purpose again because to us it’s not about ripping customers off
insurance should be about protecting the things that customers find most valuable
at some of their toughest times and so if we think about where we’re driving
customer value and customer outcomes we don’t expect customers to know or care
about machine learning artificial intelligence digital disruption we do
expect them to care about the fact that we don’t penalize them at renewal by
putting prices up we’re trying to streamline the claims process we’re
trying to take the time it takes to underwrite people down from four weeks
down to three seconds those are the things that we are using
AI machine learning to do and it’s all centered around trying to drive value
from customers so that’s the big thing for me so falling in love with the
purpose I like that Ashish I so I axe Excel which is the
commercial and specialty division of AXA and so we have sort of big risks around
the commercial insurance but also catastrophe risks or hurricanes
wildfires the kind of risk that touches a lot of industries countries big
populations and quantifying and really understanding that risk is hard we’re
also at the same time looking at the nature of evolving business models so we
have shared economy we have gig economy we have you know there was conversations
about evolving cyber risk and the whole connected nature of risk you know we
talk about smart cities all of these while really exciting propositions also
create an environment a really complex risk which is hard to understand and
what we’re trying to do is trying to unwind to fundamentally change our
ability to learn and improve the risk posture understanding a risk posture and
what does risks really mean and that in these situations and by monitoring and
helping working very closely with our clients to orlando’s point we’re able to
then create mitigating strategies and ways of working in partnership with the
clients so use of technology and data to then drive a better understanding and
helping manage risk is where we’re moving away from the very transactional
nature of just paying for the cover to then helping manage and monitor and be
partnered with the clients that’s where we are so I lead marketing at a technology
company called Satori and we’re focused on helping commercial insurers to
transform the accuracy and efficiency of their underwriting using artificial
intelligence and I wanted to start off by giving our definition of what
artificial intelligence means to us because I’m aware that it’s a very
loaded term but at site Oriya the way that we define AI is it’s the computer
having an ability to make a prediction based on finding patterns and data and
it’s Ettore we make a lot of predictions we make predictions about will this
building catch fire and if it does catch fire how much will the claim post and we
serve these predictions up to underwriters to help them make really
fast better informed decisions about the risks that they are underwriting and I
think the problem that we’re really solving here is we’re making and routing
a lot more efficient so today if you want to buy car insurance or travel
insurance you can purchase this in a matter of minutes but if you want to buy
business insurance that could take up to seven days and this means that the
customer experience isn’t great and this is because the and routing process
itself and commercial insurance today is still extremely manual so underwriters
are spending up to 50% of your time actually going out and gathering
information and assembling data sets and then trying to analyze the risk based
off that what we want to do at Cytora is serve them up these predictions so
that they can make really fast accurate decisions and that’s really great
because it means the customer gets a better experience they get a fair price
and it actually reduces the cost of delivering insurance for the insurance
company so that’s kind of what we’re focused on at Cytora is really making
risk transfer a lot more effortless making insurance a lot more easier to
obtain for businesses and that yeah that’s what we were working on hi
everyone I’m Elektra I’m the founder of the law boutique and I founded the law
boutique in 2017 after spending 10 years in large organizations and witnessing
the inefficiencies of the legal industry so we focus on startups and
tech startups that are very fast-growing and the main focus is actually data and
how do we create the companies of the future that are really tight taking data
ethics seriously and ensuring that they’re doing their bit for society in
the future so the data ethics is the new
sustainability and corporate responsibility I think so our focus is
very much on making sure that these companies understand that what the
commercial implication of that is and also how they need to be responsible in
the future great thanks very much so the first
question then and we’re going to go to the outside the true Outsiders first
Ashish Orlando so a little bit of patience my first question is right from
your different perspectives we’ll start with you Emma in a second
okay we talked endlessly in Insurance about the challenges that we face in not
just understanding technologies but working out how this changes what we do
and how we do it and why we do it just share a couple of examples from your
perspective of educating and selling stuff to insurers right of what you see
in terms of how people are thinking about this and what they’re actually
doing about it challenges that we face etc especially from a marketing
perspective as we’re asking people to really trust a technology that they’re
not very familiar with to guide their decision-making and that’s quite a scary
thing for some people and that’s definitely a big challenge that we have
to overcome and our marketing team spends a lot of time thinking about how
we can build a brand and how we can position a product that people will feel
confident and comfortable using and that people trust and for us it’s we focus a
lot on transparency so explaining exactly how our products work and when
they perform really well and when the IAA doesn’t actually perform very well
so a really good example of this is in the SME market the AI performs really
well because there’s a lot of data available for us to use to understand
those risks where is that the larger more complex end of the market the a
doesn’t perform as well because those risks are harder to assess and there’s
not as much information available so we’re super transparent with customers
about what we can and can’t do and that definitely helps to build trust thank
you so we’ll come back to the issue of how much data is enough day so hopefully
if we have enough time okay so a lecture then from your perspective looking in
insurance challenges and opportunities I’m guessing you’re going to want to
talk about privacy but go for it yeah yes I think there are some existing
issues namely gdpr which is probably not everyone’s
favorite word but it my question is is it actually a legislation that’s already
outdated in light of the technological developments currently and are you able
to build a truly GDP are compliant AI system or blockchain or anything that’s
so innovative with GD P R in mind and privacy by design in mind so Ashish
you’re holding the mic so you can obviously pick out some challenges and
opportunities from your perspective as well or pick up one of the things that
Emma Elektra has said about yeah you’re in an insurance company trying to make
this big change challenges and opportunities from your perspective yeah
so I am nearly two years in to the company but also being relatively new to
the insurance industry and as an outsider coming in or and also in some
it to some extent as an academic coming in there is a there’s a lot of
conversations that I have which don’t talk about technology is more about
demystifying the whole world of what what is emerging technology mean what
does its adoption and what is that real what does that think that you can grab
and actually work with to drive a product out of the market and a
conversation that really helps people understand and I think that first step
is very much in helping the business articulate the challenges or the
opportunities that are really trying to explore getting from there to the actual
launch or running an experiment is is the bit where I think we were getting
hopefully a little bit better but I find the challenge of helping take that
conversation from why would we want to do where what’s in it and you know why
why kind of why shake the states for how things are working and I think insurance
is now seeing the the concepts of things which are again pervasive in other
industries like design thinking and so on now coming in and creeping in and and
they’re getting quite excited about that so what do you think is making the
change and I mean it’s it partly you right or what are the drivers to unlock
people’s minds right well there’s a level of maturity that the industry is
going through so there’s you know there’s always that kind of
bootstrapping fear where there’s a lot of noise and a lot of things happening
nobody quite knows exactly what will stick what will not and there’s a poll
again we live in a world of uncertainty and there is that uncertainty of is this
technology really where we put our put our money into is this what is going to
be the next big thing and I don’t think the you need an answer for that what you
need to do is run the quite open-minded and run a whole bunch of experiments and
work with you know have skin in the game to really understand what does this
really mean in the context of applied outcomes and driving that I think is
where the industry is now maturing to a point where we’re raising the bar across
the market so we’re going to come back and talk hopefully a little bit more
about that in a moment so Orlando what do you see I think one
of the things that’s very clear we move into the insurance industry is the the
amount of technical legacy is way beyond I think other people’s expectations so
either I put a fever for two and a half years I was at money supermarket before
that and when I left money supermarket in 2016 we were in the process of
replacing our legacy database and the legacy database had been switched on in
2012 and I’ve joined a fever and we actually have operational systems that
are 50 years old 5-0 and I think when I talk to startups about that they think
I’m exaggerating when I talk to other people within the financial services
industry they they joined me mudding and maybe sobbing
because it’s really tougher and I think when I heard Reza talking I think it was
great that he talked about talent he talked about acquiring data you know he
talked about a/b testing and he talked about engineering I think on top of that
there’s a there’s a real challenge around tackling technical legacy head-on
because AI machine learning these things are powered by data and I think without
really investing in the data engineering challenges I think people will always be
left behind so um that tech legacy I’ve been in the insurance market since 2008
people were talking about that then I remember sitting opposite a CEO saying
I’m heartily sick of hearing about us Nene’s change our legacy systems he’s
not CEO anymore so so what’s gonna be what what practically can you do about
that given that this has been you know a board topic for the entire time I’ve
been in insurance and I’m sure for years before that Orlando any thoughts I think
in practice this involves typically building layers of data that are a bit
more contemporary over the top of existing policy admin systems I think
there’s going to have to come a time where people actually look to reinvent
some of the backend of insurance which still hasn’t really happened at any kind
of scale but I think you’ll have to at some point I think two electrodes point
I think GDP are kind of helps in a way because requiring you to have audit
trails of customer data requiring you to be able to delete customer data these
things fundamentally mean that you have to understand where the same customer
sits across multiple databases and so I think it kind of forces the issue so I
think it’s helpful okay thank you at this point any questions observations
from you because you’re part of this conversation – we’d like you to be so
yes yes I’m I don’t jump but if you want to come and get the microphone oh no no
no thank you for a very insightful talk and sharing your experiences given what
you said is it the technology companies who will make the more efficient and
effective insurers of the future yeah go for it I think in practice that the
places that consumers buy products and services for driven by consumers
themselves and I think insurance is a difficult market to break into it’s very
highly regulated I think it’s there’s a lot of kind of drags you know structural
drag in the system which means it’s not always a great industry to break into
but it could be a great industry if you’re an established player because
you’ve got the customers you’ve got the data you’ve got the products and given
that a lot of the data that certainly we hold is these are things that we know
about customers and we can learn about them that are very difficult for other
people to learn about so we kind of think that we’ve got an advantage there
a competitive advantage and it’s potentially quite a difficult thing to
break into I think that said some of the components of the insurance process are
very easy to be disrupted by a technology provider and I think
partnering with with technology companies is going to be the way that a
lot of people go okay and the other hand over here can you shout so getting better at getting talent is
that what you mean yeah you’re being kind aren’t you okay so she’s you’re
ready yeah yeah so it’s hard it isn’t the sexiest industry where people want
to come and work as data scientists and engineers
I think the insurance industry hasn’t articulated the you know the challenges
and the opportunities that the industry can bring to the more sort of you know
the academics or the the people who are really looking for a big challenge and
if I were to sit down and really talk about some of the things that I look at
now from sitting within the industry I would say that these are really complex
problems to solve and there isn’t a right answer but there’s a whole path of
exploration and discovery so we need to be more outspoken about how do we
present the opportunity for people to come and work in the industry and be
excited about it and and felt that they feel that they’re challenged and they’re
learning something and then rolling our products and services amok and so on so
it the onus is very much on us to try and talk more about that and I think
again going reflecting back on the conference as a whole of these three
days you hear a whole you have a whole diverse set of people who come in listen
to these talks and they take those reflections away but I think the
conversations I’ve had more one-to-one is being about how do I come and learn a
little bit more about the insurance challenges and what is it that you’re
trying to do in understanding autonomy or you know how do you look at
catastrophe risk or how do you model hurricanes and climate and how can we
participate and so there is a there’s an element of that people want to be
connected to some kind of the purpose driven side of things because insurance
does touch everything and and I think talent can come and really
a big difference the industry so working with academics to try and bring in
talent but also appealing to the market and other sectors where people move in
and I again I come up from a different industry and I’m really excited to be
here so in our preach at Orlando was saying
in that chair when he learned that you were attacked lawyer and the kind of
things that you were interested in he said we’ve got we’ve got loads of M&A
lawyers as a mature insurer but we don’t have enough people like you so what
would it take to get people like you into insurance I think it probably needs
a rebrand I mean I think I think it needs to get sexier because at the
moment it’s just it’s just not but if you bring in the tech element to it and
the impact that a I could have and you make that more known then that could
probably attract more talent well I think that leads us on to our next
question actually which we wanted to talk about data ethics and ethical
design and all of those driving better customer outcomes issue so Orlando I’m
gonna come back to you actually um when we when we had a chat you were very
eloquent about because I challenged you a little bit didn’t I off stage about
really Jill pricing loyalty penalties all of that good stuff
and does their dream of ever drive up really make a difference and you were
you had lots to say about what you and Aviva are doing to really drive better
customer outcomes with an ethical design framework in mind you want to share a
little bit about that the issue of pricing and the fact that we will know
we’ve historically charged loyal customers more than more than new
customers and I mean I think people who work within the insurance industry
understand what’s happened there I think people need price comparison sites
amongst other businesses have driven the cost of insurance down to the fact that
if you were to compete for your first year customer you want to attract a
customer you need to charge pretty much less than it costs
to actually have to run the policy and so recouping that money in later years
has become the the way that it works and I think our way through that is to use
data science and machine learning in AI actually to think about capitalizing on
our data asset how do we win customers for the long term how do we make sure
that we’re cross selling to customers how are actually doing things that are
about engaging with people and then we think we don’t have to fall into that
trap so we’ve launched new products that don’t have that feature we won’t be
charging you you more than existing customers but I think there’s a much
broader question around the ethical use of data there and I think for insurers
insurers have been pretty bad at writing down the principles by which they
operate and that’s probably not just true for insurance but it is
specifically true with with we’re such a data-driven industry and what the the
public debate about data ethics is actually forcing the in the industry to
do is is write down some of these principles and make sure that we can
hold ourselves to account and I think one of the nice things about algorithms
is that ultimately they are auditable and people talk about black boxes and
people talk about neural networks and the fact that these are difficult to
understand ultimately there are formulas there and they are auditable in a way
that’s far beyond the auditing capability of a human the ultimate black
box machine and I think you know historically we’ve made judgments that
have been very very ad hoc and without the ability to actually understand the
process by which those decisions have been made and our debate about data
ethics is forcing us to write down and stick to those those principles which
are and go far which are far deeper than just you know pricing decisions about
our specific product so obviously I’m not going to ask you details about the
insides of aviva but in general terms you know matching up that yes we need to
capture our principles and make sure that our principles stand up from a
compliance regulatory from a purpose-driven but also from a profit
orientated point of view just a couple of maybe questions not not what you’ve
been doing and the answers that you’ve had but some of the thought process
how you as an organizational thinking those complications if you want to put
it like that through will and are you happy to do that I think if you get to
the fundamental operation of an insurance business it’s kind of all
about getting the balance right between pooling risk and getting people to
contribute in terms of price in proportion to the risks that they add to
the pool and trying to understand that that trade-off is an ethical question
it’s actually trying to say how far do we want to go in terms of pooling risk
how where does it become wrong in some way how far do we want to go in terms of
getting people to pay in proportion to their price when do they become
uninsurable and so we have a data ethics forum which is sponsored by our group
exec team and we’re also partnering with Cambridge University because what we
don’t want is a situation where we’re just marking our own homework we’re just
deciding ourselves a receptacle and then we’re doing so I think it’s very
important firstly to have executive sponsorship and then secondly I think
it’s very important to get some kind of third party who’s typically and the
people that were working with the Cambridge are not machine learning
people they’re not AI people these are exes and people that are helping us with
the the ethical discussions around what we’re doing fundamentally great Ashish
coming to you then yeah so I completely I agree with that and I’ll add on to it
by saying that it’s not something we cannot think about ethics in a silo it
is something that needs to be out there in conversation and being discussed in
forums about what is the what is it that you really mean by ethics in the very
first place but also how do you how do you as a company be more responsible in
devoting the adoption and implementation of technology so we need to we need to
understand how technology works we need to be able to explain that and we need
to be quite transparent about how we’re using the data how is being ingested in
technology and what is what is actually part of the products and the services we
are reaching out to our customers with and it is there is a technology or a
demystification element to it but there is also a very important element where
we we need to educate and work together with universities and ethicists and
academics government in its adoption something
fell over but we’re still here so that’s okay Emma we were talking before about
the kinds of questions that your customers ask you about I’m gonna hand
this to you in a sec about that black box and the questions that they have
around that are you are you interested heartened it’s just another bit of the
conversation to get to a cell what’s your what’s your viewpoint as I tore it
yeah I mean we’re very interested we spend a lot of time thinking about
explain ability it’s kind of one of the things you really have to provide if
you’re providing someone with the prediction and you’re expecting them to
make a decision about what they’re going to do next
based on that prediction you have to explain how you arrived at that
prediction and so we spend a lot of time thinking about how we can build explain
ability into our models and we actually have a feature called key drivers which
basically breaks down how we arrived at a specific prediction so it might be
things like how tall is the building what is this building use for maybe it’s
a fireworks factory what’s located next to the building and
we provide these key drivers every time we provide a prediction so people can
actually understand where it’s coming from
and that’s yeah that’s something that’s really really important so lecturer
you’re sitting here listening to this are you how are you feeling about at
what response would you make I think we need more law actually which is probably
an unpopular viewpoint but I think we need a gdpr point too without all the
privacy policies into your inbox and I think we need a focus on how this is
going to work in practice I don’t think the GD P R was was done in that way and
I think that we now need to look at it slightly differently and from a more
pragmatic point of view so more law a couple of examples of that what that
actually looks like maybe or just one example for you I think we need to think
about how we create these products and how it’s great that Aviva has the
capacity and the resource to have these data ethics groups but what about this
text that don’t have that how do they know
what’s right and what’s wrong if they don’t have a think tank so it’s it’s
it’s more about guideline and there is no black and white way to say this is
how you need to do it because it’s so new and because we haven’t seen what it
can do but having a bit of guidance and an understanding of what the outcome
needs to be I think is the right approach
yeah and that’s that’s very much to your ecosystem point as well as initiation
what do you want chiming in on that so we we are part of these accelerator
programs where there’s a lot of startups that get selected as cohorts
up-and-coming and you know really good proposition so on and we go to them as
mentors to help them understand the world of insurance and you know be that
the subject matter experts if you like but quite often and item more often than
not we’re actually being reversed mentored by those companies in
understanding the use of that technology because they’re far more
forward-thinking and are actually iterating on those products and services
much quicker than we are able to understand them so I do actually like
the fact that when we have people from our company in the various business
units going out and mentoring and helping these companies understand what
does insurance mean in the context of that technology they’re they come back
with this you know this sparkle in their eye saying actually I learned a lot
about technology I didn’t know that this kind of technology existed so and then
the conversation becomes really interesting because they’re working
together to explore compliance regulation technology adoption and so on
and and that’s where I think is the sweet spot of that bi-directional flow
of wisdom and education okay coming back to you then and questions observations
yes do you mind she’ll get to you but if you
start shouting I think probably companies there is
increasing prediction accuracy what have you got to say from an insurance
perspective about non insurance of some members of society as that prediction
cost Falls and the accuracy increases yeah those who didn’t hear the question
the question was around as the cost of prediction declines and the accuracy of
prediction increases what we have to say about uninsurable people people that may
be uninsurable because we can predict to such a high degree of accuracy that they
are going to undergo some kind of claim event that we don’t want to insure them
and it’s a it’s a big issue and I think typically we see that it’s less of an
issue within general insurance so within car insurance home insurance things
where there’s a lot more intrinsic risk that we think we’re probably never going
to predict that actually it’s it’s it’s less of an issue I think where it is a
big issue is is life insurance and in particular the use of genomic data for
for life insurance and as an industry and you know our response has been
essentially we’ve all agreed not to use genomic data on the whole for pricing
there are some exceptions but on the whole we don’t use genomic data to to
price people it’s a voluntary thing there right Orlando
yeah and there are a few exceptions but on the on the whole I think that
probably isn’t going to be sustainable as an approach I think we need to
probably have an industry-wide approach and we need to be and there’s a big role
for a regulator there I don’t think individual insurers can make it call on
their I think it’s an extremely important very very serious issue that I
think requires industry-wide collaboration and regulator Kurupt
collaboration there’s an analog in the flood in flood examples there’s
something called fluttery essentially some homes are at such a high risk of
flooding that they were at risk of becoming uninsurable and the industry
response was essentially to join forces on that so I think
similar approaches for more predictable events in the future are likely to be
the way to go okay can I can I quickly add to that yeah
sure sorry so three things I think one one point that Orlando mentioned is how
far is going too far to really understand the level of risk
and what is what do you really need to use the second point is understand there
is this area of parametric insurance where you’re actually using only a set
of parameters and anything that goes above or below a threshold you get paid
out and that kind of insurance is because of the increase in accuracy of
the data and the ability to predict the power you’re able to create this much
more frictionless products that are able to detect something which you make
insurance much easier okay so there are some wins on that predictability and
using it more because of the lower costs okay
sorry quickly your question in relation to climate change any
thorough reflections on the insurers role in data because you mentioned the
models at the moment aren’t very good we know from climate scientists that these
extreme events are only gonna get sort of more extreme and more frequent and
that’s really already challenging the insurance sector and so any reflections
on the insurers role in actually collecting that data analyzing that data
and using AI would be appreciated good very briefly touch on that and then I’m
more than happy to discuss in more detail offline so one program we’ve
looked at for example and chip Cunliffe spoke at the climate stage day before
yesterday which is that we funded about 25 years of deep ocean and coral reef
research and all of that data obviously resulted in a lot of academic research
publications etc and we’re now trying to put a program together where we can try
and look at the broader benefits of the data that was collected and the research
that’s been done again in alignment with the SDG goals but also other programs
there where we can collaborate together with governments and institutions I
think it is something which is becoming more well we’ve also created something
we partner with Google to use that research to create education programs
for schoolchildren and it’s reached out to a five and a half million children
where we’re taking education about climate about ocean about coral reefs
straight into the classroom and I think that’s another way of using the data to
create an education program which i think is far it’s incredibly compelling
so it isn’t all always about the AI but certainly we can use technology as a
mean of it means of Education and and imparting back to Liz that there’s a
massive and long-standing set of data’s and analytical approaches a natural
catastrophe certainly in insurance and reinsurance so is your question and
you’re gonna let let them let you use is not the question as well right so it’s collaboration with yeah
yep and more and a more structured slightly
more formal hello to you issues number of PhDs and a lot of academic research
programs on a continuous basis so I know access Excel and Aviva are busy doing
things but you’re big you’re massive actually aren’t you is individual
companies so I suppose maybe it’s it’s again it’s about insurance as a whole
sector somehow as it did with fluttery perhaps just an idea to think about
these big problems that clever people we know we’d want to solve if only they
knew us better Ashish there’s something there to think about isn’t there okay so
we’re coming into the last couple of minutes or so so really wanted to just
get your take actually as individuals so she should you remember you said to me
the important thing and you mentioned it earlier is just start get started I’m
going to come to you first lecturer if Ashish is saying to insurance we should
just get started with this from your perspective as a tech lawyer where do
you think they should just get started what does that look like one idea that’s
a really difficult question I think consider consider the outcome always
talk about outcome consider the outcome what how do you want this to look like
and then start from there if that makes sense start from the end yeah I would
say focus on identifying areas where you can really like liberate humans from
doing repetitive tasks that a machine could be doing so that they can kind of
focus on applying creative thinking to insights to really solve hard problems
so yeah that’s where I would start Orlando yeah I agree with just getting
started I think picking a product that is going to make the most of AI that’s
going to deliver customer value I think is really important and then getting on
and delivering it and I think the you know where we’ve had some success at
Aviva is actually we try to deliver these things at scale but we don’t pick
the things that are that big so we try to pick indeed
visual products individual propositions and then launch them a scale because
that allows you to break down any of the barriers along the way and you learn a
huge amount yeah can I just ask at that point how much data is enough data in
your view Orlando because obviously specialist insurers in insurance don’t
necessarily have vast amounts of data so how big does your big data need to be do
you think I mean Aviva’s obviously mammalian but I think it’s totally down
to the truck the problem that you’re trying to solve I think all insurers
will have different types of data we’ve got 16 million customers in the UK we’ve
got 33 million globally and we might you know if some of these customers will
have several thousand data points and but I think it’s it’s less about the
size of the data it’s more about what you do and it’s kind of trying to pick
the trends trying to pick the insight it’s trying to to separate the signal
from the noise and hone in on the problem you’re trying to solve and use
just about enough data to solve that problem okay so we now have a helicopter
to add to the noise so let’s finish off then
well I asked all our panelists to do whether you turning me up or am i
shouting and was to to think about you know what on panels people really do
want to give their best to you so they think about what they’re gonna say and I
we try and have a proper conversation which I think we have had today hope you
agree but I really I’m really interested I’ve asked these guys to do something
slightly different as well and a little bit difficult is to pick one thing that
someone else has said or a question that’s been asked that’s made them think
oh I really need to look at that or think differently about that so Emma as
marketer you’re used to listening to other people’s thinking and thinking oh
yeah I need to make a note of that has anyone said anything was smart to
have any thoughts in you that’s made you think slightly differently as to when
you came and sat down here I guess not necessarily like made me think
differently but something that I definitely kind of it really resonated
was when Orlando kind of mentioned their insurance isn’t very trusted as an
industry and I think there’s a lot of research
that actually says and people trust banks and retailers and healthcare
providers more than they trust insurers and I think that sparks a really
interesting conversation around how can insurers kind of rebuild this trust with
consumers and how can technology like AI really enable them to improve the
customer journey to build that trust yeah it’s something that Emma said
actually about and the word that she used which is explained ability I really
like that works it’s really simple and it’s exactly what we mean when we say an
audit trail but that’s right transparency comes with explain ability
and being able to relay it in laymen terms so people understand why you’re
giving them a price or why you’re saying that their risk is higher so I think
that’s a really interesting point I might pick a couple but one explained
ability I think the interesting thing is that it’s a very hot topic but within
the AI machine learning community it’s quite a controversial topic there are
some people that are very antiques plain ability because it it might reduce
predictive power whereas actually amongst consumers explained ability is
typically what people want so I think finding a way through that kind of
dilemma I think is really really important and setting out principles is
a way to do that I’m gonna pick the other thing around you know we need more
law and I think it’s kind of unusual for a very highly regulated industry to say
we need more law I think one of the great things that regulators have done
and it started to upskill themselves in terms of hiring more people with
technical specialisms and I think that’s extremely helpful I think until we can
get to real clarity of the law helping people and step into these kind of
debates with some kind of technical depth is really really important and I
think it’s the it’s a good route towards getting more law which I think
ultimately we need great Ashish what about you so again I think reflecting on
the the legal side we we work with a lot of startups and clearly there’s always
issues of can we share our data outside you know off Prem and what does that
look like how do we you know how careful do we need to be are we compliant with
DDP or not and I think there’s this a growing need for its kind of enterprise
technologies or technologies companies to focus on well what does that this
data sharing ecosystem look like and how do we create so privacy security by
design there’s a lot of conversations in the earlier panel around insuring
identity ensuring privacy all of those things are going to be harder and harder
as we move forward so we do need to rethink about you know if you talk about
industrial IOT IOT system generating data all of that coming into our
environment or sitting on the cloud we don’t really have answers when we are
moving into this world of multi cloud hybrid cloud connected cloud
environments so we do need to rethink what does that it really mean or maybe
it is gdpr 2.0 or or something else but yeah maybe think thank you very much
okay we’re out of time thank you very much I hope you enjoyed that and please
show your appreciation for the panelists you

Leave a Reply

Your email address will not be published. Required fields are marked *