About this event
3 Minute Webinar Highlights
Full Webinar Recording
View Event Material:
Webinar Statistics
Webinar ID: 641-010-012
Join us for this webinar and listen to an experienced, global panel of insurance professionals present, discuss and answer your questions on the theme of “AI Powered Insurer”. This is sure to be a popular webinar so register early to guarantee your place.
Our webinar will cover viewpoints from Thought Leaders on artificial intelligence in the insurance arena.
By participating live you can help to shape the panel questions and also participate in our live poll. Registered participants who are unable to attend will be emailed a link to the recording of the webinar.
Further Information
Webinar Transcript
Timestamp: 0 min to 1 min
Hi everyone, I am Hugh Terry from the digital insurer and it’s really a pleasure to welcome you to what I think is about our 27th KPMG sponsored webinar. And today we’ve got a great topic to discuss and for fantastic panelists lined up. So, it’s going to be all about the AI powered insurers and what we want to do today is explore.
What insurers are actually doing in terms of using AI to create value in their organizations and to engage with customers better? So, if we go to the next Slide the panel that we have lined up we have as our moderator today Elisha Deol. Who’s the global insurtech manager for KPMG? She’s based in London in the UK. With her we have Kaushik Raghunandan who is director digital.
Timestamp: 1 min to 2 min
innovation here in Singapore with KPMG. We have Gillian McCain who is co-founder of WorkGrid Software. She’s based in Belfast Northern Ireland. We have Gero Gunkel the group head of AI Zurich financial services and he’s calling in from Zurich today and last but not least we have Parmjeet Kaur who is the founder and head of research ITR connect, which is a startup business on research and analytics of insurtech companies. So great panel, I think
If you’ve been on before you’ll know the format, but Elisha’s going to frame this conversation for us and then take it away so over to you, Elisha.
Thanks, you yeah, so we’ve got a really interesting topic for discussion. As you mentioned are really relevant discussion topic as well. I think it’s safe to say every conversation we have with many of our clients nowadays always comes back to Either RPA AI or any other more than any other technology. So yes.
Timestamp: 2 min to 3 min
Mentally exciting conversation and to proceed with the format is going to be we’re going to ask our esteemed panelists to first of all explain some of their slides give us a bit of a background and really spark up the debate and then we’re going to ask a couple of questions. So, we’ll be taking those questions from the Q&A chat room. So, I urge everyone who’s listening to use that and we’ll definitely be framing the questions from there. And then after that we’ll have a snap poll where we have a look at what
everyone’s General consensus on the relevant topics. So, we’re going to give all of our speakers around 10 minutes to have their time and to go through and then hopefully if we’ve got time, we’ll have a quick panel discussion towards the end. So, it gives me great pleasure to hand over now to Kaushik my colleague over in Singapore without much further Ado I hand over to you.
Thank you, Elisha. So first of all, you good afternoon everybody here in Singapore and for people in other
Timestamp: 3 min to 4 min
Part of the world. What we’re going to talk about today is the AI enabled insurer the future and that’s something that I’ve been actively involved in for the for the past 12 or years in helping organizations mainly in Asia Embrace machine learning and AI in various use cases across the business life cycle of insurance and how it’s actually playing a role in in making decisions better and more evidence-based decision-making. That’s
Been brought to bear for various use cases that we worked on. So, if you look at the next slide, please if you look at the context of what exists right in terms of the key challenges that insurers face today, and we’re talking here about both life insurers as well as non-life insurance. The first critical element is actually the customer the customer is a millennial now the customer today wants Buying Insurance to be as easy as booking.
Timestamp: 4 min to 5 min
car, right? So, the customer here is not looking at having patience. The customer today is less patient than he was earlier. She was earlier and just the expectation that the customer has in. This digitally connected world is totally different from what it used to be earlier. You’ve got ten the entire Dynamic around Insurance penetration, right? A lot of us have heard about, you know mature markets having
Hundred percent penetration of insurance or over a hundred percent penetration of insurance. Whereas economies that are still emerging have very low penetration. But what we’re finding here for example in Singapore is that everybody has an insurance policy but they’re not covered. Well, right so they are underinsured so they have some cover but they’re not fully covered. So, this is room for improvement there. What we seeing also is
great advances in terms of Technology from a data
Timestamp: 5 min to 6 min
end point from cloud standpoint also regulate the environment has changed where The Regulators are more willing for data to be moved on to Cloud. We seeing the transformation from an IoT standpoint where variables on board devices in vehicles Etc riding a lot of data points that can be used to learn Either the driver better or to learn the health of the customer better. So that’s something that’s helping us. And when you look at from more mature Market standpoint, we’ve got
Aggregators of come in so you no longer go to the insurance website or the app to buy your insurance. You can actually go to an aggregator. The aggregator has to compare prices to compare Riders helps you compare benefits and then you can buy insurance on the aggregator. So, people are moving away from traditional Insurance websites. And also, they’re born on the web insurance companies. If you look at another New Concept that’s come which is called implied insurance. So, you buy an Air ticket for example you
Timestamp: 6 min to 7 min
Buy through a wallet you’re insured already. You don’t have to buy a separate insurance policy. If you want to cancel your flight ticket, let’s say ten minutes to go for the flight. You cancel it you get your entire money back. There’s no policy that’s coming to you. Nothing is being written for you, but that’s actually implied insurance. And that’s a big Trend that we are seeing in in Asia. So, given that there are so many challenges. How does AI play a role in in making the life of an insurance company better? So that’s what we look at from a use case standpoint.
So, convergence of data and AI changing function Dynamics and Bridging the Gap between the different value chain components is what’s helping insurance companies today. So, if you look at it from a value chain standpoint broadly, the three aspects one is the sales and marketing aspect. The second is AIs the underwriting aspect and the third is from a servicing and claim standpoint. Right? So, what we’re looking at is in
Timestamp: 7 min to 8 min
Is use cases that have been adopted by insurance companies in Asia that are looking at okay from salesman sales and marketing standpoint. How do I look at? What’s the what’s the real time next best action recommendation that I can give so in markets that our agency laid, for example, we’ve got insurance agents who are using Samsung or any other android-based tablet devices that are sitting in front of a prospect taking in information in real time saying, okay. This is the age this
Is the income this is the past health history, etc. Etc. And in real time, you’re getting a pure comparison saying, you know what people of your profile actually cover themselves for a million dollars. Let’s say. What you’re looking at is 750 K. So is there room for me to increase wallet share there is right. So that’s real time next best action recommendation. For example, that’s happening to enable insurance agents to sell better to sell the right product and also to service their customers.
Timestamp: 8 min to 9 min
Better. Right that something that we’ve seen being embraced by insurance companies here. What were also seeing is how do I use external data along with the internal data that I already have to build a more complete 360-degree view of the customer. So, we’ve seen people or insurance companies here, for example, embrace wearable technology and bringing in information from wearable devices in terms of number of steps with the person takes
on a daily basis and then averaging it out to look at. Okay, he’s the person monitoring and taking care of himself or herself or no. So that’s something that we’ve seen happening in Insurance. What we’ve also seen is how do we look at social media data and can we learn from social media data may not be at an individual personal level. But at least at an aggregate level can we learn from social media data and use it in building up the profile of a particular Prospect or a customer that we can use them to make?
Timestamp: 9 min to 10 min
Appropriate recommendations. The second aspect is in terms of underwriting. So, if you care underwriting typically it’s been treated as a as a middle office function but more and more. We’re seeing AI being embraced to actually look at. Okay, what’s the what’s the risk profile of this new policy holder or Prospect that’s coming in. Is it good risk? Is it bad risk? If it is good risk in I can I embrace this good respond my book versus if it’s bad risk.
And I can I not take it on my book at all. So that’s where there are set of use cases that we worked on from an underwriting standpoint where AIs helped lastly on the on the servicing and claim site. So, we’ve seen a lot of really good use cases fraud has been has been a use case that’s been worked on for significantly long, but we’re seeing more use cases coming in now. So, for example, there’s a unique case we work recently which was looking at computer.
Timestamp: 10 min to 11 min
So can I take photographs so all of us look at taking photographs when there’s a there’s a motor accident for example write this photographs are taken by both the parties in a motor accident and they just uploaded actually there’s no analytics being done or AI being used on these photographs that are put onto the cloud. So, we are actually helping a customer look at helping an insurer. Look at these photographs that are put up and can I read from them can
I understand. Okay, what’s the extent of the damage so that I am able to foe on both parties on what the likely damage is going to be and what the likely cost is going to be in terms of repairing the vehicles. And also, can I look at the service station can the service station make sure that they actually charge me what I think is visible damage versus looking at a fraud situation. So that’s again something that we’ve helped customers with. We’re also looking at the Triads between
Timestamp: 11 min to 12 min
Persistency sales and claims. Can we look at that? Try that and see okay for other who are the good customers that we have a who are the good channels that we have which channel makes more sense for us? What we focus on the entire aspect of servicing and claims is where we were again seen a lot of use cases being deployed using AI.
What we also have a kpmg’s is an AI strategy approach. The two approaches typically adopted one is the defensive approach as you can see the others an offensive approach the defensive seems to be old school in a sense. So, we’ve got organizations that look at okay. Can I gather all my data together? And then can I figure out what’s the question? I’m trying to trying to answer. So, you’re actually looking for an answer. You’re actually looking for a question when you
Have all the answers put together into a data lake or into an Enterprise data warehouse format.
Timestamp: 12 min to 13 min
This tends to be it laid from a defensive standpoint the regulatory requirements which make you do it and it’s actually good to get all the data in if you’re looking at compliance reporting rather than actually using AI and this is a legacy approach. Whereas if you look at an offensive strategy, and that’s something that we are actively taking to our insurance clients across the world. What we are looking at is okay. What are the key outcomes or benefits that you want to derive out of AI right and then you look at a reverse?
engineering where it’s business-laid you look identify outcomes that matter to you then look at answering those questions through the through data. So, you go back you look at okay, how do I look at structuring my data? I do I need structure data. Do I need unstructured data? Do I need it on the cloud where I needed on-premise, so that’s when all these decision points come in and from a from an evolution standpoint? You look at data that giving you insights lead to decisions.
Timestamp: 13 min to 14 min
Decision lead to actions and actions hopefully will lead to better outcomes. So that’s the entire targeted approach which is the offensive AI strategy approach that that we’re taking to clients and that’s something that’s resonating a lot in terms of the next slide if we can move on, please.
Broadly, what we built is a platform using KPMG and there are several capabilities or technology blocks that you can see. These are proprietary to KPMG. Of course, they leverage an open source. They also leverage on our own proprietary algorithms that we’ve written a whole host of areas, as you can see from reasoning to knowledge-based to analysis interaction and then visualization. So, the key thing that I recently mentioned to you a few minutes ago was the entire computer vision where you’re looking at
an image and trying to trying to see for a particular car model. What’s the extent of damage that’s visible? But yeah happy too.
Timestamp: 14 min to 15 min
Talk about these and answer any specific questions that you may have as we go along. We use the whole set of extensive technology platforms to address from an insurance standpoint. So, I’ll hand it back to Elisha. Yeah. Thank you. Thank you, Kaushik. That was really interesting. So, I think we’ve got one of the poll questions. So, I’ll ask that first and then I have a couple of other questions that that would really be great if you could answer so the poll question is which of these functional areas can
Benefit most from the application of AI now so if everyone can make their choice, that would be great. And in the meantime, I have a question Crossing and actually in the chat interestingly enough. There’s quite a few around this around the implications of AI but before we get into those, I have a quick question, which is what’s the best example of AI that you’ve seen yourself personally.
for me the use case that actually very close to my heart right now is
Timestamp: 15 min to 16 min
Is the agent use case so the agent if you think about it is not the not the most highly educated person, right? He’s not the most highly educated person out there. So, to enable an agent with AI write on his or her own tablet or iPad or whatever and you’re using AI to empower this agent to sell the right product first of all or the right Ryder and to sell it at the right price for the end insurer is
we covered or the End customer is covered in terms of the right value for his or her own peer group. I think that’s a fantastic sales use case that brings in, you know ethics into it. It brings in the right amount of risk protection for the End customer into it. It also brings in real-time next best action and scoring into it. So, I think that’s the use case that I that I really like.
Yeah, I think that that would resonate actually with quite a few people. And so, thank you for sharing that and one of the other questions.
Timestamp: 16 min to 17 min
Hard that’s actually and there’s a question in the chat for a bit later. But one of the other ones that really, I think would be interesting to hear your opinion on is what you think about the implications of Greater data being on the risk pool. So for that I think what we’re trying to as I think it’s Thomas who asks is what would happen to those who were deemed perhaps not the best to ensure like very briefly and obviously AI that is utilize more alongside other forms of data what will happen to those that are deemed I guess uninsurable.
So, I think you would still see them being insured you will still see them being priced rightly. So right now, I think it’s about people who take care of themselves at least from a from a life standpoint or a health standpoint are often penalized or at least that’s the impression that they have versus people who don’t take good care of themselves. So, it’ll even out eventually but I think it’s just about taking care of yourself ensuring that the claims that that
Timestamp: 17 min to 18 min
Have is a person or the life expectancy that you have as a person is addressed and more and more data will actually play an active role where the insurer if you look at it as not just an insurance company anymore. They’re actually going to be in a position to advise you on what on how you’ve got to lead your lead your life from a from a health standpoint, right? They could provide you in terms of for example running or walking and X number of steps based on your health profile. And that’s where the engagement between the insurer and the customer will increase
Yeah, and that’s perfect. I think that’s very interesting and there’s a couple of other questions as well in the chat, but I think if we’re ready if we’re ready to have the results back from the poll we can move some of the others to the Q&A. So, if we could get the results of his book, that would be great. Oh, wow. Okay. So, we see that one of the most popular areas for functional areas where we think as a as a group the way I can be most applied as in claim. So
Timestamp: 18 min to 19 min
I’m in claims, I guess Kaushik. Do you have any other specific use cases before we move on? So, I think that’s also a question that we have in our chat. So given the results are there any specific area claims use cases you can think of
Yeah, I think the big one today is about straight through processing a lot of insurers know that that’s a pain point because typically when you have a claim it’s because the customer is in distress, right Either somebody is being hospitalized or you’ve got an accident from a motor standpoint or actually there is the unfortunate incident of a death, right? So, in that sense, you’ve got to be able to
As soon as possible while ensuring that you’re not while ensuring that you’re not paying away fraudulent claims, so that’s where straight through processing which is a combination of not just AI but also integrating AI into transactional systems in real time. That’s where that that comes in. And yeah, that’s the most popular use case that
Timestamp: 19 min to 20 min
Seeing in this part of the world.
Yeah, I think that will I think again that really that helps answer some of the Q&A questions. Actually. There’s been quite a few in the last few minutes around claim. So, I think when you get a minute kaushik, I think I’m sure you’ll be busy filling in a couple of responses to that great. Well, thank you very much. I think that that was a really interesting point of discussion and something to that said us really up quite nicely now for me to hand over to Gillian. So, Gillian’s a co-founder and head of cloud engineering and AI at WorkGrid Software. So, Gillian I’ll hand over to you.
Hi good morning, everybody just to sort of go over what WorkGrid focus is and what it how is related to insurance and is that work with so far is a wholly owned company of the Liberty Mutual Insurance Group, which is a global insurer with 50,000 employees and 800 officers around the world. And what we focus on is building intelligent software solutions for our employees and for other companies and
Timestamp: 20 min to 21 min
Here’s so it’s really how we can take different aspects of AI am really focused on using those to improve the employee experience. So, if you move on to the next slide AI is a big topic and there’s many different fields and the previous slide showed all the different types of things you can do in your computer vision or any deep learning text summarization, etc. Etc. So what I’ve chosen to do today is maybe narrow it
A specific area of AI and maybe provide some practical advice and based on our learnings of implementing conversational AI across the Enterprises and when I say conversationally I this slide gives you kind of an insight into there’s many subfields within this natural language processing. And so, there’s a lot of things around just understanding unstructured data, understand understanding documents and
Timestamp: 21 min to 22 min
Techniques like context action question-answering document comprehension, but also things like sentiment analysis and importantly the thing that I’m really going to talk about is natural language understanding and entity recognition as well as speech recognition and text speech and why I’m highlighting those is I’m going to focus on some advice around how you could get started with building chat box and voice assistants within your Enterprise.
This is really a by defining and understanding these cases for this type of technology. I am we have focused on the employee but also the Liberty Mutual and we have similar technology being used in a customer’s face. So really understand your employee understand your cause
Timestamp: 22 min to 23 min
Mark where are the pain points and also understand how to involve those select from an employee perspective really did focus on design thinking design first and having sessions with employees to try and understand. How can we improve the experience helping them simpler? How can we provide more productivity gains? And we’re coming from the coach if there’s so many different systems within an Enterprise across corporate functions, which are
And procurement Asset Management. The list is endless so I can use technology like this to aggregate that functionality in one place. So that your employee can ask questions and perform tasks through a very simple easy interface and that’s why conversational AI and conversational interfaces are powerful is because it’s a natural interface that people know how to use people know how to speak. We don’t know why they ask questions and by using the
Timestamp: 23 min to 24 min
actual language understanding to create a natural interfere Centre systems, but also take a layered approach when we first started creating Enterprise chatbot several years ago, you know, there’s lots of different ideas of what you could do with it, but you don’t have to implement them all it wants work across the different areas and I’m like, maybe Implement several it help desk use cases or some common search questions or some
Common talent management sort of issues and that’s what we did with the Liberty Mutual and was kind of provide a chat interface that would allow employees to ask questions to perform tasks across multiple systems.
And we’ve next slides.
and when you’re thinking about building chat analytics or a voice analytics is really need to understand
Timestamp: 24 min to 25 min
And what is being asked and be able to analyse that so that you can track the different conversations that and the users are having but also use it to drive forward new features and you require and constantly use that to improve the quality of your box lid actually understands more and more because this really from a machine learning perspective is the training data that you need to understand more and more of what your users are asking.
Remove the next slides.
This is a key point across what I consider AI products is user experience matters and user experience actually changes how people talk to technology is different from how people interact fire messaging and chat which is different from how you design a website on a mobile app. So, understanding that this is a new interface.
Timestamp: 25 min to 26 min
Is important and also design and conversations. It may no longer be the web designer who’s able to build a conversational interface companies are bringing in scriptwriters people who are more and focused own words rather than images. And so that’s a different Dynamic really an engineering time is to have somebody there who was focusing on the conversation flows and designing a conversation.
And also, when we’re just thinking about user experience, I know chatbots had a big hype cycle. I think we’re not in a more pragmatic and understanding of the technology. So, is it the right interface as a pure conversational interface right or is it more multimodal you see how I Alexa has progressed with the you know, they Echo docks and the device into the echo show? It has more of a UI.
Timestamp: 26 min to 27 min
So, it’s and then with the Google systems as well, you have cards and voice. So, I think we’ll see this sort of Paradigm evolved and people will be choosing to provide Customer Service employee service across a wide range of channels and interfaces. Also, when you are, you know, there’s a lot of FAQ type box really what we want to move forward with is using a conversational interface to be able to then
they interact with systems that have natural language interfaces themselves. So, you think about searched and I searched is starting to evolve into more natural language queries and people want the answers. They don’t want a link to a document. They want to be able to extract answers out of documents. So, this is where you know, we talk about conversation way. I had some of those other topics and things like question and answering the document comprehension Isabella G to evolve.
Timestamp: 27 min to 28 min
The millions of documents and across an insurance system or Insurance Enterprise and to be able to use AI capabilities to extract content out of those a layer on top conversational interface where you can ask questions of your data, you can ask questions of your systems your policies or practices so that that’s an important area and also some key things and these are partial preferences. I don’t think you should ever pretend to be human if you have a chat for you.
Should be very clear that it’s a chat Bot and you do that by when people ask you design a conversation around these types of questions while you chat Bots. Are you human? We have found people more forgiving of a technology that seems what it is around pretending to be human because if you’re not able to do it, you just come across the very stupid person. So that’s just something and also immediately if you’re thinking of in the
Timestamp: 28 min to 29 min
customer. Some people do not want to talk to chat box and some people will talk to shout boss. Again. That’s understanding your persona. But if you highlight that in a conversational way that they can talk to human at any point some will choose to do that. But also, as you design your conversation to know you should be looking at sentiment and you know the flow of the conversation and potentially
Are there a lot of them to opt out to you know, go to human for more complicated conversation or if there’s a sense of frustration that they’re not actually getting what they want. So, a lot of that is around how you design the conversations going to selves by taking into account all of those things. So, like designing and developing a chat bot and so we don’t very quickly. There’s lots of tools but designing a group Chat bot takes a bit more effort on a lot more thought and I think that’s kind of an area
Timestamp: 29 min to 30 min
In general, we talked about AI and products AI is going to change how we interact with computers. So, we need to think about how that changes the customer and employee experience.
move on to next slide
This is one of my favourites are the Sciences AI is not magic and I think there’s a couple of aspects to sort of, you know, regardless of your role in an organization. And I think there is an education that’s needed of what AI is capable of when I say, I you know, there’s so many things. What’s the area that we’re really talking about and an education at the executive level at the business level and at the tech level I think we
Started, you know when I think back something in early conversations and we three years ago. We were talking right and I’ll pay you in chat box or will just learn if I keep saying this.
Timestamp: 30 min to 31 min
To it will learn no, it won’t we have to use that as training data. We have to rebuild the model. We have to rebuild the chat box. It’s not an automatic thing. I think future technology is going to move a lot faster, but I think this is really by understanding the capabilities of technology today the limitations and big pragmatic about where it can actually really apply value and I think just
Just goes back some very simple things of yes, you do need engineer to do need people to build things. This is like an example of a pizza box a very simple thing it’ll never be able to tell you the whether until you are called and integrated with The Well of system and that’s just seems like a simple thing to say, but you would be surprised people’s perception. I think a lot and media a lot of hype around it. But I do think my is a time. I see a lot of people being more real
Timestamp: 31 min to 32 min
About the capabilities. I’m very optimistic then about the future of what could be achieved while building and delivering really good quality systems not I that can really help with the experience and with employees and customers.
That’s the end of my slides great. Thank you very much for that Gillian. And I think it was really interesting for everybody to listen to at the breakdown of AI and actually that leads us quite nicely to the snap poll question. So, before we go into a couple of other questions and I just like to ask everyone which of these AI Technologies do you think has the most potential for insurance is it natural language processing is image recognition or is it predictive modeling? So, while we’re all heading through an answering that
that’s a question that we have around sort of the what type of skills. Do you think that we need it within us?
Timestamp: 32 min to 33 min
Stations to actually deploy AI as a service and I guess that’s for this mostly relevant for insurance companies, but I guess more widely if not specific to insurance.
I think so. It’s a wide range of skills. I think we have to look at first. We see in recent years the democratization of AI in a way with a lot of the cloud providers providing out of the box Black Box Services, which are very powerful in getting Engineers interested in the subject. They’re very powerful and getting out.
The door with something quickly and but what we see is that on some of those are very engaged and it also depends on your Cloud strategy as well. And you know, so I think Cloud engineering is a big part of this we have built the team that has like Cloud.
Timestamp: 33 min to 34 min
Engineering skills, but then we have to think about everything more about machine learning and data like data engineering and is there’s a concept as well and then we have data science. So, I think an AI team and AI capability is not just a data science team. It is engineering. It is Cloud. It is dead Alex on it is dear science. So, it’s a cross-functional team of that makes sense.
And I think depending on what you’re trying to achieve you can start with different points. Yeah. Yeah, that’s definitely something that and is asking if you think there are any difficulties in enabling insurance companies and to integrate chatbot cons.
Timestamp: 34 min to 35 min
Into them into their customer service application and I suppose more widely if not specific to insurance, but it’s really interesting to see what do you think would be the inhibitions that could prevent the true benefits of chatbots and things being an eagle.
I think to go back to per design as well as I you can put people off very quickly with technology if it is properly designed, I think.
Some of the other areas that maybe it’s not what we haven’t mentioned it yet, but probably get to later things like concerning privacy and you know, AI better and to make chat box better, you know, you do need to have access to what is being said, you know, I think that cancer people but the reality is, you know, we have access to that information previously.
Timestamp: 35 min to 36 min
It’s just this is a new interface like everything that is done with such a partisan Insurance face. It is to provide faster and better customer service. I think in most cases and it doesn’t have to be big things at once. I mean, we have Daphne saying, you know, when call centres are being used to just answer questions and a smaller more or less complex situations and then hand it over to the edge.
Because it is more complicated and I think that was pretty powerful.
Yeah. Yep. I think that I think that would really help that’s really helped us understand what some of the inhibitions are and I think if we ready then it will be great to go to the snap poll results. So, we yep. Oh, wow. Okay. So, I think we’re all pretty much agreed and what we think the AI Technologies are that have the most potential for insurance so that the general consensus?
Timestamp: 36 min to 37 min
It’s around predictive modeling. And if that is that do you think that resonates quite strongly to your own views as well and Gillian? Well, I chose natural language processing and I like a very interesting modeling I think is it’s a thing that’s already happening. Yeah. I think its death of being used, you know machine learning algorithms various things are being made and I think what will be interesting to see
and this was recently at a conference in the discussion on you know, we’re using these to make predictions, but are we making the most optimal decisions? How can we enhance for decorative modeling to be more like multiple models running at the same time as so it was very interesting for that area can evolve, I do think natural language processing and understanding content and if we can solve some?
Timestamp: 37 min to 38 min
Really hard problems around language. I think it’s going to change just how we use computers, which I think it’s more generic than Insurance. Well II can see in the next five to ten years that we will just be speaking to everything that day. You know, we all start up computer. It’s nearly a year, you know. Yeah, I feel over here, you know, and it’ll be exciting to say
Yeah speaking to everything but probably less speaking to each other. Who knows and well great. Thank you very much Gillian and its site gives me great pleasure not to hand over to Gero who is actually going to give us the opinion directly from an insurance. I think it’s going to be a really fascinating insight to listen to so I’ll hand over to you Gero. And yeah, we can’t wait to hear your views.
Timestamp: 38 min to 39 min
I cherish you hear me.
so, yeah, so yeah as mentioned I will just quickly want to give an overview of typical case studies that we do then do a small deep dive into one of our case studies and then just quickly speak about the lessons learned from our work so far and looking at how it work and that actually link
very nicely to the snip what we have just seen we focus on different areas, but one common denominator from most of our work is that we are focusing on natural language processing and
Timestamp: 39 min to 40 min
The reason for that is that the vast majority of the data and insurance are text they unstructured so you can think what claims that could be claims homes or medical report. If you think what policy happen that could be policies or contract if you think about underwriting that could be emailed or broken homes and all of this is in a text format. So, if we want to address most of the data that we have using AI and machine learning well then
We need to focus on the solution in our cases basically form into three different categories. The first one is mostly round compliant and mismanagement there. The whole idea Is to use AI solutions to provide to provide a second pair of eyes. So anywhere where we will need to check do people adhere to
Timestamp: 40 min to 41 min
Total or external guidelines do they stick to the rules to the stick to the governance? Well one way to check this is through an audit or compliance review or through training and one digital way of doing that is to actually set up an AR within that is monitoring the work or that actually provided suggestions to the user whenever they think there was there was a mistake and one area where we do this is for
and consider our contract reviews where we use technology to automatically compare and contrast contract and whenever we see a misalignment where there should be no misalignment, we flag it to a user and then the user can decide if he’s doing an adjustment there or not.
The second bucket is around gaining productivity gains. And I think we had earlier already run straight through processing.
Timestamp: 41 min to 42 min
Where were you AI algorithms and a lot of coinsurance areas like underwriting and claims?
To automate the intake and processing of documents again they could be claims forms, that could be any kind of other documents that are sent to us from customer or intermediary. The second point that I just want to highlight here is it’s not just about efficiency and productivity once you’ve actually started to digitize processes and to end what you will notice is that you will get completely the.
Insight about your processor and that you can actually get these insides in real-time or near real-time. And this is why we think that’s also a big advantage of having actually something like when we call now costing so you don’t need to wait for your report and three months to see how a certain
Timestamp: 42 min to 43 min
Quarter developed based on so this is KPI if all of the documents go to the you argue and then you can automatically spot new patterns and you can there be actually basically do a trend analysis and the overall process much earlier and if you look in the case study in a minute and the third bucket is very much around risk and side and assessment. So, this is very closely aligned to the underwriting world where
we’re not so much trying to extract new insights from our existing data, but we are more trying to say how can we use these algorithms to find new insights new patterns and data that we have may be overlooked so far or said we were maybe not able to find with more traditional statistical level and we did some work around leveraging external.
Timestamp: 43 min to 44 min
Peck’s data provider is compiled and now I will just quickly like to go to the next slide to do a small deep dive into productivity and Automation and there I just quickly wanted use the old process before we look into how we use AI to change it in commercial insurance for property.
It’s been a process where customers meet a customer wants to ensure 500 location. Well, then you will use a broker and you see there on the left side who will go to Zurich. They hey, here’s my customer and the other locations you want to ensure and here’s some third-party risk reports for this location. So that means they were people
Timestamp: 44 min to 45 min
Who went to the applications and the first third and we can use the third-party reports for all the tests? And what about commercial Underwriters will do is they will send all of these reports to risk in India and the task of the risk engineer to assess all of these reports and to come back to the underwriter with a risk assessment now to issue with that is as you can imagine if something I’m a writer comes with a few.
Report then this is an awful lot of work for the risk engineer and we are local control of a process. So, the broker reached out to us when you want.
so
when we get a request a lot of reports and there’s a lot of time pressure for the risk engineer to go to all of these documents to provide the risk rating and to get back to the underwriter so he can back to the so you can get back to the vocal. So, we said well wouldn’t it be nice if
Timestamp: 45 min to 46 min
we can do the report reading for and so if we jump to the new process.
Well the overall look the same but what you see there is that when before there was this email and telephone interaction between the on the right and the risk engineer. Now we’ve implemented the algorithm where the underwriter can simply drag and drop all the risk report. And the Machine is providing an automatic assessment for each of the reports for each of the risk factor, and then it provides a summary with tech extracts. So, the risk in
this can just quickly look through the highlights of all the reports.
Can maybe do a manual adjustment where you want to do it or provide some command? And then all of this can immediately go back to the underwriter. This is huge productivity enhancement in the variable are all of these, huh?
Timestamp: 46 min to 47 min
cases are handled that make up much faster and it allows the risk engineer to really focus more on the edge cases on the more complicated locations and all the standard locations. All these data reports can all be done by Vinci and this is especially beneficial for us because during renewal period we obviously get quite a few of those broker requests and the name.
Those are even in bit Peak Seasons to respond to the broker to respond to the customer quite past despite obtaining many of these requests and we have put this solution to life for several countries already and we continue to expand this.
across our lines of businesses and I think that
Timestamp: 47 min to 48 min
even if you even if you’re not Commercial Insurance, even if you’re not even a property line.
The key takeaway for me years. We are today able to teach they are going to read at 20 to 30-page risk report written by technical expert to understand that and to return a risk assessment automatically. So, think that important to keep in mind in terms of Technology maturity, right? So now let me quickly speak about Lessons Learned in Zurich which are on the next slide.
So, if you look at a couple of Lessons Learned we have directly I think we’re a bit of an early start and the overall industry. We started with our first work and May 2014 and adore more complex cases. I’m starting in 2015 and I think if you are more the beginning of your journey the first thing that I want to recommend you to do it don’t look for the Swiss Army.
Timestamp: 48 min to 49 min
Getting your kind of looking for this one tool that can do everything and you will have a lot of its vendors out there that promise you there’s one magic dissect Machina machine coming from heaven and solving everything and Truth to the matter is that doesn’t exist. There is not one tool that can do everything. You have to go through all of the heart and neighbours work to develop a technology stack of commercial AI Solutions and open source solutions that fit you use cases.
And that fit your problem. So yeah, that’s mention doesn’t look for the one two of everything. Secondly, if you start to work with multiple external vendors be careful to not give too much based on slides and presentations because you know paper doll looks very easy. And if you just choose your partner based on based on presentations, are you’re selecting India?
Timestamp: 49 min to 50 min
The most charismatic speaker but not the best technology solution for your problem. So how we did it in Zurich is we set it up as a
The randomized control trial a bit like in medicine we gave the same task to more to the vendor with the same budget and the same data at the same time with all subject matter experts. We let them do a prototype for a couple of weeks and then we just compare them contrast with the result and it doesn’t need to be expensive. We just spent a couple of $10,000 on this and this enabled us very quickly to distinguish between signal and noise between the providers who have a big mouth and the provides we actually had some really
good capability and then the outcome of the of this work, you know, it’s not just slides to make an actual first solution for your own data. And if you if you do a very narrow scope and we focus on some very complicated problem, you will very quickly see which to work or which to that will work. So again, I mean try to do
Timestamp: 50 min to 51 min
Maybe testing try to give the same problem to multiple providers and if you narrow if you have a very narrow scope, it doesn’t need to cost a lot and you will get quite an impact. As I said, there’s none that I think is first of all working in better team. So if you have a business team and an IT team working separate the question is not if you fail but when you will fail, so I think it’s very important that it one team with little experts and it guys sitting in the same room, but the only way
to make it happen not just from a technology perspective but also from a change management perspective and then in terms of the changes that you bring we also honour I mean if you do automation to automate at all, don’t make you no, don’t try to use a lot of oil that euphemisms for I’d you can be very upfront to say hey guys. We want the machine to do this and that in the future and I think what you will realize very quickly is that people actually happier machines take over part of their job simply because
Timestamp: 51 min to 52 min
These are the kind that people don’t enjoy themselves. So, pick the biggest Champions that we very often find for project other business guys in the front line who says I have too much to do if a machine can do the reading gear for me or can do this mundane boring task for me. I’m all up for it. I will support right. That’s all for me over to the next one. Perfect. Yeah. No. Thank you very much dear. I think you really helped bring to life a lot of
Of the content that kaushik and Gillian were discussing previously as well. And I think you’re sort of rhetoric around. They’re not being a one-size-fits-all and a sort of one ring to rule them. All Swiss army knife kind of fit is really is really going to be something that is a good piece of advice for most people. So, we have a couple of questions in the panel that I’ve sort of assimilated into one which I’ll touch on in a minute, but just before that if we can go through the poll and if we can ask everyone which of the following in your view is most
Timestamp: 52 min to 53 min
Important in a successful AI project selecting a solution provider management of the solution provider quality of people deployed internally to the project scoping and planning data management data availability. So, I think if we can go if we can ask everyone to answer that and in the meantime while we do that a question that we have is for you Gero is how does AI affect business to business insurance processes and products as opposed to perhaps the business to consumer?
some processes
I think actually the changes in terms of more automation will insert a very similar. I think just the changes that we see there are a bit more delayed simply because at least in the retail space a lot of the transactions are much more standardized much simpler. So, applying all of these algorithms. It’s much easier while in the business to business pays you
Timestamp: 53 min to 54 min
If it more complex transactions complex interaction and it’s much more open relationship business. So, I think we will all the changes we see the b2c space. We will see in the B2B space in the librarian. That will be my general observation. Yeah, I think that that sounds very sensible and very applicable. Especially given what you found told us just now, so I think the other question that we have that has come through it.
Is around sort of how insurers can structure their AI assets and maximize cross-functional synergies in terms of if they don’t have a one-size-fits-all solution. So, I think what we’re looking for here, I guess from your perspective Gero is any principles to help organizations structure their assets better?
Oh, big one. It’s a toughie. Is it healthy?
Timestamp: 54 min to 55 min
So, I think the first case the first step I think this will lead to Cluster use cases together that we very similar capabilities because very often I see people saying are you know that the claims cases this is underwriting case. So, we still very much Thinking like the typical Insurance functions, but if you actually break it down, especially like a natural language processing the processing tasks, so it’s more
About an algorithm that I’m offended test and with processing a text document to extract certain key parameter P values and if you structure this when you think okay, where could we all apply this? You very quickly will see that; you know claims you analyse the claims form. You need to extract key parameters claimant name payment policy number and so on if you go then on the underwriting side missed a broker’s commission people again see something very similar sized. I’ve been trying
Timestamp: 55 min to 56 min
The customer what the policy number what do they want? So I think if you think mon capabilities of what the tool needs to be able to do with you were very quickly find a lot of commonalities between use cases and if you’ve learned from one tool that works well for your cases, you can maybe structure your overall portfolio not too much Acclaim underwriting policy admin button on the capability of reading text and extracting the following five key parameters, and then the uncertainty of moving from one use case.
So, the next actually greatly reduced if it works one area. I hope that that makes sense. Yeah. No, I think that does I think that was that was a very good response to a very tough question. And also, I think just to let you know Joe, you know, their topic that you’ve explored and some of your insights have really cause a lot of people to ask a lot of questions. So, after your after this segment, I’m sure you’ll be very busy with a few more tough ones in there and I think if we’re ready now if we
Timestamp: 56 min to 57 min
Go to the poll results. Okay, so we have not a very huge Divergence between a couple there but I think as a as a group we think that the most important thing in a successful AI project is the quality of people deployed to a project and I think I mean my personal view is that resonates quite strongly with most of world with any transformation or change program. It’s comes down to the people but is that something that you also do or think that is very fundamental so you make reference to it in yours.
Slides, but you know is that type of result something you would expect to be a key function that you consider when you’re approaching any AI projects.
We and he and his whole thing because would be that easy that don’t or won’t pay and if you have it you are you and you’re done. We would have like 500 deployment across and through the insurance industry today. The bad thing is it it’s if and so you need all of that and if you miss one you have an issue and that’s why it’s so hard.
Timestamp: 57 min to 58 min
Yeah, absolutely. Well, thank you very much Gero that that was really insightful for everybody to understand more about how AIs being applied more broadly in organizations and it gives great pleasure for me to hand over to our final and by no means are least important speaker of the of the of the chat. So, I’m glad and love to hand over to you. Thank you, Elisha. Okay so slightly on the different angle. Let’s start by looking at what the
Press has been seeing about AI and I found this article by Marsh Ventures very interesting. So, they are seeing out of the 2830 start-ups in Europe that were classified as using AI actually only 55% of them were using AI effectively in their organizations. So, you know, this gets to the point that there’s a lot of hype in the marketplace around AI and start-ups like to use the fact that they are using
Timestamp: 58 min to 59 min
Because actually increases their funding.
And then if we look at the other one, I just want to pick from these Snippets. There’s a whole ethical issue around AI, and if we look at the Amazon, so Amazon recruitment tool favoured male candidates over female candidates based on historical data. So, this is getting to the point of machine learning, right? So, to issue here was the problem of the training data that was being trained by showing it pictures of current employer.
He’s who were mostly males so you can see we’re in early stages with AI and lots of ethical issues that will trickle through into the industry that we’re going to deal with and just moving down the other one. I just want to pick from this slide is from Optimus consultancy actually quite interesting here, you know, I clearly see this huge potential of AI for our industry and they are
Timestamp: 59 min to 60 min
Meeting that the investment industry will spend 2.8 billion on the eye by 2021 expecting to cut a quarter of a million jobs from our industry. So that begs the question of what does the new world digital in shooter look like in 2014?
Okay, we’re going to the next slide. I want to start shedding some.
Statistics from our database so ITR connect as a database of 1500 ensure attacks. And if we analyse them roughly 40 percent today are active in the AI space most of these are technology companies or their technology and near blur companies are producing solutions for the insurance industry. We’ve got quite a bit of penetration there with intermediaries go to market and business models what we tend to see
see there more tactical applications of AI.
Timestamp: 60 min to 61 min
Didn’t see they’re much different than what organizations have been doing over the last 10 years. Maybe it’s faster. It’s better in the application. If we break down AI, and we look at the component parts. You can see the strong Focus there on the middle chart on machine learning and deep learning. They really are the new developments are coming out of AI and creeping in natural language processing. So, I think is Gillian is saying there’s loads of potential there, but still the industry is ours.
Early stages we’re seeing some virtual assistants coming in as well. So, you know virtual assistants are many ways robotic process Automation and that’s enabled by AI.
So, it’s a very dominant force of change for our industry the first pie chart there if we break down the ensure text, no database by are five technology families and ITR connect. You can see AI analytics represents.
Timestamp: 61 min to 62 min
49% now what’s interesting about that is if you go to the other side and you look at business process, that’s 47%
but business process is mostly enabled by AI robotic process automation chat. Bots knowledge process automation all enabled by AI saw the true penetration of AI Within These ensure tax is a lot higher.
So, they’re very small companies. Right? So, Lot of these and ensure they are an early stage. So, we look at them by the year of startup. Look at that Peak coming in 2016. So, they’ve not been around that long. They’re small in size and they’re focused on driving efficiency for the insurance industry sort of these organizations really are they’re looking at the pain points that we have today and, in some areas, they can drive greater change and other areas. They’re still learning.
Timestamp: 62 min to 63 min
So, go to next slide. I just want to share three use cases from a database that I find particularly interesting. The first one is on claims. So, we got shift technology here started up in 2014 French-based and it’s all about dealing with claims fraud huge pain point for industry and a huge cost burden. So, they’ve built a platform that samples data from multiple sources.
Has and they basically use machine learning to identify fraudulent patterns and interesting watching the CEO and doing a devil of this and he was actually noticing it. One of the fraudulent patterns was actually a stream of people within the same family and living in different locations, and he was able to get this data by just bringing in all these different data patterns.
So, this is a great business application.
Timestamp: 63 min to 64 min
You know, we’re seeing lots of clears when they’ve got a good idea to come into the marketplace, but that’s not always mean you’ll be successful. So, what’s happening with this player named months ago. They had one partnership.
Six months ago, they started building that up three months ago. They’ve got that up to nine Partnerships in place. So, they really increasing that strategic alignment. The funding has just increased. So, they’ve just secured another 60 million of funding. So, I think here we’ve got good traction for shift. I think what you almost want to watch to see how well this player will do it’s what are the use cases? It’s now going to build with these new Partnerships because that’s what’s going to prove its success in the marketplace.
We going next one Health one of my favourites. I do like the application and the health Marketplace.
Timestamp: 64 min to 65 min
It’s focused on predicting Health course offered our schema level but also at a patient level so it does this by, you know in putting large amounts of historical patient data. It takes a medical insight. It takes in Market factors. It looks at disease costing models and by using all the training in the machine learning and the Deep learning it gives Frozen all those patterns to help it predict future patient course.
So, one of many players coming into the marketplace here in this hell for arena and they’re all touching and different aspects and broadening the AI ecosystems. We’ve got Babylon there’s health wearables coming in and then there are no they’re coming together right from Partnerships. So perfect application for AI where you’ve got large data sets available with a cost of error.
Timestamp: 65 min to 66 min
The health industry is really high if they get it wrong, and that’s part of a reason certain industry. It is such a long-term care is never actually taken off the market place and at a rate are prevalent in the current system. So, AI producing actionable insights can ride real value to health insurance and actually for us for us as a society as a whole.
Okay, stick to my last example and chatbots which we’ve been discussing earlier. So, I found this example quite interesting. We’ve got speak see your key examples set up 2016 hasn’t secured any funding to date. So, what its offering is a white labelling Chat bot service to insurers and we’re seeing a lot more of this in the marketplace. Some of them are quite basic applications. Some of them are getting more sophisticated and it’s almost a level of height.
To some degree.
Timestamp: 66 min to 67 min
virtual nurse and the framework allow insurers to easily plug and play this service anywhere in that value chain.
So, it’s got huge potential this right if we can get it, right? This is where we can remove those Manual High cost tasks using robotic process Automation and leverage these chat Bots alone resources to move up to higher level higher complexity tasks. And potentially this is where we’re going to see that bone rate of resources as well. And it allows also insurers to deal with a very important pinpoint. They’ve got today which is to become much
More customer full because they will have more time to become customer focus but it can also leverage these tools to deliver that engagement with the customers and it’s quite interesting. Like I said, we started seeing these examples very much more in a customer service are in the claims area.
Timestamp: 67 min to 68 min
But we’re seeing this actually possible and to know that seals advice chat Bots its poor seals chat Bots its engagement chat Bots. I think will be chatbots everywhere soon. But what will be that integrated chat for model what will be the use case that really fits the insurance industry that I think we still need to wait and see but overall it’s Ticking out cost and it’s improving our customer experience at the industry can deliver to the customers to date?
Okay. Yes. Thanks. Thanks, Parmjeet. I think that’s really interesting. And actually, I think it was great for us to finish on Direct examples like that. So, thank you for bringing that to life and it really shows. I think the breadth of application of AI across the globe. So, it takes us to our final poll question. So technological singularity is the hypothesis that the invention of artificial super intelligence will abruptly trigger runaway technological growth resulting in unfathomable change, too.
human civilization
Timestamp: 68 min to 69 min
When do you think this will happen the next 10 years within 30 to 50 years? You’re not too sure or you don’t think it will happen. We really interesting to see what everyone thinks in terms of how quickly and how vast our how broadly this will apply to us all and so Pam Jake just while everyone’s answering that snap poll question a quick question for use out of those use cases you shared with us. What do you think the most interesting and why my personal interest is and speaks to you, but I’d love to hear what you think and why?
Been chatting too much of a chatbot. So you sure can’t pick up one I would be going towards the Miata actually because from what I can see there is this artificial intelligence is a real game changer for the industry and where I see it happening soon as health and I think you know, you’ve got a problem there right that’s looking for a solution. You’ve got an industry burdened with high Healthcare course, they’re looking for Solutions and we’ve seen
Timestamp: 69 min to 70 min
With Babylon going in trying to help the NHS and a pilot to actually provide chatbot and you know queries to answer questions from patients, you know, you’ve got Mass data dictionary as you’ve got you’ve got data records. You’ve got lots of patterns you can train the eat you can train and teach the AI to do so, that means that you know, you’re able to program those chats bolts, right? So, you know, it’s almost a very good fit.
For that solution and then you’ve got the whole gamification wearables to building that customer engagement together and that affect delivers almost ecosystem plea for in the health insurance industry. Yeah. Yeah. I think that’s I think that’s equally a really great example and an already relevant topic actually and so perfect. I think we probably have the snap poll results now and available for everyone to see. Yep. There we go. So actually, I think the general consensus from
Timestamp: 70 min to 71 min
Here is that we think it will hit within the next 10 years and I guess from what you know what you said. I think that probably is something you would agree with if not already impacting us I suppose perm. D’ Yeah, I think what we’re seeing is, you know, I agree with the comment made earlier of got on predictive modeling. I think we’ve seen AI being present in an industry where it’s rule-based algorithms for a long time and we’re seeing that transition from the marketplace. I think, you know you can the AI.
II tools are good at past information, but they’re not good at being able to train them a future event. So those elements still need to push through but I think they will start to develop and will start seeing more tangible track to move in the tent next 10 years. Yeah, I would agree with that. Yeah, perfect. Well and thank you everybody who has thank you to all of our speakers who presented today and sadly we’ve run out of time for the QA. So, I’m not sure we’re going to have
Timestamp: 71 min to 72 min
In to answer in the Q&A panel any of the questions, but I know that a lot of our panel will be answering in the chat anyway, and as they have been so hopefully, we’ll answer any burning questions and further reading obviously can be found. We’ve included in the pack. I’m now going to hand it back over to you for some wrap-up and final announcements. Yeah. Thanks, Elisha and thank you everyone on the panel. How exciting is this topic?
Listening to all the opportunities that are out there and for me, this is what digital insurance is about. It’s about exploring how we can use technology to improve the insurance industry. We put that question at the end around the technological singularity as a bit of fun and I was quite surprised actually on that result. So, you know, if we’ve got a belief that that Singularity is going to happen in the next 10 years. That’s huge right? It’s absolutely
Timestamp: 72 min to 73 min
Safe and I was on holiday last week. I was watching with my family Terminator again, right? So, what is the ethical Dimension around this are we going to have to prepare for Skynet? So, I have to say I answered don’t know on this because I’m not quite sure where things are going. But I think there’s a whole separate discussion we could have around that which is probably best done in a virtual bar. Someone using VR headsets. I think in the future we will be doing that. So, I’m just going
To wrap up now for a couple of minutes just on some announcements TI related announcements my face 2019 so bigger and better in this year ahead. We’re currently in the process of opening up and are open up for awards that we’ve got two Awards this year the same as last year the startup insurtech Awards and also the insurer Innovation Awards, so go to the website you’ll see there’s a very easy application process very simple to apply.
Timestamp: 73 min to 74 min
We’ll get publicity and then of course as you know, this is a reward where essentially you decide the winners people who come on live during Livefest decide who wins those Awards and then we wrap around that about ten or twelve webinars, which are time zones around the world using the similar format you see today, so please do sign up for that free to join. You just need to be TDI member and then last but not least we go to the last slide just
Logistics if you’ve been on before, you know, we’re recording it will email that out of which participants will be on the site emails are there for everyone to be talking today? So if you would like to reach out to them directly you can do so and planning is already well under way for our next webinar the date for the diary is the 11th of June at the topic is digital Bank Assurance topic very close to my heart and I think we’re seeing a lot of change ready in terms of the way.
Timestamp: 74 min to 75 min
The existing Bank Assurance Partnerships, which would be based on what face-to-face based selling need to be really orientated towards this digital first world as well. So, it’s a very relevant topic particularly for people in Asia and some people in Europe as well. So, we’re just about ready to wrap up. Thank you, Elisha, for great job on your moderation and to all of our panelists for their time effort today in participating and to you as well for taking the time to listen and giving us all those great.
Questions as well. So, I hope you enjoyed it and to come along to the next one tomorrow.