Revolutionizing Claims in Insurance - Interview with Niels Thoné, CEO at Sprout.ai

In this interview, Niels Thoné dives deep into how Sprout.ai has automated the claims experience with the use of contextual AI. Furthermore, he explores some of the current problems insurers are facing in claims, how we can solve them, and what the future holds for those willing to adapt.

Could you tell us a little about Sprout.ai? What was the idea or moment that led you to found the company?

Sprout.ai was founded at Imperial College in 2018 in the Imperial Enterprise Lab/incubator. At the time, Lemonade was receiving lots of press for being able to settle claims swiftly. We noticed a desire within insurance companies to be able to offer the same kind of service, which would also result in a major operational saving.

However, Lemonade is built on a new tech stack and is still a very small company compared to your average size insurance company. So, insurers couldn’t take the same approach to get to the desired result, as it would result in billions of dollars of investments and over 10 years to overhaul the entire infrastructure. So that would not be a viable route.

We envisioned a world where insurance customers get immediate and impeccable service at the point where they’re most vulnerable - i.e. when making a claim. As you only need to make an insurance claim when something ‘bad’ has happened to you. Currently, this process doesn’t work well for insurers as the average claims settlement time is 25 days.

So, to solve this problem and give customers the service they deserve, we decided to build a plug-and-play solution that would enable any insurance company to be able to settle their claims in less than 24 hours. Hence Sprout.ai was born!

Can you explain in more detail how your AI-based claims software works?

There are 5 modules to our solution:

1. Extraction (OCR & NLP)

We developed proprietary and state of the art extraction capabilities, which leverage OCR (optical character recognition) and NLP (natural language processing). We’re actually the world leader in handwritten OCR at the moment and are in the process of patenting our software.

The extraction happens on the myriad of unstructured documents and data underlying claims, such as PDFs, images, and the free form notes in the claims database. Without access to this ‘dark data’, currently unused by many insurers, you can’t inform the claim and therefore will experience sizeable time lags on the settlement. 

From the extraction phase we get what we call ‘data sprouts’, these are all important data points that allow us to enrich the claim with external data in subsequent stages and contextualize the claim as soon as it hits the claim department.

2. Enrichment (Data pipeline)

You can compare the data sprouts to coat hangers in the sense that they enable us to attach external data points to the claim in real time in order to provide a much better data capture from the start, verify the customer statement and validate the claim. 

We’ve gathered over 50 data categories from the external data ‘stratosphere’ in the past 18 months. Examples of data sources include weather, geo-location, medication information, business information, etc. and they’re gathered via API networks and by aggregating databases via our own web-crawlers.

The data-network is continuously growing by aggregating more databases and websites as well as having global contracts with data providers. As a result, we should hit +150 data categories by 2021, which is unparalleled.

3. Policy checking (NLP)

In order to know whether a claim is valid or not you always need two checks. One of them is checking whether the claim is covered. For this purpose and on a global client’s request, we’ve developed a proprietary NLP solution that can automatically check for coverage at the moment that the claim is made.

 The power of our policy checking solution lies in the strength of the NLP algorithms. For example, it can understand policies like an experienced claim handler would, including all the different synonyms for the same word - e.g. septic tank, septic system, sewage disposal, etc. It also understands legal vocabulary since we’ve connected it to various legal and ombudsman databases and can even predict the likelihood of a claims dispute occurring - a very powerful asset for easy and instant dispute resolution. 

4. Prediction (DL/ML)

This is where a lot of the magic happens. At this stage, we combine the historical claims data with the enrichments we’ve brought in. The combination of data then gets fed into our deep learning AI algorithms, which then predict the next best step for the claim and pair it with a clear justification (hence the name Contextual AI). The combination of internal and external data is proving to be a winner - with +30% performance increase on the predictive capabilities of the AI as a result. This is a big part of our ‘secret sauce’. 

The results you’ll experience as a claim handler are that a big bulk of your incoming claims on a daily basis will be re-routed to a fast-track since they’ve already been completely validated by our platform. This means that now the handler can focus on where you’re needed most, the complex claims that require a human touch, and high level of expertise. 

5. Continuous guidance & improvement

Sprout.ai’s Contextual AI guides claim handlers through every step of the way with recommendations and justifications. When a new piece of information comes in for the claim, the system automatically recalculates and updates the recommendation per claim.

When claims are settled, the system gets that feedback as well and hence continuously improves its performance according to the volume of claims that come in.

The current pandemic has forced many insurers to review their strategy, particularly in regard to digitisation of the business. What are some of your expectations and/or hopes for the industry post-Covid-19?

Insurers are accelerating their database migration to the cloud which will enable them to innovate more.

Claims specifically related to Covid-19 have and will continue to increase insurers' claims volume, ultimately reducing settlement time and therefore customer experience. At Sprout.ai, we are looking into developing a specific Covid-19 solution for insurers in order to settle these claims both efficiently and accurately considering the lack of experience claims teams may have in pandemic coverage. We will use our proprietary Contextual AI capabilities to enrich claims with specific Covid-19 related information in real-time. 

Some insurers were caught out when they had to work with distributed teams. So, for us it’s obvious that having an intelligent claims automation platform will enable them to make that transition more efficiently and allow claims handlers to continue to do a terrific job regardless of their workspace. 

What is the problem that Sprout.ai set out to solve? Why and how did your organization go from BlockClaim to Sprout.ai, as well as from deploying activities in the insurance sector and motor insurance, to working on health claims, and now moving into other lines of businesses? 

Problem: Claims settlements in insurance are long and tedious for customers and insurers alike. Customers are unhappy with the service at claim, when they’re most vulnerable and need help. Insurers have a lot of manual touchpoints and high operational costs + less than ideal reputations.

The problem lies in the limited data capture when a claim is made, which is called First notice of loss. This initial problem pervades the whole workflow and slows everything down. We solve this problem by bringing in external data sources in real-time to enrich the limited claims data and validate automatically if the claim is safe from the get-go - which allows them to settle immediately. Our combination of data enrichment and predictive analytics allows insurers to settle claims in record time, save money and provide unprecedented customer service.

BlockClaim to Sprout.ai: Originally, we had a hybrid solution of a private blockchain for data gathering across legacy IT systems (which are everywhere in insurance) and AI for predictive analytics. Throughout our journey and by working with global tier 1 carriers we quickly realised that procurement is very rigorous and time-consuming and they did not embrace a new technology such as blockchain - so implementation cycles become very lengthy. Therefore, we developed alternative solutions to pair with legacy systems as a plug in and focused solely on our world-class deep learning predictive analytics. Hence the name change.

Lines of business: We originally started in P&C (motor, property) but have always been opportunity-driven. So, when the chance came to work with a very reputable health insurer on their claims challenges, we leaped at the opportunity - knowing that we could provide a transformational solution with our data enrichment. The results were extremely positive and our health claims solution is now one of the highlights in the product suite. For now, we will continue to focus on Health and P&C insurance. We get a lot of requests from other types of insurance but for now, are holding them off. We are keen to stay focused and do what we do best.

How is Sprout.ai changing claims processes, benefiting the work of companies and assessment experts, and improving outcomes for people — compared to the current approach?

Our mission is to enable any insurance company to settle claims in less than 24 hours (instead of the 25-day average!). 

Improving outcomes for customers: less hassle, faster and less biased service, more accurate pay-outs. 

Improving outcomes for insurers: by leveraging data enrichment and proprietary deep learning algorithms we do the following:

1. Automatically check for coverage by reading the policy (via NLP) as soon as the claim comes in.

2. Create a fraud referral stream at the start of the claim, informed by the relevant external data (deep learning on macro and microdata). This allows insurers to catch more fraud, save money, and frees up precious claim handler time to deal with the actual valid claims and customers. Because it’s always the honest customers that suffer from the actions of fraudsters.

3. Cognitive claims recommendations. Using deep learning we can predict that the claim is a valid one before even all the information is available (a first in the industry) - therefore speeding up payments to customers without making a trade-off in accuracy. 

In a nutshell: using Sprout.ai allows insurers to fast-track the majority of claims, immediately spot the complex claims that require extra assistance, and double down on fraudulent claims. 

When it comes to significant innovation within the insurance sector, such as that offered by Sprout.ai – do you find that it is the case that the innovation comes first and the demand later to match it, or vice versa?

Very good question. I believe the movement goes both ways and there is an important dynamic. It’s pertinent that you listen to your customers and prospects. 

• They have one piece of the puzzle: they know their problems better than anyone. For example, lots of manual touchpoints, etc. They have a general awareness that there are / should be solutions out there but don’t know exactly which ones or to what extent those solutions can help or where the technology has grown to.
• We have the other piece of the puzzle: cutting edge technology and the knowledge that goes with it. 

So, when we speak it’s our job to listen to their problems. This can take up to 4 or 5 sessions to dig deep and figure out their issues from several angles. Once we have a good overview, we can then inform them what technology can do and what our solution specifically does, and how it can solve their issues. When this process is done right, it’s extremely gratifying because you can really see the synergy points emerging throughout your conversations. The ‘aha’ moments they have when they see how great technology can be in solving their real-world problems is brilliant. That’s partly what makes me love this job/industry so much. You can really make people and your users happy with solving issues that they’ve had for ages and that really made their life difficult (and costly). 

On top of that, I also believe that like any other industry, there are trends. Until about 2018, the big trend was fraud detection. The press writes about it, some promising startups get traction in the market, VC’s invest in the space etc. Fraud was sexy. But then people started seeing that claims, so often overlooked, was an even bigger pain point for the industry. Maximum 10% of claims are fraudulent. But once that’s done there’s still about +90% of regular claims that are managed very inefficiently, with bad customer experiences and high operational costs for insurers as the result. 

So I feel that for the past 2 years or so, since we hit the market, this awareness has grown massively and there’s an exciting wave right now in terms of changing the claims space and making everyone’s life better - both insurers and customers!

Do you think there is still a role for the human in the insurance sector – even in areas where great automation is possible?

Absolutely. And in more than one way. 

First of all, we believe strongly in explainable AI - that’s why we named our solution Contextual AI. It’s vital to provide context to your users (in this case claim handlers). Black box Ai decisions don’t work. Users might not trust them and you are talking about significant sums of money when paying out claims. 

AI is very good at doing specific tasks (often even better than humans). But it still requires a human in the loop for complex claims. For several reasons.

They have the wealth of soft skills and experience to manage a sensitive claim and the stakeholders (e.g. solicitor, customer, third party, adjuster, etc.) for more complex claims. Sometimes you need a human touch.

This also relates to the customer experience. When something awful has happened to you (death of a relative, big car crash, your home has gone up in flames, etc.), you’d ideally want a human to interact with you. To show compassion, have empathy, and help you in the best way possible. No one relates to humans better than humans. We are a social species and need that contact in dire times. So AI can play a major part in supporting claim handlers to do that job as best they can (e.g. settle claims quicker, find better solutions, etc.).

False positives & negatives. Even when you have a very high level of confidence in your AI decision, e.g. 97% (which is our operational benchmark), there’s still 3% of cases that you might not be sure about. In these cases, it’s important to have user feedback. A person can go into the decision, look at the supporting data (from our Contextual AI) and agree, disagree, or adjust the recommendation/decision. This serves everyone well:

• The customer gets the right service
• The AI takes this feedback loop as a learning opportunity and will not have to ask the same thing again in the future
• The user/claim handler has full transparency in the recommendations made by the AI and a collaboration ensues
• It’s extremely important to do this and we have a real-time user feedback loop installed with all of our customers.