“Louise, I have a problem.”

If you have heard of WeGroup, chances are that you have heard of Louise, our virtual insurance assistant! She represents the front of WeGroup technology, which allows her to assist insurance providers with analysing consumer’s insurance need, setting up policy propositions, automating back-office work, but also providing help to work towards claims automation.

Automatic claims handling enables claims managers to spend more time on the difficult and intricate cases.

Recently, I got the chance to talk about the latter— claims automation — at the 8th Annual Insurance Claims Management Conference in Lisbon.

Personally, I am not a big believer of the concept of robots are taking over our jobs. I am into technology enabling us to be better at what we do, and that is exactly what WeGroup is destined to do. It enables a claims manager, for example, to be able to spend more time on the difficult and intricate cases, and as such attaining higher customer satisfaction, while the affairs that have become routine are automated.

We all know that computers are great at working with standardized and structured data like spreadsheets, or database tables. They are able to process data much faster than humans. However, we do not communicate in structured data. We communicate using words, which is a form of unstructured data. The problem here is that a simple computer program might be able to capture the words, but it won’t know what to do with them. It will understand the syntactics, but not the semantics. And this makes language understanding a rather difficult problem. Similar words might have different meanings and different words might have the same meaning. So, in order to understand the meaning behind a word, you must always explore it in its context.

This is exactly what Natural Language Processing (or NLP) does. It aims at exploring the given text or sentences and relating words to each other in their context. This way, the computer will have a better understanding of the structure of the text, its entities and its intents. Eventually, NLP can be seen as a tool that helps a computer understand our unstructured data, and transform it into structured data.

In conversation with a chatbot or virtual assistant, the computer should be able to understand the intent of the user and reply accordingly. However, this technology is not exclusive to chit-chat conversation.

Within WeGroup, we use NLP technology to enhance our property claims management. First, we ask ourselves the question, what information do we actually need to submit a property claim? Well, first of all, we need to know what type of risk occurred. Was it caused by leakage? A small fire or burglary perhaps? Secondly, we should know what objects were damaged or taken. Third of all, we should know when this happened with a specific time and date. Next, to that, it is also very interesting to be able to understand what caused the incident and in what area of the property it occurred.

Now the question is, as speech is brought in an unstructured form, how would a computer be able to extract these specific key information from their context? Using NLP technology, we were able to build a system that takes text as an input and gives back the found key information as a result. Coupled with our Louise, we can then dive deeper into the unanswered questions, to extrapolate the key information needed from the customer. Let’s give an example of how we teach a computer to extract this information.

One of the key features needed is the date of occurrence. Extracting time from a textual context is a relatively simple problem, as there are only so many ways one can tell the time. For example, you could type: “It happened Wednesday, at five past half two in the afternoon.” or “It happened the 23rd of January at around 1:35 pm.” or as it is text based, in a more digitally structured way.

However, other information such as extracting the type of risk occurred is a quite difficult problem. To solve this, we use a machine learning algorithm that predicts the type of risk based on certain word pairs extracted from the text. The algorithm relates the words “kitchenrobot” and “exploded” in a sentence, which implies a fire-related risk. If it were to extrapolate the word pair “bathroom” and “wet”, this would imply a water-related risk.

We can even take this one step further. Because we now have our claim in a structured format, we can check two big flags for approval. That is, first, if the claim is valid under its given policy, and secondly, if the claim is not deemed fraudulent.

In order to check if the claim is valid under its given policy, we use NLP technology to analyse the policy terms of a given policy to extract what is actually covered and what is not. In doing so, we are able to check the claim against the policy terms and quickly conclude whether or not the damage that occurred is deemed covered. For example, if the claim concerns water ingress and an open window and we check it against a given property policy, we can assess that the claim is deemed invalid as this occurrence is not covered.

Secondly, once we know that the claim is deemed as covered by its policy, we must check for fraudulent activity. As the claim is already given in a structured form, it is much more straightforward to check for inconsistencies. These inconsistencies are mostly based on factual information in relation to what could be deemed fraudulent. However, as enough use-cases are gathered, machine learning algorithms are able to find unseen anomalies, from which the system learns again.

At WeGroup, we do not only aim at building technology to enhance the current way of working, but we also work towards future proof solutions. I have shown you that current technologies enable us to bring another way of handling insurance claims. However, with this technology, we can also tackle the products of tomorrow. We truly believe that, in the future, many of the digital applications that we know today will be enabled with voice technology via home assistants such as a Google Home or Amazon Alexa. These devices allow us to capture speech in a new, fun and exciting way. However, that speech is then converted to text, which implies that we are able to attach our Louise enabled with NLP technology to these devices.

Let’s say that someone wants to file a claim with a Louise device. The person will say: “Hey Louise, I want to file a claim.”. Starting from that point on, a conversation will be started with Louise, who will answer from the Alexa: “Hey Bjorn! Hopefully, you are alright? What happened?”. The user will then explain what happened and from that, Louise will once again extract the key information. If some information is unknown to her, she will ask directly for that information and the user can reply to the given question accordingly. However, this goes beyond claims automation. Imagine having bought an electric bike and wondering if your insurance actually covers liability for it. One could ask Louise: “Hey Louise, am I covered for liability in regards to my electrical bike?”. Hopefully, she will respond that you do, but in the case you don’t, she could recommend you the optimal insurance policy that does cover this specific part and insure you on the spot.

So basically, insurance should not be seen as a bothersome activity, but should be integrated within our daily routines. This is why we truly believe that using the technology of today, we can enhance processes from the past, in order to build a better future and make insurance great again!