Claims settlements are an insurer's biggest expense. So it would seem natural for firms to understand as much about them as possible. Analysing claims in all sorts of different ways is surely an all round good thing. Yet is it? Some of the uses to which analytics are being put in claims functions fail the most basic of ethical tests, and risk confirming the public's worst fears about the sector.
There are different types of analytics being used in insurance claims at the moment. Descriptive analytics are being used to work out what happened. An example of this is image data to establish the extent of damage to a motor vehicle. Then there's diagnostic analytics, to work out why something happened. Telematics data can tell you why two cars crashed. These have obvious benefits for both insurer and policyholder.
Predictive analytics is different. It looks beyond the past and present, seeking to work out what might happen in the future. Yet aren't claims events in the past? Well, yes and no. The claim is in the past but the claimant is in the present and future. So the questions that claims functions are asking of their predictive analytics are along the lines of…
- How likely is this claimant to claim again
- Could we have incomplete underwriting data on this claimant?
- Might this claimant accept an immediate but reduced settlement offer?
- How likely is this claimant to become fraudulent
Real Questions Now
And we're seeing evidence of how insurers are using what their predictive analytics is telling them. Around 'likelihood to claim again’, we're seeing insurers increase motor renewal premiums even though the policyholder's non-fault claim was recovered in full.
With the likelihood of incomplete underwriting data, we're seeing the rise of 'underwriting at claim stage’. And with ‘immediate but reduced settlement’, we're seeing the emergence of claims optimisation.
And that question about 'becoming fraudulent’ has many layers, several of which have a clear ethical dimension. I explored them in this post back in 2017.
What is clear then is that predictive analytics can be put to all parts of uses, some important, others questionable. So this means insurers have to think carefully about how they use it. And by ‘think’, I mean something more than ‘can we do this’. Question like 'should we do this’ have to be just as much to the fore.
One Question to Rule Them All
One way to address that 'should we do this’ question is through a very simple ethical test. It involves asking yourself if you would be happy to see the decision on how to deploy predictive analytics in claims described on the front page of the Financial Times. Or perhaps a more populist paper like the Daily Mail. If you'd be happy to see your decision in the spotlight, then take it forward to its next stage. If not, then go back to the drawing board.
I suspect that several uses of predictive analytics would fail this test. So will they be taken back to the drawing board? That will of course depend on a special category of decision maker, namely those identified on SMCR responsibility maps as the Senior Management Function holder for claims. They will need to be able to show evidence that their predictive analytics projects have addressed questions such as ‘are the outcomes being generated fair for customers? And ‘is it being deployed in the best interests of each client?’
These are ‘here and now’ questions, given recent regulatory outputs (this for example). They are certainly ‘here and now’ issues for US insurers. Earlier this month, the National Association of Insurance Commissioners’ big data working party designated ‘predictive analytics in claims’ as one of its two new focus areas.
On the Radar
What about the FCA? On data and analytics issues in the past, the UK regulator has tended to lag US regulators by about a couple of years. That lag may not be so great now, and might not even exist. The pricing review will be consuming much of their data and analytics expertise, but I believe that predictive analytics in claims is at least on their radar, if only in a queue.
Meanwhile, the public's experience of the outcomes that predictive analytics in claims is producing for them will continue. And their assessment will I believe be a harsh one. For predictive analytics in claims feeds directly into the accusation often laid before insurers, that they are always looking for ways to reduce or avoid claims.
So how might the sector respond? Their response will undoubtedly be centred on insurance fraud. Yet the narrative around the investigation of claims fraud needs to be very carefully thought through. Of course fighting fraud is hugely important and ethically justifiable, but at the same time, so is the way in which fraud is fought. Insurers emphasise the former. A new narrative needs to acknowledge the latter.
Three Steps Awaited
Consider these three steps that I’ve suggested in earlier posts. Firstly, calls from their trade body to 'do whatever it takes’ to identify and tackle fraudsters must never be repeated. Such carte blanche narrative does more harm than the sector realises.
Secondly, the closed room in which the sector's overall fraud strategy is directed must be opened up to an independent voice or two. Governance at the Insurance Fraud Bureau exposes the whole drive to combat fraud to accusations of secretive group think.
And thirdly, performance management of fraud programmes within individual insurers must be ethically audited. I've been told of performance metrics in use at big insurers that beggar belief.
Insurers contest such questions with protests that fraud is fraud and that only they understand its technical nature. Yet the credibility of such a response was undermined last year, when Liberty Mutual were fined for, amongst other things, using predictive analytics to identify claims fraud without any real thought as to the training, testing, monitoring or oversight of that analytics. As a result, huge numbers of claimants were mislabelled as fraudulent. Only their complaints caused the insurer to start checking how the levers of their analytics had been set.
What this tells us is that oversight of predictive analytics in claims needs to be robust, informed and independently minded. I believe insurers have a couple of years to make sure that their use of predictive analytics is in order. We will see issues come out of the NAIC's work, and the detail of these shared in closed conversations with the UK regulator. This will then inform output from the FCA as to how they expect insurers to be managing their predictive analytics. This could be diagnostic work or risk based conversations with people on responsibility maps.
The danger for the senior management function holders with whom those conversations are held comes in the form of rationalisations. Replies such as 'the firm needed this from me’, or 'no one is really worse off’, or those five most dangerous words in business 'everyone else is doing it’, are easy to fall back on but disastrous for careers. Claims directors need to be well prepared, not just with better answers, but with more nuanced understandings of the questions.
One forum for this might be the recently formed Society of Claims Professionals. Set up by the Chartered Insurance Institute, it is now home to the tens of thousands of its members working in claims. The committee overseeing it is full of the very people with the authority to steer insurance claims in the right direction. Yet is there the appetite within the Society to take on challenges like the uses to which predictive analytics appears to being put in claims? Time will of course tell and the Society judged (not least by the FCA) accordingly.
Finally, while writing this post, it struck me that I'd done some predictive analytics of claims myself in the past. Back in about 1987, the broking firm I worked for had a very safety conscious client who sent it all sorts of data, including on both near miss and actual accidents at work. No one in the broking firm did anything with the near miss data, so one dull day, I loaded it all into my PC (one of only two PCs in the entire firm) and started modelling it.
Patterns started to emerge and the subsequent report told the client to watch out for three combinations of events. They loved it. It confirmed what they already suspected and helped them deliver an evidence based response. It also showed them that their broker understood what mattered to them and was supporting them in that. The result were fewer accidents and a loyal client. I was predicting claims, in order to avoid them. Yet I was doing so in the best interests of this client.
My point is that predictive analytics in claims is not automatically a bad thing. What does matter is the direction that is given to it and the leadership around how that direction is followed. Both have an important ethical dimension that the sector needs to fully grasp and deliver.