Using AI to Estimate Claim Exposure

In-house legal teams are under increasing pressure when valuing claim exposure. Their company’s earnings guidance can be imperiled if an audit shows an unexpected jump in value in their claim reserves. That can put counsel on the defensive if they’re asked to justify where the numbers came from. Rough estimates don’t cut it anymore.

AI tools, which are already being used in legal functions to simplify creation of frequently used documents and speed up electronic searches, can help with claims valuation too, but only if they’re used properly.

As legal professionals, we need to understand how these tools complement rather than replace human intuition.

Although computers excel at functions humans find challenging, such as complex counting, information retrieval and rapid calculations, they’re still no match for human ingenuity and intuition. That means the best results come when we take our best human advice about a situation and let AI tools calculate the implications of that advice. 

This same formula can turn legal advice into higher-quality claim valuation.

Claims usually are made up of subcomponents on which legal counsel can speak from considerable experience: the likelihood that a jury would find against the client, or that the alleged mistake injured the claimant, or how wide the range of damages a claimant is likely to receive is, or what the likelihood of a defense eliminating those damages is. Experienced lawyers are good at these assessments. It’s why clients hire and pay them money, whether they are at an outside law firm or in a specialized role within a company.

AI is most valuable when a client requests a bottom-line risk value for the case. Without AI, it is ordinarily the lawyer’s job to handle this estimation, and it is a thankless task.  

Although counsel will usually have already done an exceptional job evaluating the possible answers to each legal question in the case, estimating a final valuation goes further than this: it must account for the financial implications of the intersecting probabilities across each of these questions, which for complex claims is exceptionally difficult for human beings to do, much less to do so consistently or quickly.  

The discomfort of the task is not surprising: counsel never want to underestimate risk, which tends to incentivize conservative (higher) case valuations and, to the frustration of many clients, a disturbing frequency of 50-50 probabilities of success. The client, of course, can flip a coin themselves at no charge if this is always the bottom line. When in-house counsel are grilled by the financial team about these assessments, the combination of high valuations and uncertain odds makes for an uncomfortable meeting.

The problem is that the client is asking counsel to go outside their core expertise – legal advice – and effectively advise on the mathematics of combinations. Remember that this is what a risk valuation is: the average of the good, the bad, and the ugly outcomes for any claim.  

Given the role AI can play, what the client should be able to expect from their counsel is an estimate of the probability of success at each step of the verdict form. Other counsel experienced in statistical programming can then incorporate those probabilities into an AI calculation system.  

Now, the computer identifies the various combinations of outcomes, based on counsel’s advice, and further handles the counting, averaging, and range of bottom-line valuations. Counsel’s assigned probabilities must be correctly inputted, but once that is done, the computer can simulate hundreds of thousands of hypothetical verdicts in minutes.

When AI is incorporated in this way, everybody is happier. Counsel can offer their best assessments on individual verdict questions without worrying about punishment for getting a complex math estimate wrong. The AI system relies exclusively on counsel’s advice, so it’s showing the client only the mathematical consequences of what their lawyer has already told them. This minimizes concerns about the system being a black box that is simply making up numbers. Frequently, the valuations that are returned by the AI system are around what counsel expects, which is a good thing. But sometimes the values are very different, occasionally by millions of dollars. This can mean that counsel’s inputs need to be rethought. But it can also mean that the client is about to be spared a potentially devastating mistake. AI, therefore, helps complete the puzzle of case valuation. 

It offers other benefits too. Once an AI script has been written for a case, it can be updated as facts change or to test the effect of a development that appears to be significant.  

Not all claims require AI, but companies would be smart to use it for their more serious claims, effectively managing their portfolio of risk. 

The next time the finance team asks where these values are coming from, in-house counsel can provide a calculated summary of minimum, maximum, and most typical outcomes for each claim. And because each claim, regardless of assigned counsel, goes through the same final AI step, all claims are valued consistently, providing confidence in the results.  

Of course, the design and programming of an AI system for any case is still legal advice; the computer code has to be a legally reasonable summary of counsel’s advice, not some random programmer’s take on what counsel seems to be saying. Fortunately, there are attorneys with this combination of legal and statistical expertise, ready to help clients and their preferred counsel make these final assessments.  

*This article was originally published on Legaldive.