Predicting What People Will Do, Without Knowing Much About Them

Gilad opened with a note: this was his first guest with a PhD in anthropological sciences from Stanford. Avi Tuschman is the founder and Chief Innovation Officer of Pinpoint Predictive, a Stanford StartX entrepreneur and an internationally recognized expert on heritable psychometric traits.

About Avi Tuschman

Avi Tuschman founded Pinpoint Predictive, which uses AI-powered psychometric data to help carriers and MGAs predict behaviors such as cancellation, loss and misrepresentation. He wrote a book on the science of political orientation, which he describes as part of the broader field of human personality.

From Anthropology to Insurance

Early in his career, Avi advised banks and, he said, about 18 current and former heads of state on people analytics. The question was what data best predicts what an individual will do in the future when you don’t have many directly related behaviors, and quantitative psychology kept coming up as powerful but hard to get, the rare earth metal of data. Around 2014 and 2015, academics who are now Pinpoint’s advisors, at Stanford and Columbia Business School, showed that with enough of the right behavioral footprint, AI can predict personality scores more accurately than a person’s own spouse, and at great scale. Pinpoint set out to build a data enrichment machine with explainable AI. After a couple of years of R&D, insurance turned out to be the best product-market fit: lots of new data, smart people and problems the technology solves with important efficiencies.

Avi sees Pinpoint as the next layer of enrichment after credit, telematics and property intelligence, adding the granular individuality of the insured and human behavior across the value chain, with the aim of improving lifetime value, removing risk and personalizing. Most of the work is in personal lines, where insurance helps people protect themselves and their families, and he stressed the responsibility to use psychological models carefully and privacy-safely.

Capitol Hill, Cambridge Analytica and Privacy

Gilad recorded the episode the week after the storming of the Capitol and asked about the book he’d read on the Cambridge Analytica story. Avi said he has followed the field closely and knew some of the original researchers, who decided to have nothing to do with that company. It is a cautionary tale about what not to do on privacy and applications, and Pinpoint does everything differently: no social media scraping, patent-pending privacy safeguards, formal agreements with clients, review by its privacy team, and data used only for approved purposes. He noted that the efficacy of psychometric data is documented in peer-reviewed papers, including in the Proceedings of the National Academy of Sciences, and that many people argue about it without doing their homework.

What the Model Is

The core is the five-factor model, which Avi said is behind essentially all peer-reviewed personality research, used from the Air Force to the National Institutes of Health. It quantifies stable individual psychological differences as continuous, normally distributed dimensions, dozens of them, that correlate with future behavior. His example: a facet related to impulse control.

He compared it with credit, one of the biggest efficiencies insurers have adopted. Credit doesn’t cause risky behavior directly; it picks up parts of conscientiousness that happen to be available at scale. Pinpoint fills in the rest of that gap, in an explainable and data-driven way.

How Carriers Use It

It starts with enriching consumer records. Pinpoint works with a carrier’s data science team, providing privacy-safe enrichment features so they can quickly validate behavioral models, such as the propensity to cancel within a time window or non-catastrophe loss cost per premium. Where there is no robust data science team, Pinpoint builds a complete predictive model, with professional services followed by an automated batch process or real-time API. Results come with explainability at the aggregate level, and it can comply with regulation at the individual level.

Avi shared results from a case study: it increased the identification of true cancellations by 52 percent over the baseline. In the riskiest group for non-cat losses, it improved prediction of dollars lost by 15 to 30 percent in some areas. A back-test, in which the client knows actual outcomes, can show value within a couple of weeks.

Underwriting, Marketing and Personalization

For admitted carriers, where filing a new variable with regulators is heavy, Pinpoint is easier to use at acquisition: filtering who gets a quote, deciding where to allocate marketing spend, personalizing communications, or deciding about direct mail. Knowing that someone is in the riskiest five or ten percent changes the risk profile of new business. On credit and zip code, Avi said carriers are asking about moving in a more behavioral direction because they believe it is the trend and more ethical, and he calls it “greener and cleaner” than putting people in boxes by geography. Its models supplement whatever the carrier uses today.

Fraud Before the Claim

Avi said Pinpoint sees a strong signal of fraud propensity before a claim is filed, at the point of acquisition, which is where it is most valuable: once a claim is filed there is already an expense, and carriers already have powerful models. Using only its own enrichment features, Pinpoint can say how much more likely a person is to misrepresent themselves than the average. In the explainable model he pulled up, traits associated with misrepresentation include higher imagination, lower cautiousness and higher depressive tendency, and he said the explanations usually feel intuitive.

StartX and Accelerators

Pinpoint has been around about five years and spent the early years on R&D, with the last year and a half focused on insurance and scaling up its next cohort of carriers and MGAs. It is part of Stanford’s StartX, which Avi said accepts about 5 percent of alumni-founded companies and cites a very high survival and acquisition rate (about 92 percent, in his words), and it also went through the InsurTech New York accelerator in the fall.

Asked what an accelerator is worth, he gave advice from the carrier’s point of view: if a carrier joins just to check a box or go on an innovation safari, it isn’t a good use of time. Carriers should ask whether the startups fill a need and whether they have budget, and he suggests participants carry a sunk cost for proofs of concept. For startups, he said, you have to be somewhat transactional in a scaling phase and bring a healthy set of qualifying questions.

Advice: The Book of Why

Avi recommended The Book of Why by Judea Pearl about causal inference, and why statistics ignored causation for so long, noting how hard it was to prove the causal link between smoking and cancer without these tools. Gilad has been following previous guests’ recommendations, including The Art of Thinking Clearly, and listened to the new Hunger Games prequel about young President Snow. Avi’s lockdown tip: get outside, get vitamin D, and carve out family time.

Key Takeaways

• Pinpoint adds psychometric enrichment, based on the five-factor model, to carrier and MGA data in order to predict cancellation, loss and misrepresentation.
• In a case study, it identified 52 percent more true cancellations than the baseline; for riskiest non-cat groups, it improved dollar-loss prediction by up to about 30 percent.
• Back-tests can validate results within weeks, and enrichment is mostly applied at acquisition, marketing and retention rather than filed rating.
• The strongest fraud signal comes before the claim, at the point of acquisition.
• Avi stresses privacy safeguards: no social media scraping, formal agreements and approved uses.
• Carriers should join accelerators with a need and a budget, not for an innovation safari.