What if you priced risk based on how people actually drive?
In 2011, Progressive Insurance was a mid-tier player in a commodity market – about 10x smaller than their size today. Auto insurance had been priced roughly the same way for decades: actuaries looked at your age, your zip code, your gender, your driving record, and they put you in a bucket. The bucket determined your premium. Nobody knew how you actually drove. Nobody had ever tried to find out.
Progressive asked a different question.
“What if we could price risk based on how people actually drive instead of demographic proxies?”
That question became the Snapshot program; and Snapshot became one of the most consequential competitive advantages in the history of the insurance industry. Progressive today is roughly ten times the size it was when that question was asked. The causality is not incidental.
The data no one thought to collect
The industry consensus, unexamined for generations, was that individual driving behavior was unobservable and therefore irrelevant to pricing. If you couldn’t measure it, you couldn’t price it; and if you couldn’t price it, you used the best proxies available. Young male in a sports car in a dense urban zip code: high risk. Retired woman in a suburb: low risk. The model worked well enough that no one questioned its foundations.
Progressive questioned them anyway.
Telematics technology had matured enough to make it practical to put a small device in a car and transmit real-time data: hard braking events, miles driven, time of day, speed patterns. The hardware cost had dropped far enough that the economics made sense. All that was missing was someone willing to treat that behavioral data as a strategic asset rather than operational noise.
Progressive saw exhaust data where others saw nothing at all. The industry had effectively decided that granular driving behavior was irrelevant. Progressive decided it was the only thing that actually mattered.
How the flywheel started turning
The initial Snapshot proposition was straightforward: plug in the device, drive safely, potentially earn a discount. For low-risk drivers, it was a good deal. For Progressive, every device in every car was generating something more valuable than premium revenue; it was generating training data.
The pricing models Progressive built from that behavioral data were categorically different from anything their competitors had. Traditional actuarial models could tell you that a 24-year-old male in Atlanta had a certain statistical risk profile. Progressive’s models could tell you that this particular 24-year-old male in Atlanta braked hard an average of three times per hour, drove 60% of his miles between 11pm and 2am, and regularly exceeded posted speeds by more than 15 mph. The granularity of that signal was not incrementally better than demographic proxies; it was a different kind of information entirely.
Better data produced better models. Better models produced more accurate pricing. More accurate pricing attracted the best risks in the market; the safe drivers who knew they were safe and were tired of subsidizing everyone else’s premiums. More safe drivers generated more behavioral data. More behavioral data produced even better models.
The flywheel had no natural stopping point. Competitors watching from the outside could see the program; they could not replicate the data advantage it had already accumulated. You cannot buy twelve years of driving behavior data. You can only generate it, one mile at a time.
What Progressive actually did right
The Snapshot story is often told as a technology story. It was not. The technology was available to every insurer in the market. Telematics devices were not proprietary; the transmission protocols were not trade secrets. Several competitors eventually launched their own usage-based insurance programs. Most of them remain marginal.
Progressive did something harder than deploying a device. They reframed what insurance data was for.
The industry had always thought about data in terms of risk pools: how do we assign individuals to groups with known characteristics? Progressive started thinking about data in terms of individuals: how do we build a model of this specific person’s actual behavior? That conceptual shift drove every subsequent decision, from product design to pricing architecture to the long-term commitment to growing the dataset even when early returns were uncertain.
They also made a commitment that most organizations find genuinely difficult: they were willing to disadvantage their existing customers to build the future model. Snapshot’s best value proposition went to safe drivers who had previously been subsidizing riskier drivers in their demographic cohort. Pulling those safe drivers out of shared risk pools meant the old actuarial model’s cross-subsidies started showing up as losses. Progressive accepted that transition cost because they understood what they were building.
The willingness to let the old model break in service of the new one is rarer than it should be.
The Innovative Question your company hasn’t asked yet
Progressive’s question — “what if we priced risk based on how people actually drive?” — is an instance of a broader class of question that unlocks category-level competitive advantages: “What can we personalize with data and AI that we couldn’t personalize before, and how much would our customers appreciate that?”
The reason this question is powerful is that personalization historically required human judgment at scale, which meant it was expensive and therefore rationed. A private banker can personalize service for 50 clients. An insurer with 15 million policyholders cannot personalize anything without a systematic data advantage. AI changes the economics of personalization so fundamentally that many assumptions about what is and isn’t personalizable at scale are simply wrong now.
The follow-on question, which most CEOs haven’t asked in a structured way, is this: “What data are we generating every day that we’re currently treating as exhaust?”
Progressive’s answer was real-time driving behavior. The data existed; it was just flowing to nowhere. The value was latent, not absent.
Your answer might be different. Consider what Progressive’s logic looks like in adjacent domains.
A regional distributor has transaction timing data for thousands of customer orders. They know that certain customers order in anticipation of price volatility, certain customers order reactively when inventory gets tight, and certain customers follow patterns correlated with their own downstream demand cycles. That behavioral signal, accumulated across five years of order history, is a better predictor of customer health and retention risk than any survey or sales call. It is probably sitting in an ERP system that no one has analyzed.
A specialty manufacturer has warranty claim data with failure timestamps, product configuration codes, and geographic deployment patterns. Individually, each claim is a cost center and a nuisance. In aggregate, that data contains a predictive model of where the next failures will occur before they become claims; which means it contains the raw material for a proactive service offer that competitors offering reactive warranty programs cannot match.
A professional services firm has interaction logs: which clients scheduled calls proactively versus which ones had to be chased, which projects generated repeat requests on adjacent topics versus which ones ended cleanly, which partners generated referrals and which didn’t. That interaction pattern data is a map of relationship quality at a level of granularity no CRM was designed to capture. It could drive client expansion strategies with a precision that annual relationship reviews cannot.
The pattern is the same in every case. Data that appears to be operational residue, the byproduct of doing the actual work, turns out to contain a model of customer behavior more accurate than anything built from demographic proxies, survey responses, or sales team intuition. The companies that recognize this early build pricing advantages, service advantages, and retention advantages that compound over time. The companies that recognize it late spend years trying to close a data gap that grew while they weren’t paying attention.
The personalization lever most mid-market companies aren’t pulling
Mid-market CEOs sometimes look at the Progressive story and file it under “things that work at scale.” That instinct is understandable and mostly wrong.
Usage-based pricing required scale because the underlying insurance product required a large enough pool to remain actuarially sound. But the data flywheel logic does not require scale; it requires commitment. A company with 400 customers that builds a behavioral data model of those 400 customers owns something no competitor can acquire: twelve months of granular interaction history that reveals which offers land, which service moments drive expansion, and which early signals predict churn.
The question isn’t whether your company is big enough to benefit from behavioral data. The question is whether you’ve started collecting it with enough intentionality that it’s actually useful.
Most mid-market companies are collecting data in the sense that their systems generate logs. They are not collecting data in the sense that anyone has defined what behavioral signals matter, designed instrumentation to capture them cleanly, and built the analytical infrastructure to turn them into decisions. The gap between “we have data” and “we have a data asset” is where most of the opportunity lives.
AI accelerates the value of closing that gap because the marginal cost of building a personalization model on clean behavioral data has dropped by an order of magnitude in the last three years. The analysis that once required a team of data scientists and six months of development time can now be approximated in weeks. The economic case for investing in data quality and coverage has never been better; and the competitive risk of deferring that investment has never been higher.
The question worth putting in front of your leadership team
Progressive’s Snapshot program succeeded not because they had better technology or more capital than their competitors. They succeeded because someone in a leadership meeting asked a question that hadn’t been asked before, took the answer seriously, and committed to building an organization capable of acting on it.
That sequence is replicable. The question is available to any CEO willing to sit with it long enough to generate an honest answer.
What data are we generating every day that we’re currently treating as exhaust? What can we personalize with that data and AI that we couldn’t personalize before? And how much would our customers actually value that?
The companies that answer those questions well in the next five years will have built something their competitors cannot easily buy. The companies that don’t will spend the following five years watching someone else build it.
Progressive started asking in 2011. The compounding has been running ever since.
