What separates a number that impresses from evidence an underwriter can price.
Any vendor can produce a percentage. “Our technology cut claims by twenty per cent” is a sentence assembled in a pitch deck, and on its own it commits no one to anything. The question an underwriter asks is never whether twenty per cent is a good number. It is whether that number can bear weight — whether it can survive being leaned on in a pricing decision that will be audited, challenged at renewal, and defended to a regulator.
Most evidence cannot. It fails not because the technology does not work, but because the demonstration of it was never built to be priced. The gap between a number and a priceable number is not a matter of degree; it is a set of specific, nameable checks, each of which removes a specific way of being fooled. There are five. An underwriter applies them whether or not they are written down. This is them, written down — what each one tests, what it achieves, and what clearing it unlocks.
The framing that matters throughout: in a market moving from risk transfer to risk control, evidence is the currency. These gates are what turn a claim into currency the market will actually accept.
Attribution: did your intervention cause the effect, or merely accompany it?
What it checks
Is there a counterfactual — a matched control group, or a difference-in-differences design — that isolates the effect of the intervention from everything else that changed at the same time?
What it achieves
It converts correlation into causation, the single most consequential conversion in the entire chain. A before-and-after chart shows that claims fell after the technology was deployed. It does not show that the technology is why they fell. Fleets renew vehicles, hire different drivers, change routes, and benefit from market-wide safety improvements, all while a telematics programme runs. Without a counterfactual, the measured improvement is a coincidence you cannot rule out, dressed as a result.
What it unlocks — and what failing it caps you at
This is the master gate, and it is unforgiving: fail it and nothing downstream can rescue you. A claim with no counterfactual is capped at correlation, however large the sample, however tight the statistics. More data buys a more confident correlation, not an attribution. Clearing it is what first makes an effect yours to claim — the difference between “claims fell” and “we reduced claims,” and only the second can be priced.
An honest baseline: is the improvement real, or is it the mean returning?
What it checks
Was the comparison period a representative stretch, rather than an unusually bad one selected precisely because it was bad?
What it achieves
It strips out regression to the mean — the most seductive way a genuine analysis lies. Insurers and vendors alike are drawn to intervene where the numbers look worst: the depot with the spike, the cohort having a terrible quarter. But extreme periods tend to be followed by ordinary ones regardless of any intervention, simply because they were extreme. Measure from an outlier-bad baseline and a large share of the “improvement” is the mean returning to where it always sat. The intervention gets the credit; arithmetic did the work.
What it unlocks
It makes the size of your effect defensible rather than merely impressive. This is the gate that most often turns a headline figure into a smaller, honest one — and the smaller honest figure is worth more, because it will not collapse when an actuary re-bases the analysis against a representative period. Passing here means your number is a floor you can stand on, not a ceiling you happened to touch.
Persistence: is this a durable effect, or a good quarter?
What it checks
Does the effect hold across multiple periods, or does it rest on a single strong window?
What it achieves
It distinguishes a real behavioural or structural change from noise that happened to fall the right way once. A single period can move for a hundred reasons that will not repeat. An effect that persists is one the underwriter can expect to see again — and expectation of recurrence is precisely what a rate is: a forward-looking price on a repeatable pattern.
What it unlocks
Durability lets an effect be priced forward rather than merely observed backward. A one-period result, however clean, is a historical curiosity; a persistent one is a rating assumption. This is also the gate that converts a pilot into a programme — the point at which “it worked once” becomes “it works,” and the conversation shifts to terms.
Exportability: can the evidence leave the building?
What it checks
Is the evidence ownable and portable — can it be taken to the market, in a usable and verifiable form, at renewal — or is it trapped inside a vendor’s dashboard?
What it achieves
This is the commercial gate, and the one most often overlooked because it is not a statistical property at all. Evidence that cannot be exported cannot be priced, no matter how rigorous, because the party who sets the price cannot hold it, verify it, or carry it into the rating file. A beautifully controlled, persistent, material effect that lives only as a login to a supplier’s platform is, to an underwriter, evidence that does not exist.
What it unlocks
Exportability turns evidence into an asset the insured owns rather than a feature they rent. It is the difference between a discount that evaporates the moment the relationship changes and a credit that travels with the risk. For a risk owner, it decides whether prevention spend builds equity in their own risk profile or simply funds someone else’s product.
Materiality: is the effect large enough to move a price?
What it checks
Is the effect big enough to shift a rate, a loss ratio, or a retained-loss line — not merely large enough to be statistically detectable?
What it achieves
It enforces relevance, along an axis volume cannot rescue. Statistical significance and material significance are different questions: with enough data, a trivially small effect becomes detectable with great confidence, and remains trivially small. An effect too minor to move a price does not become priceable by accumulating more claims. Materiality is orthogonal to credibility, and conflating the two is how programmes end up with an immaculately evidenced result no underwriter will pay for.
What it unlocks
Materiality makes the whole exercise worth running. It connects the evidence to money — the confirmation that clearing the other four will produce not just a defensible number but a consequential one. Pass it, and the effect is large enough that pricing it changes the economics for everyone in the chain.
The opportunity
What clearing all five unlocks
Taken singly, each gate removes a way of being wrong. Taken together, they move an effect up the proof ladder from a number to a priced risk reduction — and the position an effect reaches on that ladder is the position from which it can be commercially claimed.
An effect that clears all five is no longer a marketing statistic. It is a credible, defensible, durable, portable and material demonstration that the risk is genuinely lower — which means the correct price is genuinely lower. That destination has a specific commercial character worth naming plainly.
It compounds. Because it is exportable and persistent, the evidence does not reset at renewal; it accrues, strengthening year on year rather than being re-argued from scratch. It is defensible under scrutiny, because attribution and an honest baseline mean it survives an actuary’s or a regulator’s re-examination rather than dissolving under it. And it reprices rather than discounts. A discretionary discount is margin a carrier concedes to win business, and it disappears the moment the market hardens. A credit earned through these five gates is different in kind: nobody is conceding anything, because the expected loss cost is actually lower. That is a technical premium reduction — the only kind that lasts.
This is the prevention-to-pricing bridge in a single idea. Prevention that cannot clear the gates is a cost centre defending a frequency reduction it cannot cash. Prevention that clears them becomes an asset that reprices the risk — for the vendor, a proposition an underwriter will buy; for the risk owner, equity in their own loss profile; for the insurer, a rate credit that is defensible, durable, and honestly earned.
The order is not incidental
How the gates relate
Attribution comes first and dominates. There is no point strengthening a sample, extending a pilot, or polishing a dashboard if there is no counterfactual underneath — you would be building a taller structure on no foundation. If only one gate can be addressed, it is this one.
Materiality is orthogonal to everything else. It cannot be reached by accumulating data, and it will not improve because the other four gates were cleared. If the effect is too small to move a price, the honest conclusion is that this is not a rating story, and no amount of rigour changes that. Establish it early.
Exportability is technical to satisfy but commercial in consequence, and the one most often left until it is too late — designed out of a pilot from the start, or discovered, at renewal, to have been designed out. Build the right to own and export the evidence into the arrangement before the data starts flowing, not after.
The gates are not hurdles placed in the way of good evidence. They are the specification for it. Meeting them is what turns a number that impresses into currency the market will accept — and in a market where evidence is the currency, that is the whole game.
See where a specific claim sits against these five gates — what is holding it back, and what clearing the rest would unlock.
Open the Evidence Credibility & Prevention Bridge →This is the companion piece to the Evidence Credibility & Prevention Bridge — the interactive tool that assesses where a specific claim sits against these five gates, what is holding it back, and what clearing the remaining gates would unlock.