The Evidence Premium
Season 2's sixth Friday poll asked what wins the bid once every seller can claim performance: 37 ballots put Verifiable outcomes first at 54%, Lowest price second at 21%, Proprietary data 13%, Exclusive supply 10%. The finding is the comparison, not the margin — the same idea polled 21% seven days earlier as verifying results, a job an intermediary does, and 54% here as a property of the offer. Sold as a service, proof polls like overhead; built into the offer, it wins the bid. The essay names the mechanism, the Evidence Premium: the price gap between a claim a buyer takes on faith and the same claim carrying proof the buyer's own side accepts. The asset is acceptability rather than accuracy — a standing arrangement with whoever rules that a number counts — and the accreditation layer already documents the parts list: a definition agreed before the trade, an examiner with nothing riding on the answer, re-examination on a schedule, and the power to withdraw the certificate. Drop the fourth and the other three are decoration. The author voted Verifiable outcomes, then concedes that no body certifies a deal delivered what it promised, so until proof can enter the auction it clears on price and the 21% are right. Checkable close: a rate card quoting the same inventory twice, attested and not.
In English, please
A recurring reader poll asked what wins a sale once every seller can claim its product performs. Four options: being the only place to get something, holding data nobody else has, charging the least, and being able to prove the result actually happened. The question came out of a cost collapse that has already happened. Forecasting how an advertising campaign will perform used to take a team and most of a week; an AI program now writes a believable forecast while it is still reading the request. A forecast became cheap to make and just as cheap to fake on the same day.
Of 37 voters, proving the result won with 54%. Charging the least came second at 21%, exclusive data 13%, being the only supplier 10%. Thirty-seven votes is a small poll of people who follow the subject for a living, so the running order is the finding and the small gaps between the bottom three are not. What the essay dwells on is a comparison with the week before, when the same readers were asked what a middleman still gets paid for and checking results scored 21%. Same idea, same audience, seven days apart. Sold as a service someone else performs, proof looks like overhead; built into the thing being bought, it wins the sale.
The essay names that gap the Evidence Premium: the difference in price between a claim a buyer has to take on trust and the same claim carrying proof the buyer's own side will accept. The premium does not go to whoever has the best number. It goes to whoever can produce a number the other side accepts without reopening the argument, and that takes a standing arrangement with whoever gets to rule that a number counts — something almost no seller has, because a claim checked by the party who profits from it is not really a check. Advertising already built this machinery one layer down: an industry body created in 1963 at the request of the US Congress re-audits measurement companies every year and can take the approval away. Read off that model, proof needs four parts — an agreed definition of what is being measured, an examiner with nothing riding on the answer, a re-check on a schedule, and the power to withdraw approval. Drop the fourth and the other three are decoration.
The author voted for proving the result, and declares an interest: he co-leads an industry working group on measurement and runs a product in that layer, so he is arguing for work he is part of. His concession is bigger than the disclaimer. Nobody currently certifies the thing this poll is about — the existing bodies approve the measuring instrument, not whether a particular deal delivered what was promised — and there is still no way for a seller to attach that record to an offer, or for a buyer's software to refuse an offer that lacks one. Until that gap closes the sale is decided on price, which makes the 21% who picked cheapest a better description of the next two years than his own vote is. The test he applies to his own claims is a small one: take the performance line in your sales deck and ask whose word a buyer would take for it. If the only acceptable answer is your own systems, you are selling reporting and pricing it as evidence.
On this page
The cold open
Here’s the question as it ran: when every seller can claim performance, what wins the bid? Options: Exclusive supply. Proprietary data. Lowest price. Verifiable outcomes. Core thesis at launch: prediction becomes abundant, evidence becomes differentiating.
The question was built around a cost collapse that has already happened. Forecasting performance used to take a planning team, a modelling stack and most of a week. A seller’s agent now produces a plausible projection while it is still parsing the brief. A prediction is easy to make and just as easy to fake, and those two properties arrived on the same day. So the poll wasn’t asking whose pitch is most convincing. It was asking which half of a seller’s pitch survives contact with a buyer who can manufacture the other half in-house.
Last week’s essay closed by putting a marker on this poll. That room had just ranked verifying results at 21% as an intermediary function, and the marker was that the same idea, asked as a buying question, would do better. It did considerably better than that.
The vote
The poll closed with 37 votes.
| Answer | Share |
|---|---|
| Verifiable outcomes | 54% |
| Lowest price | 21% |
| Proprietary data | 13% |
| Exclusive supply | 10% |
Thirty-seven ballots is a straw poll of people who follow this subject for a living, and it forecasts nothing. The order carries the information; the gaps between adjacent options do not. Here’s the demonstration: at 34 votes the leader sat on 56%, and the last three ballots took two points off it. Treat the bottom three as one bunched minority and the top line as the finding.
The number worth sitting with is 54 against 21. Seven days ago this room priced verification at 21% when it appeared as a job an intermediary does. This week the same idea, worded as verifiable outcomes and placed on the buy side of a bid, took 54%. Same feed, seven days apart, two and a half times the support.
That is not a room changing its mind. It’s a room answering two questions that happen to share a noun. Last week verification was a role somebody else fills: a third party, a line item, someone else’s QA team. This week it was a property of the thing being bought. Sold as a service, proof polls like overhead. Built into the offer, it wins the bid. Which is an uncomfortable result for anyone building a verification business, because buyers will pay a seller more for a proven claim and will not pay a verifier much for proving it.
The Evidence Premium
The Evidence Premium: the price gap between a claim a buyer has to take on faith and the same claim carrying proof the buyer’s own side would accept. It is not collected by whoever has the best number. It goes to whoever can produce a number the counterparty accepts without reopening the negotiation.
That capability is an asset, and it is worth naming which one, because most sellers assume they already own it. Inventory is a supply position. Data is an input to a forecast. Acceptability is neither. It is a standing arrangement with whoever gets to rule that a number counts, and almost no seller holds one. A claim checked by the party who profits from the claim is not a check, however good the checking is.
The industry has built this machinery once already, one layer down, and the parts are documented. The accreditation layer separates two questions that agents will collapse if nobody stops them: the research question is whether a measurement method is scientifically sound, and the accreditation question is whether a specific vendor’s number is trustworthy enough to move real money against. Research proves a method is valid. Accreditation proves a number is transactable. The Media Rating Council was established in 1963 at the request of the US Congress, and the detail that matters here is not its scope but its clock — accreditation runs on an annual independent CPA audit and a vote of the MRC board, renewed every year rather than won once. It can lapse. The US Joint Industry Committee, announced in January 2023, does something narrower and more commercial, certifying cross-platform premium-video and TV measurement vendors as fit to transact real money against, with buyers and sellers setting the certification criteria jointly. That last part is the piece vendors keep trying to skip, because a standard written by the party being measured is a brochure with a methodology section.
Read those two bodies as a parts list and an evidence premium needs four components. A definition of what is being measured that both sides accepted before the trade. An examiner with nothing riding on the answer. A re-examination on a schedule. And the power to withdraw the certificate. Drop the fourth and the other three are decoration.
Note also who walks into the audit. The party making the claim is the party that applies, which is why the premium in this vote accrues to sellers rather than to auditors. That is the same conclusion the underwriting essays reached about where demand for protection actually sits. It also has a market-structure consequence most decks have backwards. Producing acceptable evidence has the cost shape of infrastructure: expensive to stand up, expensive to keep current, close to free on the marginal deal. Assets with that shape consolidate. A handful of parties will carry the cost of being acceptable, everyone else will rent acceptability from them, and the rent will arrive inside the media price whether or not anybody calls it a measurement budget.
Run the four
Exclusive supply. Ten percent, last, and the only claim on the ballot that proves itself. Either the inventory is available from somebody else or it isn’t, and no auditor is required to settle it. That is also its ceiling. A self-proving claim carries no evidence premium because there is no doubt for evidence to resolve, and the moment exclusivity has to be defended on performance rather than access, it becomes an ordinary claim needing ordinary proof. Exclusivity commoditises from the other end too: agents are efficient substitute-finders, and most exclusivity survives only until a substitute is described well enough to bid on.
Proprietary data. Thirteen percent for the input-side bet: if the data can’t be copied, the prediction built on it can’t be copied either. The bet is reasonable and it walks into a wall. Proprietary means the buyer cannot inspect it, which makes it an explanation for a forecast rather than evidence for one, offered into a market that already discounts forecasts. There is a real bind inside this option that nobody has solved commercially: transparency destroys the asset and opacity destroys the proof. The accreditation layer is the institutional answer to exactly that bind. Audit the method under confidentiality, publish the verdict, keep the data. Sellers who think their moat is the data should notice that the workaround for their problem was built sixty years ago and does not belong to them.
Lowest price. Twenty-one percent, second, and the most economically literate minority vote of the season. It is not a claim that cheapness is a virtue. A buyer who cannot tell two claims apart is holding two identical goods, and the correct move on identical goods is to take the cheaper one. Commoditisation never required the products to be the same, only indistinguishable. So the 21% are not voting against evidence. They are forecasting that evidence will not arrive quickly enough to matter, and that agents will do what buyers have always done without a quality signal, only faster and without a relationship to slow it down. They are describing the next two years more accurately than the 54% are, and I’ll pay for that concession below.
Verifiable outcomes. Fifty-four percent, and the adjective is carrying the vote. Most of what ships as verification today is seller-reported measurement with a dashboard bolted on, which settles nothing at the moment of the bid. The version that wins a bid has to answer the accreditation test in miniature, before the trade clears: measured against whose definition, examined by whom, valid until when, and what happens when the check fails. Week 3 named the settlement clause nobody had built. Week 4 found the room fighting hardest over what counts as success. This week’s 54% is those two findings with a price attached.
My vote
I voted Verifiable outcomes, and I have an interest to declare: I co-lead AdCP’s Signals and Measurement working group and run a signals agent, so I am arguing for a layer I work on.
The concession is larger than the disclaimer. No body accredits the thing this poll is about. MRC accredits measurement products. The US JIC certifies currency vendors. Both validate the instrument, neither stamps this deal delivered what it promised, and there is no agent-readable way today for a seller to attach a settlement record to a bid or for a buyer’s agent to reject on its absence. Until that gap closes, proof cannot enter the auction at all, so the auction clears on price, and a buyer optimising this quarter should behave exactly as the 21% say. My vote is a claim about where the margin sits once the gap closes. It says nothing useful about next year’s median transaction.
The harder concession is what the jump from 21% to 54% implies about who funds the check. It suggests verification may never become something buyers purchase separately, and a proof layer that exists only as a seller’s feature is a proof layer the seller controls. The independent examiner still has to sit somewhere and still has to be paid. This vote says it gets funded as a cost of selling, which is approximately how the audit profession is funded, and approximately how that profession fails when it fails. The test I would run on my own claims is the small one: take the performance line that appears in every deck you send, and ask whose word a buyer’s agent would take for it. If the only acceptable answer is your own systems, you are selling reporting and pricing it as evidence.
What next week tests
No. 15 is live: if an agent chooses the brand, who owns the customer relationship — the consumer, the brand, the agent platform, or the data holder? It moves the season from what settles a trade to who holds the asset once the trade is done. Watch the fourth option, because “the data holder” is this week’s winner in ownership clothing. Whoever holds the record of what was promised and what happened has a claim on the relationship whether or not anybody granted them one.
The reading here is checkable, and not by waiting for a vendor announcement. There are two tells. The first is a rate card that quotes the same inventory twice, one price with an independently attested outcome and one without, with buyers paying the spread. The second is a field on a machine-readable offer naming who attested to a claim and when that attestation expires, in a form a buying agent can reject on parse. My prediction is that the first sellers charging an evidence premium will be ones already submitting to a recurring audit, because they have paid the fixed cost of producing an evidence trail and everyone else still has to build one. If the premium shows up first on self-attested claims, then buyers are paying for the adjective, and the lowest-price bloc will be right for a good deal longer than this room thinks.