Nobody Sells an Outcome

The cold open

Here’s the question as it ran: what will autonomous buying agents ultimately buy? Options: Impressions. Audiences. Attention. Outcomes. Core thesis at launch: the unit of trade may move from media delivery toward defined commercial risk and results.

The hidden word in the question is “buy.” A unit of trade needs a seller, and a seller needs to be able to deliver what the invoice says. An impression can be delivered. An audience can be claimed. Attention can be measured, more or less. An outcome — a sale, a signed contract, a customer who stays — can only be warranted, because the seller doesn’t fully control whether it happens. Warranting something you don’t control has a name in every other industry: underwriting. So the poll was secretly asking whether media will grow an underwriting function, and 57% of the room said yes without anyone using the word.

The room has said it before. Season 1’s Week 7 poll asked what an agent should optimize for, and Business Outcomes took 54% of 51 ballots. That essay argued an outcome can’t be a single number an optimizer chases, because it’s judged after the fact by someone with a P&L. This week asks the harder version: not what the agent aims at, but what the money actually purchases. Same instinct, one layer deeper into the plumbing — and the majority grew a little.

The vote

The poll closed with 73 votes. (Shares are LinkedIn’s rounding, which is why they sum to 98.)

Seventy-three ballots is one of the largest turnouts of the run — only the two Season 1 closers drew more — and 57% is the clearest majority Season 2 has produced. The first two weeks of this season went against the launch thesis: Week 1 picked the incumbent asset over the mandate, Week 2 split down the middle at 16 votes. This one endorsed the thesis, at four and a half times the turnout. Whatever else the season shows, it isn’t a room that votes for whatever the question implies.

The number worth staring at is the 27%. A quarter of the room, offered the future, voted for the unit the industry already trades. The easy reading is nostalgia. I think the accurate reading is that they answered a different and better question — not what do buyers want but what can actually be sold — and by the end of this essay I’ll argue both camps are right about their own question. What connects them is the machinery this poll never named.

Outcome Underwriting

Outcome Underwriting: for an outcome to become a unit of trade, four clauses have to hold at once. A definition both sides accept — what counts as the outcome. A measurement both sides trust — who counts it, with what access. A settlement window — when the count is final and disputes close. And underwriting proper — who pays when the outcome doesn’t happen. The first three are measurement work. The fourth is capital.

Media has spent twenty years working on the first three and almost no time on the fourth, because the fourth is the expensive one. The seller of an outcome is exposed to everything between the ad and the result: the product’s price, the landing page, the sales team, the competitor’s launch, the customer’s mood. Underwriting means pricing all of that risk into the unit, the way an insurer prices a policy or a bank prices a loan. Nobody in the media supply chain currently holds reserves against outcomes that fail to arrive, and a guarantee without reserves behind it is a slogan.

There’s a second, quieter problem stacked on top. An agent spending money doesn’t want outcomes; it wants incremental outcomes — results that wouldn’t have happened without the spend. iROAS is a negotiation, not a number, precisely because incrementality is a claim about a counterfactual world, and counterfactuals can’t be invoiced. You can count conversions. You cannot count the conversions that would have happened anyway, only estimate them, and every estimation method is a term to be argued over. A market where the unit itself is the output of a disputed model needs its settlement machinery before it needs its demand — and demand is the only part this poll measured.

Run the four

Impressions. The only option on the list with all four clauses built. Defined to the pixel and the millisecond, counted by systems both sides audit, settled on a schedule, disputed within known bounds — a century of plumbing, from the gross rating point to the viewable impression, all of it built so that two companies who don’t trust each other can close the books. That’s what the 27% voted for, and it isn’t nostalgia. It’s an accurate description of the only unit that can currently survive an argument between a buyer and a seller who both hired lawyers.

Audiences. A claim about who saw it, priced as a data fee rather than a guarantee — which tells you how much the sellers themselves trust the clause structure underneath it. Verification is thin, decay is fast, and no audience seller refunds you when the segment turns out to be someone else. Eight percent reads about right for a unit the market has already quietly repriced from promise to input.

Attention. Measurable, increasingly well, and unwarrantable, permanently. A seller controls the opportunity for attention — the placement, the format, the sound-on rate — but not the attention itself, and a unit of trade priced on something neither party controls is a proxy wearing a lab coat. The 6% who picked it are measuring something real. They’re just measuring a diagnostic, not a currency.

Outcomes. What buyers actually want, and the only option nobody can currently sell at scale, because clause four has no institution behind it. Fifty-seven percent is a demand signal, not a market description. Demand signals matter — markets get built toward them — but the distance between this vote and a tradable unit is the distance between wanting insurance and founding Lloyd’s.

The private outcome markets

Here’s the wrinkle that makes the 57% less speculative than it looks: outcome buying already exists. It just doesn’t exist between companies.

Inside Google’s and Meta’s auctions, a buyer can bid a target cost per acquisition or a target return on ad spend, and the platform’s machinery reprices toward it. That is outcome optimization with the first three clauses present, definition, measurement and settlement, operated by the same company that sells the media, and with a loop dense enough to adjust price before the next cycle. (Revised August 22, 2026 — this paragraph originally described the platforms as “self-insuring” the fourth clause “across billions of daily events.” That was wrong, and the follow-up essay corrects it: their own filings say no fixed price per action and no guaranteed target. Owned loops optimize and reprice; they do not insure the outcome. Nobody does.) The affiliate channel figured out a cruder version decades ago: pay only on the conversion, and the publisher eats the delivery risk. Both work. Neither is a market. One is a product feature inside a walled garden; the other is a niche that never scaled past the bottom of the funnel, partly because its own settlement clause — the last-click — was never really agreed either, just tolerated.

So the real question under this poll isn’t whether outcomes can be bought. They’re bought every day. It’s whether outcome contracts can exist across a company boundary, between a buyer’s agent and a seller’s agent who don’t share an owner — with a definition neither controls, a measurement both can audit, and someone underwriting the gap. That is exactly the layer the Season 1 finale found missing when it stacked eight weeks of polls into one diagram and marked the ninth layer — Issuance, the institution that mints, prices, and revokes the guarantee — as the one that exists nowhere. Fifty-seven percent of this room just voted to need it.

What would build clause four

Three mechanisms, in rising order of ambition.

Holdouts as contract terms. The counterfactual problem shrinks the moment an agreed holdout design is written into the deal rather than argued after it — a percentage withheld, a duration, a pre-registered readout both sides accept before a dollar moves. The methods exist; what’s missing is their promotion from analytics practice to contract clause. Measurement for agentic commerce is largely the story of that promotion.

Verification with standing. Somebody neutral has to hold the count. Not a dashboard the seller renders, and not a model the buyer tunes — a third party whose read is final within the settlement window, the way a clearinghouse’s read is final. The agentic protocols are already circling this: when buyer and seller agents negotiate a deal, the question of whose measurement settles it is the hardest clause on the table, which is precisely why it’s this season’s Week 4 question.

Capital that prices the miss. The genuinely unbuilt part. Someone — an insurer, a guarantee fund, a platform willing to put capital behind a target rather than merely price toward it, a seller large enough to warehouse its own risk — has to hold reserves against outcomes that don’t arrive, and charge for it. The day a media contract carries an underwriting premium as a line item is the day “buy the outcome” stops being a slogan. Watch for the premium, not the press release.

My vote

I voted Outcomes, and the honest version of that vote is: I voted for where the market ends, not for what’s tradable this quarter.

The uncomfortable reading of the 57% — my own ballot included — is that an operator voting “outcomes” is voting to hand risk to a counterparty who hasn’t agreed to take it and currently has no way to price it. That’s what buying an outcome is: risk transfer. The reason it feels like a purchasing preference rather than an insurance demand is that the walled gardens have spent a decade making outcome bidding feel ordinary while keeping every clause of the contract in-house. The feeling is real. The institution behind the feeling doesn’t exist outside their walls.

And the impression doesn’t die in this future — it gets demoted. When outcome contracts do get written, the miss still has to be reconstructed from something, and the something is delivery: what ran, where, when, seen by whom. The unit stops being what you buy and becomes what you settle with, the way a shipping manifest stopped being the product and stayed the paperwork. The 27% weren’t defending the past. They were describing the receipt layer of the future, accurately.

What the next weeks test

This result handed its own hardest clause to next week, and next week answered. Week 4 closed asking what buyer and seller agents negotiate hardest, and the room picked success criteria over price by two points — clause one of the outcome contract, named as the fight. Success criteria and liability are clauses one and four of this contract by other names, and the follow-up essay reads the near-split between semantics and economics. Week 5 then asks which intermediary function survives when agents transact directly, and two of its four options — verifying results, assuming risk — are the settlement and underwriting layers wearing job titles.

The reading in this essay is checkable the ordinary way. If, by 2028, outcome-priced deals between independent companies remain a rounding error while impression-settled contracts carry on — no underwriting premium anywhere as a line item — then the 27% were right about the decade and the 57% only about the destination. If a premium shows up — if anyone starts charging, visibly, to warrant a result they don’t control — clause four is under construction and this room called it early. Both are public facts on a normal disclosure schedule. The poll’s question was what agents will ultimately buy. The market’s question is who sells it, and that one is still open.