Operating Playbook

Outcome Underwriting.

Operating model for turning performance promises into priced guarantees — outcome definition, neutral observation, settlement windows, restatement policy, and the capital that carries the miss.

We can promise the outcome. We cannot price the miss.
Evgeny Popov
Scope this playbook → See the four clauses
Fast read · Best for

Sellers, intermediaries and platforms being asked to promise a business result they do not fully control.

Evgeny Popov Evgeny Popov
playbook/outcome-underwriting Operating playbook
Not for
Teams that want a performance narrative, a new dashboard, or best-effort language rebranded as a guarantee.
Primary buyer
CEO, CRO, CFO, general counsel, commercial finance, measurement, and partnership leaders.
Primary output
Outcome definition, observation regime, settlement and restatement policy, carrier and reserve model, and the term sheet that carries them.
Main risk
Warranting a result with no reserve behind it and no window in which the count goes final.
Best next step
See the six-phase sequence, or scope the operating model. →
Executive summary

A promise is not an instrument.

An outcome can be delivered by nobody. It can only be warranted, because the seller does not control everything between the exposure and the result — the price, the product, the landing page, the sales team, a competitor launch, a policy change. Outcome underwriting is the work of pricing that gap into the unit instead of leaving it with whoever has least leverage when the invoice comes due.

Plate 01 · Signature figure
Outcome underwriting as a signed instrument: four clauses, the fourth sealed A certificate with four clauses: definition, measurement and settlement signed in a thin pen line; underwriting signed heavy and sealed, because it is capital. Behind it, the same paper as it usually ships — promised, unsealed, no price for the miss. The sold instrument reads: one definition · one count; one window · one payer. Promised: no price for the miss Promised Outcome, sold 1 · Definition: what counts as the outcome — measurement work 1 Definition what counts as the outcome 2 · Measurement: who counts it, with what access — measurement work 2 Measurement who counts it, with what access 3 · Settlement: when the count is final — measurement work 3 Settlement when the count is final 4 · Underwriting: who pays for the miss — capital 4 Underwriting who pays for the miss Underwriting is capital — the clause nobody in the supply chain holds 4 one definition · one count one window · one payer
Four clauses converging into one tradeable unit. Definition, measurement and settlement are measurement work and draw as thin line; underwriting is capital and draws heavier, because it is the one nobody in the supply chain holds. Hover a block for detail.

Buyers want outcomes. Sellers keep promising them. What almost nobody has built is the machinery underneath: a definition both sides accept, a measurement both sides trust, a window in which the count goes final, and a party that pays when the outcome does not arrive. The first three are measurement work, and the industry has spent twenty years on them. The fourth is capital, and it is the expensive one.

Warranting something you do not control has a name in every other industry. This playbook does that work for commercial teams: what gets warranted, who counts it, when the count closes, what the remedy is, and whose balance sheet is behind it. Selecting the pricing model in the first place is a different job — see Commercial Productization.

  • Define the outcome down to a counted event before anyone prices it.
  • Name who holds the pen on the count, and write that into the contract as a term.
  • Set the settlement window deliberately — its width is a price, and somebody pays it.
  • Write the restatement policy before the first correction arrives, not after.
  • Name the carrier. A guarantee with no reserves behind it is a slogan.
  • Screen what you will warrant; steerable and correlated outcomes are the ones that break.
  • Record forecast against delivered on every deal — loss history is what lets a premium be repriced.
  • Watch for the premium as a line item, not the press release.
A guarantee with no reserves behind it is a slogan.
Evgeny Popov
Symptoms

You need this when…

The tell is rarely a weak product or a bad measurement stack. It is a promise that was never written down as an instrument, and it shows up as the same operating symptoms every time.

  • Sales is promising a business result the contract never defines.
  • Both sides are counting, and the two numbers do not reconcile.
  • The count that settles the invoice is rendered by the party being paid.
  • A campaign closed months ago and the number is still being argued about.
  • A correction landed after the invoice and reopened the whole deal.
  • Finance cannot say what a miss would cost, because nobody priced it.
  • The word "guaranteed" appears in the deck and nowhere in the master agreement.
  • Misses settle in more inventory, and nobody in the building calls that a claim.
  • Incrementality is asserted in the pitch and renegotiated in the review.
  • Nobody can name the party who is out of pocket when the outcome does not arrive.
Scope

This is not a pricing exercise.

Most attempts to fix this reach for a billing unit. But the billing unit is the easy half — the hard half is what happens when the unit does not arrive.

  • Not a pricing-model selection. Commercial Productization selects the model — subscription, usage, spend-linked, outcome-linked — and stops there. This playbook starts where that stops: it designs the warranty, the observation regime, the settlement window, the reserve, and names who eats the miss.
  • Not a measurement project. Definition, observation and settlement are three of the four clauses, and no measurement stack supplies the fourth, because the fourth is capital.
  • Not a performance-marketing narrative. Optimising toward a target is effort. A guarantee is answerability.
  • Not a rebrand of best-effort. A best-effort clause obliges work, not a result. Nothing in it can be claimed against.
  • Not an incrementality methodology. Every method for estimating the counterfactual is itself a negotiated term, which is why the method belongs in the contract rather than in the appendix.
  • Not an insurance product you can buy off a shelf today. It is the operating work that has to exist before one can be written.
  • Not a protocol or a schema. A specification can carry a fixed price on a specified event. It cannot make anyone pay when the event does not come.

Commercial Productization selects the pricing model and stops. This playbook starts there: it designs the warranty, the observation regime, the settlement window, the reserve — and names who eats the miss.

The instrument

Four clauses, holding at once.

This is a conjunction, not a sequence. The four clauses are piers under one beam: a definition both sides accept, a measurement both sides trust, a settlement window where the count goes final, and underwriting that pays for the miss. Pull any one of them and the beam drops — the outcome reverts to a promise. The asymmetry is in the footing. The first three rest on measurement work the industry has been building for two decades. The fourth bears exactly the same load with nothing visible beneath it.

Three clauses are a measurement programme. Four clauses are an instrument. The difference between them is a balance sheet.

Plate 02 · The four clauses
THE FOUR-CLAUSE INSTRUMENT all four, at once — or none The unit of trade — an outcome that can be invoiced, counted, settled and collected on. It is not a property of the outcome itself; it is what the four clauses hold up. Remove any one pier and the beam drops. THE OUTCOME AS A UNIT OF TRADE bears the same load +++ Definition — what counts as the outcome. A pier of the instrument: schemas, event taxonomies and match keys, still written bilaterally rather than as a standard. Without it there is nothing for either side to count. 1 Definition what counts asthe outcome both sides accept it WITHOUT IT nothing to count Measurement — who counts it, and with what access. A pier of the instrument: attribution, holdouts and incrementality methods, each of which is itself a negotiated term. Without it the count belongs to whoever renders the dashboard. 2 Measurement who counts it,with what access both sides trust it WITHOUT IT no count both sides own Settlement — when the count is final and disputes close. A pier of the instrument: a window, say T+30, after which the number stands or is formally restated. Without it a result can be reopened forever, and a number that can always be reopened cannot settle an invoice. 3 Settlement when the count is finaland disputes close an agreed date, then restated WITHOUT IT it never goes final Underwriting — who pays when the outcome does not happen. This pier carries the same load as the other three, and nothing stands under it: no reserves against outcomes that fail to arrive, no price on the failure case, no institution inside a wall or outside one. The first three clauses are measurement work. This one is capital. 4 Underwriting who pays when theoutcome does not happen capital, held in reserve WITHOUT IT nobody pays the miss What clause 1 rests on: measurement work — schemas, audits and dispute practice the industry has worked at for twenty years. The ground is real, and it is still partly built. measurement work partly built What clause 2 rests on: measurement work — schemas, audits and dispute practice the industry has worked at for twenty years. The ground is real, and it is still partly built. measurement work partly built What clause 3 rests on: measurement work — schemas, audits and dispute practice the industry has worked at for twenty years. The ground is real, and it is still thin. measurement work thin The missing footing — the fourth pier rests on nothing. No insurer, no guarantee fund, no platform putting capital behind a target rather than merely pricing toward one. A guarantee with no reserves behind it is a slogan. capital no institution Any one clause missing and the outcome is a promise, not a unit of trade. The first three clauses are measurement work. The fourth is capital.
The four clauses of an outcome contract
ClauseWhat it settlesThe testFootingState of the artWithout it
1 · DefinitionWhat counts as the outcome.Both sides accept it.Measurement workPartly built — written bilaterally, deal by deal, rather than as a standard.Nothing to count.
2 · MeasurementWho counts it, and with what access.Both sides trust it.Measurement workPartly built — attribution, holdouts and incrementality methods, each itself a negotiated term.No count both sides own.
3 · SettlementWhen the count is final and disputes close.A stated width, then restatement.Measurement workThin — dispute windows exist in direct deals, thinly at scale.It never goes final.
4 · UnderwritingWho pays when the outcome does not happen.Capital, held in reserve.CapitalNo institution behind it — the clause the rest of this playbook is about.Nobody pays the miss.
The distinction

A performance promise and a priced guarantee are not separated by effort.

Both carry the same target, the same best-effort clause, and the same optimisation loop. The promise stops there. The guarantee carries those three forward and adds the three the promise structurally lacks: a neutral observer, a settlement date, and a party that pays when the outcome does not arrive. Those three are not harder work. They are capital and answerability, which is why no amount of additional effort converts one column into the other.

Only the third kind pays the buyer when the result misses. The first two make the seller work harder, or thinner, and leave the outcome risk exactly where it started.

Plate 03 · Promise vs guarantee
OUTCOME UNDERWRITING what has to be added before an outcome can be traded Performance promise — a target, an effort clause and a loop. Everything in it is work the seller controls, which is why nothing in it can be claimed against. PERFORMANCE PROMISE what the deck says Priced guarantee — the same three clauses, plus a neutral observer, a settlement date and a party that pays the miss. Only this column can be claimed against. PRICED GUARANTEE what can be claimed against The target — the number both sides aim at. A target is a statement of intent. Both a promise and a guarantee name one; naming it settles nothing. 1 The target the number both sides aim at Best-effort clause — commercially reasonable efforts. Best effort is the clause that converts a number into a direction. It obliges work, not a result. 2 Best-effort clause commercially reasonable efforts The optimisation loop — reprice, retarget, try again. Optimisation moves toward the target and reprices inside the seller system. It is effort, and effort is the part both columns already have. 3 The optimisation loop reprice, retarget, try again A neutral observer — absent from a performance promise. Answerability, part one. A count held by a party with standing, final within the window. A dashboard the seller renders is not a neutral count. absent A settlement date — absent from a performance promise. Answerability, part two. Without a date on which the count is final, there is no moment at which anyone can be held to the number. absent A party that pays the miss — absent from a performance promise. Capital. Someone holds reserves against the outcome that does not arrive, and charges for holding them. A guarantee with no reserves behind it is a slogan. absent carried over unchanged the difference is not effortit is capital and answerability The target — the number both sides aim at. A target is a statement of intent. Both a promise and a guarantee name one; naming it settles nothing. 1 The target the number both sides aim at Best-effort clause — commercially reasonable efforts. Best effort is the clause that converts a number into a direction. It obliges work, not a result. 2 Best-effort clause commercially reasonable efforts The optimisation loop — reprice, retarget, try again. Optimisation moves toward the target and reprices inside the seller system. It is effort, and effort is the part both columns already have. 3 The optimisation loop reprice, retarget, try again A neutral observer — a counter neither side employs. Answerability, part one. A count held by a party with standing, final within the window. A dashboard the seller renders is not a neutral count. 4 A neutral observer a counter neither side employs + A settlement date — agreed in advance — books close, disputes end. Answerability, part two. Without a date on which the count is final, there is no moment at which anyone can be held to the number. 5 A settlement date agreed in advance — books close, disputes end + A party that pays the miss — reserves held against the miss. Capital. Someone holds reserves against the outcome that does not arrive, and charges for holding them. A guarantee with no reserves behind it is a slogan. 6 A party that pays the miss reserves held against the miss + 3 clauses of effort the same 3, plus 3 that bind someone We can promise the outcome. We cannot price the miss.
Performance promise compared with a priced guarantee
ClauseWhat it isPerformance promisePriced guaranteeWhy it matters
The targetThe number both sides aim at.CarriedCarriedA statement of intent. Naming it settles nothing.
Best-effort clauseCommercially reasonable efforts.CarriedCarriedConverts a number into a direction. It obliges work, not a result.
The optimisation loopReprice, retarget, try again.CarriedCarriedMoves toward the target inside the seller system. This is effort, and both columns already have it.
A neutral observerA counter neither side employs.AbsentAddedAnswerability, part one. A dashboard the seller renders is not a neutral count.
A settlement dateThe books close and disputes end.AbsentAddedAnswerability, part two. With no date on which the count is final, there is no moment at which anyone can be held to the number.
A party that pays the missReserves held against the outcome that does not arrive.AbsentAddedCapital. Someone holds the reserve and charges for holding it.
Show all 6 rowsShow fewer rows

Three things an outcome-priced product can be

The industry uses one word for all three, which is why deals that look alike settle so differently.

Three kinds of outcome-priced product
KindWhat it isWho holds the riskThe tell
Optimisation toward an outcomeThe seller observes results, reprices and retargets toward your goal.The buyer keeps the outcome risk. The seller carries commercial exposure — thinner margin, lower spend next quarter.The target is described as a goal, and the documentation declines to guarantee it.
Action-based billingThe seller buys exposure and is paid only when a defined action fires.The seller carries the variance between the exposure and the action — real risk transfer, at one layer.Eligibility screens: enough events, arriving fast enough, each small enough.
A guarantee of the outcomeA counterparty pays when the business result does not arrive.A named party holds reserves against the miss and charges a premium for holding them.A premium on the invoice, and a remedy that moves money rather than inventory.
Clause one

An outcome cannot be priced until it has been defined down to something countable.

Four rungs, and each one supplies the thing the rung below was missing. The annotation at every climb is the decision that had to be taken to get there: which result in whose P&L, what event evidences it, then who counts it and when the count closes. A price line sits between the third rung and the fourth. Everything below it can be promised. Only what sits above it can be underwritten.

A business aspiration has no counterparty, no counter and no unit. It can be agreed with forever and never traded once.

Plate 04 · The definition ladder
OUTCOME DEFINITION LADDER Each rung supplies what the one below lacks ↑ Business aspiration — grow the business, win the category. Where the brief arrives. An aspiration has no counterparty, no counter and no unit — it can be agreed with and never traded. 1 Business aspiration grow the business, win the category MISSING: A CLAIM Claimed outcome — a promise, with nothing to count yet. A claimed outcome can be written into a deck and even into a contract, but nothing in it says what would have to happen for the promise to have been kept. 2 Claimed outcome a promise, with nothing to count yet MISSING: AN EVENT Decided: which result, in whose P&L one claim replaces the aspiration Observable event — visibly happens; nobody owns the count. An observable event is visible to both sides and owned by neither. Without a named counter and a closing date it produces an argument, not an invoice. 3 Observable event visibly happens; nobody owns the count MISSING: OWNER + WINDOW Decided: what event evidences the claim a promise becomes a thing that happens Counted event — one counter, one agreed window, and it closes. The only rung above the price line. A counted event with a named owner and a settlement window is a unit: it can be quoted, warranted, and paid against when the miss happens. 4 Counted event one counter, one agreed window, and it closes CAN CARRY A PRICE Decided: who counts, and when it closes and what a restated count does to the invoice The price line — below it an outcome can be promised but not underwritten, because there is no agreed count, no owner of the count and no window in which it becomes final. Above it the outcome is a unit, so a premium for the miss can be quoted. THE PRICE LINE counted is not the same as incremental — still a term to argue A brief arrives on the bottom rung. Nothing below the price line can be underwritten. We can promise the outcome. We cannot price the miss.

  1. Business aspiration

    Grow the business, win the category. Where the brief arrives.

    MissingA claim. An aspiration has no counterparty, no counter and no unit — it can be agreed with and never traded.

  2. Claimed outcome

    A promise, with nothing to count yet.

    Decidedwhich result, in whose P&L.

    MissingAn event. It can be written into a deck and even into a contract, and nothing in it says what would have to happen for the promise to have been kept.

  3. Observable event

    It visibly happens; nobody owns the count.

    Decidedwhat event evidences the claim.

    MissingAn owner and a window. Visible to both sides and owned by neither, it produces an argument rather than an invoice.

  4. Counted event

    One counter, one window, a date the count closes.

    Decidedwho counts, when it closes, and what a restated count does to the invoice.

    Above the price lineA counted event with a named owner and a settlement window is a unit: it can be quoted, warranted, and paid against when the miss happens.

Clause two

Who counts it, and with what access.

One disputed count between a buyer and a seller, answered three ways. Seller-counted is the cheapest read and the first one attacked once money rides on it. Buyer-counted costs the buyer and is trusted by exactly one side, because an agent employed by one party cannot check the other party numbers. Independently counted costs a line item and is the only read that binds inside a settlement window. What you pay to be counted is what you buy in dispute resistance — which makes choosing the observer a commercial decision, not a technical one.

If the count only steers spend, the cheap read is enough. If the count triggers a payment, only an independent read survives the argument.

Plate 05 · Neutral observation
WHO COUNTS, AND WITH WHAT ACCESS The count is the outcome contract’s second clause: who counts it, and with what access. Both sides book against the same number, and only one of them holds the pen. BUYER SELLER the count one number, two sets of books Seller-counted — the seller renders the number. Access: complete access, and no audit from the other side. Cost: cheapest — folded into the price. In a dispute: disputed once money rides on it. Use it when nothing is owed on the number: it steers spend, it never settles it. Seller-counted the seller renders the number WHO HOLDS THE PEN the seller, out of its own logs ACCESS complete access, and no auditfrom the other side COST cheapest — folded into the price IN A DISPUTE disputed once money rides on it USE IT WHEN nothing is owed on the number:it steers spend, it never settles it Buyer-counted — the buyer brings its own counter. Access: partial — blind where the seller’s walls start. Cost: carried by the buyer, off-invoice. In a dispute: trusted by exactly one side. Use it when the buyer is deciding its own spend, not billing a counterparty. Buyer-counted the buyer brings its own counter WHO HOLDS THE PEN the buyer, or an agent it employs ACCESS partial — blind where theseller’s walls start COST carried by the buyer, off-invoice IN A DISPUTE trusted by exactly one side USE IT WHEN the buyer is deciding its own spend,not billing a counterparty Independently counted — a neutral holds the pen. Access: only what both sides grant it, agreed before the money moves. Cost: a line item someone has to carry. In a dispute: binding inside the settlement window. Use it when the count triggers a payment: a bonus, a make-good, the miss. Independently counted a neutral holds the pen WHO HOLDS THE PEN a named third party — accredited,or contractually neutral ACCESS only what both sides grant it,agreed before the money moves COST a line item someone has to carry IN A DISPUTE binding inside the settlement window USE IT WHEN the count triggers a payment:a bonus, a make-good, the miss Cost and dispute resistance rise together. The cheap count is the one the other side attacks first; the count that survives an argument is the one somebody paid to make neutral. COST RISES → → SO DOES DISPUTE RESISTANCE Accreditation says a method is sound. It does not say who is right about last Tuesday’s count.That takes a counting source named in the contract, and a window after which its read is final. The observer is a commercial decision, not a technical one.
Three observation regimes
RegimeWho holds the penAccessCostIn a disputeUse it when
Seller-countedThe seller, out of its own logs.Complete, and no audit from the other side.Cheapest — folded into the price.Disputed the moment money rides on it.Nothing is owed on the number: it steers spend, it never settles it.
Buyer-countedThe buyer, or an agent it employs.Partial — blind where the seller boundary starts.Carried by the buyer, off-invoice.Trusted by exactly one side.The buyer is deciding its own spend, not billing a counterparty.
Independently countedA named third party — accredited, or contractually neutral.Only what both sides grant it, agreed before money moves.A line item someone has to carry.Binding inside the settlement window.The count triggers a payment: a bonus, a make-good, the miss.

Accreditation and adjudication are different goods. Research says a method is sound; accreditation says a number is transactable. Neither says who is right about a particular period — that takes a counting source named in the contract, and a window after which its read is final. See research and measurement science for how that layer is built.

Clause three

The settlement window is a priced tradeoff, not an administrative detail.

The timeline runs from the event through observation, attribution close, the dispute window, a final count, and the release of money. Hold the window open and cash stays trapped, exposure keeps running, and settled counts reopen. Close it early and late-arriving truth lands outside the count, so the miss settles understated. Whoever writes the width is pricing it, whether or not they say so.

Money cannot move before the count is final. That is why the width of the window is a cash decision before it is a measurement one.

Plate 06 · The settlement window
The settlement window — open costs cash, early costs truth THE SETTLEMENT WINDOW Leaving the window open is not free. An unfinal count is an unreleased payment, a live exposure and a permanently arguable number — and a book that never closes never produces the loss history an underwriter would need to price the next one. COST OF HOLDING THE WINDOW OPEN Cash stays trapped escrow will not release Exposure keeps running the guarantor stays on risk Disputes reopen a settled count moves again The book never closes no loss history to price from Left open — the count never finalises. The span runs off the end of the timeline because there is no mark at which the argument stops. Money never moves. LEFT OPEN · no final count, ever The window you price — event to final count. Its width is the term being negotiated: wide enough to catch truth that arrives late, narrow enough that the money is not held hostage. That width is a price, and somebody pays it. THE WINDOW YOU PRICE · event to final count Event occurs — The outcome happens. Nothing is countable yet, and nothing can be settled. Event occurs Observation lands — The count arrives, late and partial. Observation is not the event; the gap between them is the first reason a window has to exist at all. Observation lands Attribution closes — No new credit is assigned after this mark. Everything that arrives later has to be handled as a restatement rather than as a count. Attribution closes Dispute opens — Either side may challenge the count. Until this window shuts, the number is provisional and the money behind it cannot be released. Dispute opens Dispute closes — The last restatement allowed. After this mark a corrected count has nowhere to land. Dispute closes Count is FINAL — The number stops moving. This is the mark the whole contract is built to reach: a count both sides have agreed to stop arguing about. Count is FINAL Money moves — Escrow releases and the miss is paid. Money cannot move before the count is final, which is why the window width is a cash decision. Money moves Closed early — the count is declared final the moment attribution closes, before the dispute window has run. The span is short, the cash moves fast, and everything that arrives afterwards is simply not in the number. CLOSED EARLY · final the moment attribution closes Late-arriving truth — a conversion, a correction, a restated count. The window you price is wide enough to admit it. The early-closed window is not, so the miss settles understated and the difference is borne by whoever wrote the shorter term. late-arriving truth — shut out by the early close Closing the window early is not free either. The cash moves, but it settles against a count that is missing whatever had not arrived yet — and the party holding the guarantee is the one that keeps the difference. COST OF CLOSING THE WINDOW EARLY Late truth excluded it lands after the count closed Long cycles cut off slow outcomes never get counted The miss is understated you settle on a partial count No home for restatement a corrected count arrives late The window is not an admin detail. Its width is a price, and somebody pays it.

Cost of holding it open

  • Cash stays trapped

    Escrow will not release.

  • Exposure keeps running

    The guarantor stays on risk.

  • Disputes reopen

    A settled count moves again.

  • The book never closes

    No loss history to price the next one from.

Cost of closing it early

  • Late truth excluded

    It lands after the count closed.

  • Long cycles cut off

    Slow outcomes never get counted at all.

  • The miss is understated

    You settle against a partial count.

  • No home for a restatement

    A corrected count arrives with nowhere to land.

Clause three, continued

The correction that arrives after the book is closed.

Measurement error is the one exposure no pause button removes, and a late restatement is not a crisis if the contract already says what happens to it. The correction routes through two pre-agreed gates — materiality, then the dispute window — into either a restated count with a named remedy, or a logged non-event carried into the next baseline. Written in advance, the same correction is a procedure. Written after the miss, it is a renegotiation, and leverage decides the number.

Plate 07 · Restatement policy
THE RESTATEMENT POLICY Without a restatement clause the correction has nowhere to route, so the counterparties reopen the terms of the deal and leverage decides the number. RENEGOTIATION the whole deal reopens no clause written Settled — the count was final, the invoice went out, and both sides closed the book on the period. SETTLED the count wasfinal, invoiced A late correction — the counting source restates the number after the book is closed. Measurement error is the one exposure no pause button removes. LATE CORRECTION the counting sourcerestates the number Restated — the count is reissued and the remedy named in the contract runs. A restatement is an operation, not an argument. RESTATED the count isreissued policy runsyesyes Materiality — is the change big enough to reopen a closed period? The threshold is a contract term, agreed before anyone knows which way the error will run. MATERIAL? vs. the threshold no Under the threshold — the correction is recorded and the book stays closed. Reopening a settled period for a small correction costs more to administer than it returns. Logged, not reopened under the agreed threshold The dispute window — corrections inside the agreed window settle against this period. The window is what makes a closed book actually closed. IN THE WINDOW? T+30, as agreed no Outside the window — the book stays closed and the corrected count carries into the next period as the new baseline. Real, but not this invoice. The window has closed carried to the next baseline THE REMEDY, NAMED BEFORE THE MISS Credit — value returned against a future period rather than cash out the door. The oldest guaranteed media settle their misses this way. Credit value returned against a future period True-up — the invoice is reissued at the corrected count, and the difference moves in whichever direction the correction runs. True-up reissued at the corrected count Clawback — money already paid comes back. The remedy with the most teeth, and the one that most needs to be agreed before the miss. Clawback money already paid comes back Which remedy applies is decided in the contract, not in the argument that follows the correction. Written before the miss, a restatement is a procedure. Written after it, a renegotiation.

  • Materiality

    Is the change big enough to reopen a closed period? The threshold is a contract term, agreed before anyone knows which way the error will run.

    If not: Under the threshold, the correction is logged and the book stays closed. Reopening a settled period for a small correction costs more to administer than it returns.

  • The dispute window

    Did the correction arrive inside the agreed window? The window is what makes a closed book actually closed.

    If not: Outside the window, the corrected count carries into the next period as the new baseline. Real, but not this invoice.

The remedy menu, priced into the contract rather than argued after the miss

  • Credit

    Value returned against a future period rather than cash out the door. The oldest guaranteed media settle their misses this way.

  • True-up

    The invoice is reissued at the corrected count, and the difference moves in whichever direction the correction runs.

  • Clawback

    Money already paid comes back. The remedy with the most teeth, and the one that most needs to be agreed before the miss.

A remedy that moves inventory instead of money is a make-good. It is a reasonable commercial gesture, and it is not risk transfer.

Clause four · capital

Who carries the miss.

An outcome is not a number. It is a distribution: an expected result, variance around it, and a tail of misses the seller does not control. The tail needs a reserve, and a reserve is capital held by a legal person with an incentive not to lie about either. Three candidate carriers, tested against those three requirements, and each answer prices differently.

"Nobody" is not the absence of a price. It is an unpriced tail, sitting quietly with whoever bought the promise.

Plate 08 · Reserve and capital
WHO CARRIES THE MISS ONE PROMISE, DISTRIBUTED WHO HOLDS THE RESERVE The distribution behind one promise. An outcome is not a number, it is a spread: an expected result, variance around it, and a tail of misses the seller does not control. likelihood delivered outcome The expected outcome — the number that gets written into the promise, and the only part of the distribution anyone quotes. expected outcome Variance — the spread the promise never mentions. Everything between the ad and the result moves it: price, product, the sales team, a competitor, a platform policy change. variance The miss and its reserve. The shaded tail is the outcome that does not arrive; the reserve is capital held against it. A guarantee with no reserve behind it is a slogan, not an instrument. the miss RESERVE capital held against the tail The seller — warrants its own promise. Capital only at its own scale, a legal person to sue, and no incentive to grade its own miss honestly. Price: a premium the seller sets on a count it also operates. The seller warrants its own promise SELF-PRICED capital legal person incentive not to lie a premium the seller sets on a count it also operates. A third-party underwriter — carries the tail for a fee. The only carrier that supplies all three at once — reserves, a balance sheet to hold them, and no stake in how the count lands. Price: media, plus a premium, on a count it does not operate. A third-party underwriter carries the tail for a fee STATED PREMIUM capital legal person incentive not to lie media, plus a premium, on a count it does not operate. Nobody — the current state. No reserve, no counterparty, no premium line. The tail did not go away; it went unnamed, and it sits with whoever bought the promise. Price: looks cheapest — the tail is unpriced, so the buyer keeps it. Nobody the current state UNPRICED capital legal person incentive not to lie looks cheapest — the tail is unpriced, so the buyer keeps it. A guarantee with no reserve behind it is a slogan. Software knows; capital carries.
Candidate carriers of the miss
CarrierWhat it isCapitalLegal personIncentive not to lieHow it prices
The sellerWarrants its own promise.Part — only at its own scaleYesNoSelf-priced: a premium the seller sets on a count it also operates.
A third-party underwriterCarries the tail for a fee.YesYesYesStated premium: media, plus a premium, on a count it does not operate.
NobodyThe current state.NoNoNoUnpriced: looks cheapest, because the tail went unnamed and sits with whoever bought the promise.
Eligibility

What can actually be underwritten.

An underwriter does not open with what you would like to guarantee. It opens with which promises are eligible at all. Every credible version of this market screens hard, and the screens are the same ones any book of correlated, steerable risk has always needed.

Show this sectionHide this section

  1. Enough events

    A book with too few outcomes cannot lean on the law of large numbers. Below that threshold the variance is not a risk around the deal — it is the deal.

  2. Fast enough

    Outcomes that arrive long after the window closes settle against a partial count. Latency between the exposure and the result is an underwriting parameter, not a reporting detail.

  3. Small enough

    A cap on the value of any single claimed outcome is what stops one deal from becoming the whole book.

  4. Not steerable by the claimant

    An outcome the party being paid can move is the one a market refuses to price. Which is why credible versions settle against screened cohorts and holdout-based triggers rather than an index anyone can trade.

  5. Not purely correlated

    A platform-policy change or a downturn hits every seller at once. Settling relative to a cohort stops a common shock from being priced as one seller’s failure.

  6. Evidenced at the point of sale

    This works where a purchase happens and can be evidenced. Long-cycle brand outcomes stay exactly as uninsured as they are now, and no escrow account changes that.

The vehicle exists one industry over: specialty insurers already write warranties against defined algorithmic performance failures, underwriting only after a technical review, with their own balance sheets carrying the shortfall. Nobody has pointed it at media outcomes — and media outcomes are steerable and correlated in ways that would have to pass the same test.

If the outcome cannot be screened, capped and collateralised, it cannot be underwritten. It can only be promised.

Demand

The first premium is probably paid by the sell side.

The assumed buyer of outcome protection has been the advertiser for twenty years. But the advertiser already holds the cheapest hedge in media: it can pause a campaign mid-flight, at close to zero cost, the moment the numbers disappoint. The protection is built into the product. It is imperfect — sunk creative, launch windows and seasonal inventory all leak through it — and it is good enough to cut willingness to pay a real premium for cover on something you can simply stop buying. Worse, the buyers who would pay are disproportionately the ones who already know the campaign is in trouble.

Show this sectionHide this section

  • The receivable

    Risk typeCredit risk

    Who is exposedThe seller who is paid on the conversion.

    Why the pause button does not helpThe work was done and the action happened, and payment still depends on a counterparty balance the seller cannot see. No pause button removes that.

  • The spread

    Risk typePerformance risk

    Who is exposedThe intermediary that buys exposure and sells results.

    Why the pause button does not helpIt carries the gap between what it bought and what it sold. The gap is priceable only while the inputs behind the prediction hold — and those inputs are usually rented.

  • The restatement

    Risk typeMeasurement error

    Who is exposedWhoever relied on the count.

    Why the pause button does not helpWhen the count itself is wrong, the damage is real money moving the wrong way, and it shades into professional liability rather than performance risk.

These are three different risks — credit, performance, and measurement error shading into professional liability — which is why the fourth clause will not arrive as one policy. It gets assembled from narrower protections around payment, counting, and residual performance risk.

Sold honestly, the near-term package is counterparty hygiene rather than outcome insurance: money escrowed, counts collateralised, miscounts with a written remedy.

The failure modes

How outcome deals break.

They do not break for mysterious reasons. Each of the four ways they fail is the visible symptom of one missing clause, and the trace runs backwards cleanly every time.

Three of these are measurement work and can be fixed with a contract term. The fourth cannot be drafted around, because it is the one that costs money.

Plate 09 · Failure modes
Four failure modes, each traced back to the clause that would have held it OUTCOME UNDERWRITING · FAILURE MODES Four ways it breaks. Each one is a clause that was never written. HOW THE DEAL BREAKS THE CLAUSE THAT WOULD HAVE HELD IT STATE Definition drift — the deal named an outcome that each side read its own way, so the argument arrives disguised as a data discrepancy. The definition clause is what would have held it: the counted event, written down and agreed before a dollar moves. Definitions today are still bilateral rather than standard, so the drift is re-litigated deal by deal. Definition drift both sides counted different things traces to 1 Definition what counts as the outcome, fixed before the spend BILATERAL Observation capture — the number came from a party with a stake in the number. A seller-rendered dashboard or a buyer-tuned model is not verification; it is an argument with extra steps. The measurement clause is what would have held it: a named counter whose read binds both sides, independent enough that neither has to take the other on trust. Observation capture the counter had an interest in the answer traces to 2 Measurement who counts it, with what access and what standing CONTESTED Window abuse — the count stayed open, so restated numbers kept arriving and the money never released. Hold the window open and exposure keeps running; close it early and late truth lands outside the count. The settlement clause is what would have held it: a stated width, after which the number stops moving and the file is shut. Window abuse the count never closed traces to 3 Settlement when the count goes final and disputes end THIN No carrier — nobody had agreed in advance to absorb the shortfall, so it settled on whichever party had least leverage when the invoice came due. The underwriting clause is what would have held it: a named party, paid a premium, holding reserves against the miss. This is the clause with no institution behind it — the first three are measurement work, and this one is capital. No carrier the miss landed on whoever blinked traces to 4 Underwriting who is out of pocket when the outcome does not arrive UNBUILT Clauses one to three are measurement work. The fourth is capital, and a guarantee with no reserves behind it is a slogan. We can promise the outcome. We cannot price the miss.
Show this sectionHide this section

  • Definition drift

    Both sides counted different things, so the argument arrives disguised as a data discrepancy. The definition clause is what would have held it: the counted event, written down and agreed before a dollar moved.

  • Observation capture

    The number came from a party with a stake in the number. A seller-rendered dashboard or a buyer-tuned model is not verification; it is an argument with extra steps. The measurement clause is what would have held it.

  • Window abuse

    The count stayed open, so restated numbers kept arriving and the money never released. The settlement clause is what would have held it: a stated width, after which the number stops moving.

  • No carrier

    Nobody agreed in advance to absorb the shortfall, so it settled on whichever party had least leverage when the invoice came due. The underwriting clause is what would have held it — and it is the one with no institution behind it.

The artifact

The outcome term sheet.

Fourteen terms that convert a performance promise into something a finance team can carry. None of them requires an industry-wide standard, a regulator, or a new protocol. They require two counterparties willing to write down what they already argue about afterwards.

outcome-term-sheet 14 terms
01 · Counted event
The outcome named as an event rather than an aspiration: what counts, written before the spend.
02 · Named counting source
Who holds the pen, named in the contract rather than inherited from whoever renders the dashboard.
03 · Access grant
What each side lets the counter see, agreed before money moves rather than after a dispute opens.
04 · Holdout design
The counterfactual promoted from analytics practice to a pre-registered term: what is withheld, for how long, read out against a design both sides accepted in advance.
05 · Attribution close
The mark after which no new credit is assigned. Everything later is handled as a restatement.
06 · Dispute window
How long either side may challenge the count, and the date on which challenging ends.
07 · Final count
The mark at which the number stops moving — the mark the whole contract is built to reach.
08 · Materiality threshold
How large a correction has to be before a closed period is reopened at all.
09 · Remedy menu
Credit, true-up or clawback, named in advance rather than negotiated after the miss.
10 · Escrow
The buyer funds the account. The seller receivable stops depending on a balance sheet it cannot inspect.
11 · Count bond
Collateral the dispute clause can seize, so a counting failure pays cash instead of starting an argument.
12 · Carrier and premium
Who holds reserves against the miss, what they charge for holding them, and where that appears on the invoice.
13 · Screens and caps
Which deals are eligible, and the ceiling on what a single claim can be worth.
14 · Loss record
Forecast against delivered, recorded on every deal, in a form the next premium can be priced from.

Every papered deal also produces the first input to a loss history — standardised evidence of what was forecast, what arrived, and what the gap cost. Without that record, nobody can price the next premium at all.

Deliverables

What ships.

The work is not a workshop. The work is an underwriting system a commercial team, a finance team and a general counsel can run together.

  1. 01
    Outcome inventory and definition audit

    Every outcome currently promised, taken down to what would actually have to be counted.

    Used byExec / Commercial / Legal

  2. 02
    Outcome definition ladder

    The four-rung climb from aspiration to counted event, applied to your own promises.

    Used byProduct / Measurement

  3. 03
    Observation regime design

    Who holds the pen, with what access, and what each option costs in dispute resistance.

    Used byMeasurement / Commercial

  4. 04
    Holdout-as-a-clause template

    The counterfactual written as a pre-registered contract term instead of a post-hoc argument.

    Used byMeasurement / Legal

  5. 05
    Settlement and restatement policy

    Window width, materiality threshold, dispute close, and the remedy that runs when the count is restated.

    Used byFinance / Legal

Show all 10
  1. 06
    Carrier and reserve model

    Who carries the miss, screened against capital, a legal person, and an incentive not to grade its own homework.

    Used byCFO / Exec

  2. 07
    Underwriting screens

    Which outcomes are eligible, which are refused, and the reason in each direction.

    Used byCommercial / Risk

  3. 08
    Outcome term sheet

    The fourteen terms that turn a performance promise into an instrument.

    Used byCommercial / Legal

  4. 09
    Loss-record schema

    Forecast against delivered, captured in a form a premium can later be priced from.

    Used byFinance / Measurement

  5. 10
    Pilot deal package

    One counterparty, money escrowed, the count bonded, the remedy written rather than argued.

    Used byCRO / Legal

Each artifact is part of the broader Artifact Library — the canvases, taxonomies, and operating models this playbook produces.

The motion

The miss gets priced last.

Six phases, in order, along one track. A carrier cannot reserve against a number that is still being argued over, so the measurement clauses close first — what counts, who counts it, when the count is final — and only then can the carrier and reserve model be built, a pilot deal papered, and the thing rolled out.

  1. Phase 1

    Outcome inventoryMeasurement work

    • What counts, in writing
    • One list, both sides
    • Clause one closed before anyone argues about who pays
  2. Phase 2

    Observation regimeMeasurement work

    • Named counting source
    • Access agreed before money moves
    • Holdout written as a term
  3. Phase 3

    Settlement and restatementMeasurement work

    • When the count is final
    • Materiality threshold and dispute close
    • The remedy if it is restated
  4. Phase 4

    Carrier and reserve modelCapital

    • Who pays the miss
    • Screens, reserves, premium
    • Cannot be priced until phases one to three close
  5. Phase 5

    Pilot deal paperedIn market

    • One counterparty
    • Escrow and count bond
    • Remedy written, not negotiated after the miss
  6. Phase 6

    RolloutIn market

    • Forecast against delivered, every deal
    • Loss history accrues
    • Premium appears as a line item

Phases one to three are measurement work. Phase four is capital. Running them out of order produces a guarantee nobody can price and nobody will honour.

Show the sequence as a diagram
Plate 10 · The sequence
OUTCOME UNDERWRITING SEQUENCE MEASUREMENT WORK CAPITAL IN MARKET Outcome inventorydefinition audit Observation regimedesign Settlement andrestatement policy Carrier andreserve model Pilot dealpapered Rollout Phase 1 — Outcome inventory and definition audit. Clause one: write down what counts as the outcome, in language both sides sign, before anyone argues about who pays for the miss. 1 Phase 2 — Observation regime design. Clause two: name the counting source as a contract term and promote the holdout from an analytics practice to a pre-registered clause. 2 Phase 3 — Settlement and restatement policy. Clause three: when the count goes final, how disputes close, and what the remedy is when the counter restates. 3 Phase 4 — Carrier and reserve model. Clause four, the capital one: who holds reserves against outcomes that do not arrive, which deals are screened out, and what the premium is. It cannot be priced until the first three phases close. 4 Phase 5 — Pilot deal papered. One deal, one counterparty: money escrowed, the count collateralised by a bond, the remedy written rather than negotiated after the miss. 5 Phase 6 — Rollout. Every papered deal records forecast against delivered, and that loss history is what lets the premium be repriced. The tell that the clause is real is a premium on the invoice, not a press release. 6 what counts, in writingone list, both sides named counting sourceholdout written as a term when the count is finalremedy if it is restated who pays the missscreens, reserves, premium escrow and count bondone counterparty loss history accruespremium as a line item The miss is priced last, because it cannot be priced until the count is settled.
The open question

How you would know this clause is being built.

This playbook rests on a reading of the public record, and the reading should be stated precisely rather than loudly.

Show this sectionHide this section

Stated precisely The platforms that come closest to selling an outcome optimise toward a target their own documentation declines to guarantee. They reprice; they do not indemnify. The broader claim — that nobody currently holds reserves against advertising outcomes — is a reading, not an audited fact, and it is offered here as a falsification test with a date and a tell rather than as settled market description.

  • The premium as a line item

    The day a contract between two independent companies carries a priced premium for warranting a result the seller does not control, the fourth clause exists. Watch for the premium, not the press release.

  • A named counting source, settled from escrow, by end of 2027

    An insertion order that names its counter as a contract term and settles from escrow, with a cash remedy when the count fails, is clause two and clause three arriving together.

  • The same inventory quoted twice

    One price with an independently attested outcome, one price without, and buyers paying the spread. That spread is evidence being priced rather than asserted.

  • Which side buys first

    If protection scales on the sell side first — a receivable warranty, a quota-share on an intermediary spread — the reading here holds. If an advertiser-facing product scales first, the pause-button argument is wrong.

  • The shape it arrives in

    Screened, capped and collateralised, or not at all. An unconditional guarantee of a business outcome across a whole book would be the genuine surprise.

  • The negative test, dated 2028

    If, by 2028, outcome-priced deals between independent companies stay a rounding error while impression-settled contracts carry on, with no underwriting premium anywhere as a line item, then the demand was never there and the problem was never plumbing.

Watch for the premium, not the press release. A line item is the only evidence that somebody actually put capital behind a result they do not control.

Straight answers

FAQ.

What is outcome underwriting?

It is the operating work of making an outcome tradable. Four clauses have to hold at once: a definition both sides accept, a measurement both sides trust, a settlement window in which the count goes final, and someone who pays when the outcome does not happen. The first three are measurement work. The fourth is capital.

How is this different from outcome-based pricing?

Outcome-based pricing selects a billing unit and stops. Underwriting decides what happens when the unit does not arrive: who observes it, when the count closes, what the remedy is, and which party is out of pocket. Selecting the model is the Commercial Productization job. This playbook starts where that one stops.

Is a target CPA or a cost cap the same as a guarantee?

No. Optimising toward a target is effort — the system observes results and reprices toward your goal, and the risk stays with you. A guarantee is answerability: a neutral count, a date it becomes final, and a party that pays the difference. The platforms that come closest to selling an outcome optimise toward a target their own documentation declines to guarantee.

Who actually buys outcome protection first?

Probably not the advertiser, because the advertiser already holds the cheapest hedge in media — it can stop buying mid-flight at close to zero cost. The exposures that cannot be paused sit on the sell side: the receivable, the intermediary spread, and the restatement of a count somebody relied on.

What can and cannot be underwritten?

Outcomes that are numerous enough, fast enough, small enough, not steerable by the party being paid, not purely correlated across the market, and evidenced at the point of sale. Long-cycle brand outcomes fail most of those screens, and no escrow account changes that.

How wide should the settlement window be?

Wide enough to admit truth that arrives late, narrow enough that the money is not held hostage. Hold it open and cash stays trapped while exposure keeps running; close it early and the miss settles against a partial count. The width is a price, and whoever writes it is setting that price whether or not they say so.

Do we need a third-party verifier?

It depends on what the count does. If the number only steers spend, a seller-rendered count is the cheapest read and it is enough. If the number triggers a payment — a bonus, a make-good, the miss — only an independent count binds inside a settlement window. Accreditation says a method is sound; it does not say who is right about a particular period’s count.

What is a restatement policy?

The clause that routes a correction arriving after the book has closed. It tests the correction against an agreed materiality threshold and an agreed dispute window, then either restates the count and runs a pre-named remedy, or logs it and carries it into the next baseline. Written in advance it is a procedure. Written after the miss it is a renegotiation, and leverage decides the number.

Does anyone underwrite advertising outcomes today?

The honest answer is that the clause has no visible institution behind it. The platforms that come closest optimise toward a target their own documentation declines to guarantee, and specialty insurance already writes warranties on defined algorithmic performance one industry over without having been pointed at media. Whether anyone currently holds reserves against advertising outcomes is a reading of the public record, and this playbook stakes it as a falsification test rather than asserting it as an audited fact.

What does the engagement actually produce?

An outcome inventory and definition audit, an observation regime, a settlement and restatement policy, a carrier and reserve model with its screens, a term sheet carrying all of it, a loss-record schema, and one pilot deal papered with escrow, a count bond, and a written remedy.

Where to go next

Where this connects.

Market references last validated: September 12, 2026. Revalidate before pitch use.

On the operator’s shelf285 books
Cover of Skin in the Game by Nicholas Taleb
Cover of Incerto by Nassim Nicholas Taleb
Cover of Thinking, Fast and Slow by Daniel Kahneman
  • Skin in the Game · Nicholas Taleb
  • Incerto · Nassim Nicholas Taleb
  • Thinking, Fast and Slow · Daniel Kahneman
Next step

Ready to turn a performance promise into something you can price?

The playbook defines the outcome down to a counted event, designs the observation regime, writes the settlement and restatement policy, builds the carrier and reserve model, and papers the first deal with escrow, a count bond, and a written remedy.

Scope this playbook → See the term sheet See related operating playbooks