Pay for Performance Only Works If You Agree What Counts

Month four of a performance deal. The agency’s dashboard reports one conversion number. GA4 reports a smaller one. The client’s CRM reports a smaller one still, because a third of the leads were never called. The contract says the bonus is paid on conversions. Nobody wrote down which system counts. Both sides are now convinced the other is being unreasonable, and both are reading their own screen correctly.

That is where outcome-based pricing actually breaks. Not at the percentage, which gets negotiated for weeks, but at the definition of the thing being counted, which most contracts leave to a single undefined noun. What the term means and how to interview an agency claiming it is a separate question, covered in what performance-based actually means and how to test for it. This is about the deal you sign afterwards.

Three systems, three numbers, all of them correct

Meta’s default attribution credits conversions that happen within seven days of a click or one day of a view. Google Ads counts conversions on its own surfaces using its own model. GA4 applies a different model with different session rules, and routinely files a Meta view-through conversion under direct or organic instead. Your CRM knows only what the person answering the phone typed into a field.

Four systems, four answers, none of them lying. A contract that pays on conversions has named none of them, which means the number that determines the fee is whichever one the person writing the invoice opens first.

Three lines fix most of it. Name the system of record, and make it the one closest to money you have actually collected. Name the attribution window and the lookback, in days, matched to your real sales cycle rather than the platform default. State the deduplication rule for when two platforms both claim the same sale, and state what happens when the system of record breaks, because tracking outages are not hypothetical and somebody has to be paid during one.

Then set the baseline. Outcome pricing without an agreed starting point pays an agency for demand you already had. Fix the pre-engagement period, the seasonal comparison, and the data source for both, in writing, before the first campaign launches. It is a fifteen-minute conversation in month zero and an unwinnable argument in month six.

The bottom of the funnel is the model’s built-in exploit

Tie a fee to cost per acquisition or return on ad spend and you have told the agency exactly where the profit is: harvesting demand that already exists. Bid on your own brand terms. Retarget people who were already in the funnel. Push budget toward the audiences closest to purchase. Blended numbers improve within weeks, the gate gets hit, and new-customer acquisition quietly stops.

Almost none of this is fraud. It is the incentive working as written. The same dynamic is what makes a healthy-looking blended figure so unreliable, which is the argument in why branded search flatters a blended ROAS number.

The contractual answer is to split branded from non-branded reporting from day one and to set the gate on new customers rather than on blended efficiency. If the agency objects that they cannot control how many people search your brand name, they are right, and that is the point: the metric they are paid on should be the one they can actually move.

Write the incrementality test before anyone needs the answer

The only way to settle whether paid media created demand or intercepted it is to turn some of it off. Holdout regions, staggered pauses, matched markets. The method matters less than the timing of the agreement, because nobody consents to a holdout test while their bonus depends on the result.

So agree it at signature: which method, who runs it, in which quarter, over how many weeks, and how the outcome adjusts the fee. Add a standing rule that campaigns which failed get reported alongside the ones that worked. An agency that only presents winners has made your fee structure decorative, whatever the contract says.

Percentage of spend is the model everyone attacks, and it is still the least bad one

The objection is obvious and worth taking seriously. Paying an agency a share of media spend pays them to spend more, and the incentive points away from efficiency at exactly the moment efficiency matters.

What contains it is structural rather than moral. The client owns the ad accounts and pays the platforms directly, so no agency can move budget on its own initiative: every increase is a decision you approve and a charge that appears on your own card. A base retainer covers the actual labor, so the percentage is not the agency’s survival. A ceiling above which the percentage steps down removes the pay rise from scaling. And the performance gates sit on acquisition quality, not on volume, so spending more without results costs the agency money rather than earning it.

Ours is a monthly retainer plus 15 percent of managed spend, with budgets running from the client’s own advertising accounts and the agency fee billed as a separate line item. It is defensible for a plain reason: it is the model easiest for you to audit, because every input sits on invoices you receive directly from Google and Meta rather than inside a report we produce.

Compare the failure modes rather than the promises. A flat retainer makes the agency indifferent to results until renewal season. A pure share of revenue or a pure cost-per-acquisition deal makes the agency an underwriter of your conversion rate, which they will price for by chasing whatever converts fastest. Milestone structures turn every month into an argument about whether the gate was met. There is no model without a failure mode. The question is which one you can supervise from your side of the table.

The clauses that decide whether this survives month six

The commercial terms most disputes turn on are rarely the headline fee:

• System of record, attribution window, and deduplication rule, named explicitly
• An agreed baseline period and seasonal comparison
• Who can change budget, and by how much, without written approval
• A quality floor: what counts as a qualified lead, defined by your sales team, not by a form submission
• A ramp period before gates apply, derived from your own median time to close
• A spend ceiling with a stepped percentage above it
• What happens when the client side breaks: site outage, tracking removed by a developer, leads not contacted within the agreed window
• Ownership of accounts, pixels, audiences, creative files and historical data on exit
• Term length and notice, with renewal tied to the same metrics the bonus is

The clause people skip is the one about their own obligations. Outcome pricing assumes the agency is the only variable. In practice the traffic is often fine and the loss happens after the click, in a form that asks for eleven fields or a callback that arrives two days late. If that is where your money is going, the path between a form submission and a first real conversation is the fix, and no fee structure substitutes for it.

When outcome pricing is the wrong instrument

Some businesses should pay a straight retainer and stop negotiating. If your sales cycle is longer than your contract, the outcome lands after the term ends, and you will be arguing about who owns pipeline created in month two that closes in month fourteen. Either pay on a mid-funnel gate both sides genuinely trust, or accept a retainer and judge it annually.

New products in new markets have no baseline, which means no honest way to say what the agency changed. Budgets too small to generate signal produce noise that neither side can attribute, a threshold worth reading about before you propose gates. And in regulated categories the outcome is partly a compliance artifact rather than a marketing result: for investment advisers, what you are permitted to show about performance is set by rule, and those rules move, as they did when the SEC relaxed its net performance guidance. Writing a bonus against numbers a regulator defines is a different exercise from writing one against sales.

The through-line is unromantic. Outcome pricing is not an alignment philosophy, it is a measurement contract, and it works exactly as well as the definitions inside it. Get the definitions right and almost any fee shape holds up. Get them wrong and the most elegant structure in the market turns into a quarterly argument neither side can win. If you want the arrangement pressure-tested before you sign, that is the conversation to have while setting up the media program, not after the first invoice.

Frequently Asked Questions

Is charging a percentage of ad spend a conflict of interest?

It is a real tension, and the honest response is structural rather than reassuring. Ask three questions: who holds the advertising accounts and receives the platform invoices, whether budget increases require written client approval, and whether the percentage steps down above an agreed ceiling. If the client owns the accounts and controls the budget, the agency cannot act on the incentive unilaterally. If the agency holds the accounts and reports spend back to you, the conflict is live regardless of what the contract says about alignment.

What percentage should an agency charge on managed spend?

The number matters less than what it covers and where it stops. Ours is 15 percent of managed spend on top of a retainer, with the retainer covering strategy, creative direction and account management, and the percentage covering the operational load that scales with budget. What you should test in any proposal is whether the percentage is charged on top of platform fees or inside them, whether it applies to every channel or only paid media, and at what budget level it steps down.

Should the bonus be a share of revenue instead?

Only if revenue is cleanly attributable and both sides accept the same attribution model, which in practice means ecommerce with a short cycle and solid tracking. For considered purchases and lead generation, revenue arrives too late and passes through too many hands to serve as a fee trigger. A defined qualified-lead gate, agreed with your sales team, is usually a better instrument even though it feels further from the money.

What happens if the agency hits its KPI but revenue does not move?

That outcome means the KPI was the wrong proxy, and the contract should say what happens next rather than leaving it to goodwill. Two clauses handle it: a review trigger when the agreed metric and revenue diverge for two consecutive periods, and a right for either side to renegotiate gates once during the term without penalty. Agencies confident in their measurement rarely resist this, because divergence usually points at a tracking problem they would rather find early.

How long should the ramp be before performance gates apply?

Derive it from your own data rather than accepting a default. Pull the median time from first touch to closed deal over the last year, add the platform learning period, and set the gate to begin after that. For fast ecommerce this may be weeks. For considered B2B purchases it can be a full quarter or more, and starting gates before the first cohort has had time to convert guarantees a fight over numbers that were always going to look bad.