The agency incentive problem nobody fixes

Studios have spent fifteen years trying to design an agency fee that makes the agency want what the studio wants, and every model on the menu fails in a predictable direction. That is not because agencies are dishonest. It is because the studio and the agency are optimising different objective functions, and no fee structure can make two different objectives into one. The useful response is to stop searching for the aligned contract and start designing the relationship around the misalignment you have chosen.

The question has become more pressing this spring. Liftoff's flotation this month and AppsFlyer's reported fundraising conversations are reminders that the vendor layer is consolidating and capitalising, and a lot of mid-size studios that used to buy through networks directly are now routing more of their budget through agencies who manage automated platforms on their behalf. The agency's incentives now sit between the studio and a black box.

Three fee models, three predictable failures

Percentage of spend is the oldest model and the easiest to criticise. The agency earns more when the studio spends more, so the agency has a structural reason to argue for scale and against pauses. Most practitioners know this. What fewer notice is the subtler effect: percentage of spend punishes the agency for finding efficiency. A buyer who cuts wasted spend by a fifth has just cut their own firm's revenue by a fifth, and their account lead knows it.

Fixed retainers remove the volume incentive and introduce a different one. The agency's margin now depends on hours spent, so the rational move is to do the minimum that keeps the account. Retainer accounts drift toward maintenance: the same channels, the same creative rotation, the same weekly report. Innovation happens when the client threatens to leave, not before.

Performance fees, usually a bonus tied to ROAS or cost per payer targets, sound like the answer and fail in the most interesting way. The agency is now paid to hit a number the agency partly controls the measurement of. Attribution window choices, view-through settings, the decision about which cohorts to exclude as anomalous: each is a legitimate judgement call, and each one nudges the reported number. Nobody has to cheat. The number simply gets defined in the agency's favour one small decision at a time.

The underlying mechanism

Step back from the fee models and the structure is clear. The studio wants the maximum profit from its game across its whole lifetime, which includes the option to spend nothing this quarter. The agency wants the maximum profit from this account across the account's expected lifetime, which is shorter than the game's and ends the day the studio spends nothing. No fee model closes that gap because the gap is not in the fee. It is in the time horizon and in the fact that the agency's best outcome includes the account existing.

There is a second, quieter mechanism. Agencies serve many clients, and the knowledge they gain on your account has value on other accounts. That is legitimately why you hire them. But it means the agency's incentive to run a genuinely novel test on your budget is lower than yours, because a novel test that works becomes an insight they can sell to everyone, while a novel test that fails is a conversation with you.

What the usual fixes do

Studios respond in three ways, and each carries a second-order cost that rarely appears in the procurement deck.

Bringing buying in house removes the fee gap and replaces it with a hiring problem and a knowledge problem. The internal team's incentives are aligned, but a team of three seeing one game's data cannot match an agency's cross-account pattern recognition, and it will take a year to notice what it does not know.

Splitting the account across two agencies creates competition and destroys learning. Each agency now has a reason to hoard insights and to make the other look worse in the shared review, and the studio pays a coordination tax nobody itemised.

Hybrid fees, a small retainer plus a performance bonus plus a spend cap, try to cancel each incentive with another. In practice they produce an agency that cannot predict its own revenue and therefore staffs the account conservatively. Complexity in a fee structure is paid for in the seniority of the people working your account.

Design around the misalignment instead

The workable approach accepts one misalignment on purpose and fences it. A decision rule: pick the fee model whose failure mode is cheapest for you to detect, then build detection into the contract rather than trying to remove the incentive.

If you choose percentage of spend, the failure mode is over-spend, which is the easiest of the three to detect. Fence it with a studio-owned marginal ROAS threshold that the agency does not set and cannot argue against in the weekly review, and with a clause that pauses are never a breach.

If you choose a retainer, the failure mode is stagnation, which is detectable if you measure it. Require a fixed share of budget in tests the agency has never run for you before, and audit the list quarterly. Pay a modest bonus for a documented failed test, because a retainer agency's rational move is otherwise never to fail.

If you choose performance fees, the failure mode is measurement drift, which is the hardest to detect and therefore the one to avoid unless you own the measurement. Take attribution settings, window definitions and cohort exclusions out of the agency's hands entirely and put them in a document that only the studio can change. Better still, tie the bonus to a metric the agency cannot influence the definition of, such as finance's reported net revenue against finance's reported spend, with a lag long enough that nobody can time it.

The illustration studios skip

Consider an illustrative mid-size studio spending a material monthly sum through an agency on a percentage model. It moves to a performance model to fix over-spend, and reported ROAS improves within a quarter. Finance's net revenue does not move. Nobody did anything wrong. The attribution window widened from seven days to fourteen, a legitimate choice the agency made when the platform offered it, and the reported number rose with it.

The lesson is not that the agency gamed the studio. It is that the studio moved from a misalignment it could see to one it could not, and called that progress. The contract that works is the one whose failures you will notice.

Related archive reading

These articles provide related context and remain subject to their stated review status.

• First-party data is overrated for most studios
• The creative strategist role is misdefined