Performance is a promise about accountability: the agency claims its work can be traced to results rather than judged on impressions. It is a good promise and it is only as strong as the definitions underneath it. Whoever defines a conversion, chooses the attribution model and controls the analytics property effectively decides whether the campaign looks successful. Buyers who skip those three questions end up debating a dashboard for a year. This page sets out how performance engagements are structured, where reports flatter themselves without anyone lying, what a fair fee model looks like, and the specific contract terms that keep the evidence checkable by you rather than only by the agency.
Define the result before you agree the fee
A conversion can be a form submission, a phone call over a length threshold, a booked appointment, a qualified opportunity or closed revenue, and the numbers differ by an order of magnitude between the first and the last. Agree which one the contract refers to, write the definition down, and exclude the obvious contaminants: spam submissions, existing customers using a contact form, recruitment enquiries, test submissions and duplicate entries from the same person. Then agree who counts them and from which system. If the agency's own platform is the sole source of the count, you have outsourced the scoreboard to the team being scored. Ask for the definition to be implemented in your own analytics property, which you own, and for the raw events to remain exportable. Google's developer documentation for its analytics platform sets out how events and conversions are collected, and any competent performance agency can walk you through exactly which events they will create and why. A candidate who cannot is proposing to report on a system they do not really understand.
Where attribution quietly flatters everyone
Attribution models decide which touchpoint gets credit, and reasonable models disagree substantially on the same data. Last click favours whichever channel sits closest to the purchase, which is usually branded search, and branded search would often have converted anyway. First click favours the top of the funnel. Data driven models are better but need volume to be stable. None of this is fraud, and it becomes a problem only when an agency reports on the model that favours the channel it runs and never mentions the alternative. Ask each candidate which model they report on, why, and what the picture looks like under a different one. Ask specifically how they separate branded search from everything else, because branded clicks bought in paid search are the single most common way a performance report gets inflated. A firm that volunteers this distinction before you raise it has been asked the question before and answered it honestly, which is a better signal than any case study.
Fee models and what each one rewards
A flat retainer is predictable and neutral, and it requires an agreed workload or it becomes vague. A percentage of media spend rewards spending more rather than spending better, which is workable with a cap or a step down at higher levels. A cost per lead or cost per acquisition model looks perfectly aligned and creates two pressures worth watching: an incentive to count marginal enquiries as leads, and an incentive to stop working once the monthly target is met. A hybrid of a modest retainer plus a performance element usually behaves best, provided the definition of the result is airtight, which brings you back to the first section. Whichever model you choose, buyers evaluating performance marketing services should ask what happens to the fee if spend pauses, what the notice period is, and whether the rate is reviewed as volumes grow. Get the answers in the agreement rather than in a proposal deck that will not survive a change of account manager.
The terms that protect the evidence
Four clauses matter more than the rest. Ownership of the advertising accounts, set up under your billing with the agency granted access, so history and spend records stay with you. Ownership of the analytics property and the right to export raw data at any time. Ownership of creative assets and the underlying licences for stock, music and talent, including term and territory. And a reporting clause that names the metrics, the definitions, the cadence and the source system, so a change of personnel does not quietly change the scoreboard. Add one more if you can: the right to have a third party review the accounts once a year. Reputable firms agree without discomfort, because they expect to be audited eventually and would rather it happen while things are going well. Resistance to any of these four is the most informative signal available in a pitch process, and it costs nothing to test before you sign.
Questions people ask about marketing performance agency
Is a cost per lead deal safer than a retainer?
Only if the definition of a lead is tight and you control the counting. Loose definitions turn a performance deal into a volume deal, and you pay for enquiries your sales team discards. Define quality, exclude contaminants, and agree who arbitrates disputes before the first invoice rather than during the first argument.
How long before a performance engagement can be judged?
For paid channels with meaningful volume, one quarter usually produces enough data to draw a conclusion, provided tracking was correct from week one. For anything with a long sales cycle, judge on pipeline created rather than closed revenue, and agree that measure in advance so nobody redefines success halfway through.
Should the agency hold our advertising accounts?
No. Set them up under your own billing and grant access. You keep spend history and learning, you can see actual costs against actual results at any time, and changing supplier becomes an access change rather than a rebuild. Reputable firms accept this readily.
What belongs in a monthly performance report?
Spend by channel, results using the agreed definition, cost per result, what changed during the month and why, what is planned next, and a link to the source data you can open yourself. If the document cannot be checked against a system you control, it is a summary rather than a report.