How to Audit an Agency’s Performance Claims Before You Sign

A slide in an agency pitch reads: 4.2x ROAS, 63 percent lower cost per acquisition. It has a client logo on it and a nice chart. Three questions decide whether it means anything. Over what period. Measured in which system. Compared to what.

Most performance claims do not survive all three, and the reason is rarely dishonesty. It is that the number was assembled for a deck by someone who was not in the account, using whichever export made the strongest case. What follows is the diligence that separates a claim you can rely on from one that is technically true and commercially useless. The commercial terms come afterwards, and they have their own failure modes, covered in how to structure a performance deal that survives contact with attribution.

Ask which system produced the number

Platform dashboards, analytics, the CRM and finance will each give a different answer for the same period, because each one counts differently and none of them is wrong on its own terms. A return on ad spend figure taken from Meta and a revenue figure taken from a bank statement are not the same claim, and the gap between them is often the entire margin of the case study.

So ask for the source, then ask to see it. A screenshot of the live account with the date range, the attribution setting and the conversion columns visible takes about thirty seconds for anyone who actually ran the campaign. Client confidentiality covers the name and the spend, not the shape of the report, and figures can be redacted. If producing it takes a week and comes back as a rebuilt chart, the original number was made in a slide rather than in an account.

Ask one follow-up: what was the attribution window, and was it the platform default? A seven-day click window and a one-day click window describe the same campaign very differently, and the choice of window is usually where a mediocre result becomes a headline.

A percentage without a denominator is not a result

A doubling is a doubling whether the lead count went from four to eight or four hundred to eight hundred. One of those is a rounding error and the other is a business. Percentages travel well in decks precisely because they hide scale, and the same trick appears in charts drawn without a y-axis.

Ask for absolutes. Spend, conversions, cost per acquisition, and the period they cover. Then ask the harder question: what was the trend before the engagement started? Growth that began three months before an agency arrived belongs to whatever caused it, and any campaign launched into a seasonal peak will look like genius until the peak ends.

The fair comparison is the same period a year earlier, not the month before launch. An agency that has thought about its own results will already have it. An agency that has not will treat the request as hostile, which is itself an answer.

Ask how much of it was branded search

The fastest way to produce an excellent-looking blended result is to spend on people who were already coming: your own brand terms, retargeting pools, audiences deep in the funnel. Efficiency rises, the case study writes itself, and no new customers were acquired. That mechanism is worth understanding in detail, because it also flatters ongoing reporting, not just pitch decks: why branded search subsidises a blended ROAS number.

So the question for any case study is simple. What share of that result came from branded search and retargeting, and what did the non-branded prospecting numbers look like on their own? If the agency cannot split them, they either did not measure it or would rather not.

Ask for the client who left

Reference calls with three delighted current clients tell you what the agency is good at selecting for. The useful request is different: a client whose engagement ended in the last year, and permission to speak to them. Agencies with genuine accountability usually have one, and they will frame it honestly, because engagements end for reasons that are nobody’s fault.

Pair it with a question about failure. What did they do the last time a campaign missed its targets, and what changed as a result? The answers cluster into two types. One describes a diagnosis, a decision and a consequence. The other describes an algorithm update, a seasonal dip, or the client’s landing page. Landing pages genuinely do kill campaigns, but an agency that noticed only after the results came in was not watching, and if the path from click to first conversation was the problem, the time to say so was week two.

One category-specific note. If an agency shows you performance figures for a regulated client, particularly in financial services, ask whether the presentation complies with the rules that govern such claims. Those rules are specific about what can be shown and they change, as they did when the SEC revised its net performance guidance. An agency casually publishing a client’s returns without knowing the rule is telling you how it will handle yours.

Find out who would actually be in the account

The final claim to audit is not about results at all. Ask for the names and seniority of the people who would run your account day to day, how many other accounts each of them carries, and which parts of the work are subcontracted. Then ask what happens to that staffing if the agency signs a client three times your size next quarter.

Pitch teams and delivery teams are often different people, and the gap between them explains more disappointing engagements than any pricing model does. The strongest single request in an evaluation is to spend twenty minutes with the person who would be in your weekly call, without the founder in the room.

Once the claims hold up, the argument moves to fee structure, definitions and gates. That is a separate exercise and it deserves the same scrutiny, whether you are building a paid media program from scratch or replacing an incumbent.

Frequently Asked Questions

What should an agency be able to show me from a live account?

A screenshot or screen share of the account with the date range, attribution setting and conversion columns visible, with client identity and absolute spend redacted if needed. That is enough to verify the shape of a result without disclosing anything confidential. Rebuilt charts, exported summaries and slides do not serve the same purpose, because the number has already passed through someone’s judgement about what to include.

Are case studies with named clients more trustworthy?

A name confirms the relationship existed, not that the numbers are sound. Plenty of named case studies rest on platform-reported figures over a favourable window with no baseline. An anonymized case study backed by a live account view and a straight answer about branded share is worth more than a logo wall. Use names to verify the client is real, then audit the number the same way regardless.

What if the agency says the results are confidential?

That is often legitimate, and it does not have to end the conversation. Confidentiality covers the client and their commercial figures. It does not prevent an agency from describing methodology, showing a redacted account view, explaining what share of performance came from branded traffic, or having a former client speak to you with permission. An agency that treats every one of those as confidential is using the word as a shield.

Should I pay for an audit before hiring?

A paid audit is a reasonable way to see how an agency thinks before committing to a term, provided you own the output. Judge it on whether it identifies problems the agency cannot fix and would not profit from fixing. An audit that concludes you need exactly the services on their price list, with no mention of your tracking, your offer or your intake process, is a proposal wearing a different name.