Marketing Agency Accountability: SLAs and Exit Terms

Picture a performance clause drafted in a single line: Agency will maintain a cost per acquisition under $120. No platform named. No conversion action named. No measurement window named. It reads like a commitment and works like decoration, because nobody on either side can compute that number the same way twice. Marketing agency accountability starts one step earlier than the target. It starts with the definition of the thing the target is attached to.

Why marketing agency accountability dies in the definitions

Google Ads documents the default plainly. If you do not customize the click-through conversion window when you create a conversion, the default window is 30 days. You can set the conversion window anywhere from 1 to 90 days, and if you set it to 7 days, any conversion that happens more than 7 days after the ad interaction will not be recorded.

Read that as a contract problem. Same campaign, same spend, same buyers: move that one setting from 30 days to 7 and the reported conversion count falls while the reported CPA rises. Nobody cheated. The clause simply never said which number it meant.

Write the measurement into the clause: platform and account, the conversion action by name, the attribution window in days, and the report the number gets read from. Say who is allowed to change any of it. Then add the line that earns its keep six months later, when someone edits a conversion action and the trend breaks. If the window or the event definition changes, the baseline is restated in writing and both sides sign off on the new one.

Fix one source of truth before you add KPIs

The instinct when an engagement feels unaccountable is to ask for more metrics. The evidence points somewhere else. Nielsen reports that 85% of marketers express confidence in their ability to measure ROI, but only 32% actually measure ROI holistically across traditional and digital media channels. Forrester puts it harder, finding that only 37% of B2B marketing decision-makers trust their measurement and analytics enough to inform decisions while four of the five most common criteria that organizations use to evaluate marketing rely on proof of engagement.

Those samples are enterprise marketers and B2B decision-makers, so read them as the direction the industry is facing rather than a benchmark for a founder-led business spending five figures a month. The structural lesson holds anyway. A longer KPI list does not buy trust. One number, computed one way, from one report both parties can open, has a chance of it. So name the source of truth in the agreement, whether that is the ad platform, the CRM or a warehouse table, and say which one wins when they disagree. They will disagree. If the CRM wins, the performance clause has to be written against CRM-stamped events, and the partner needs the access to stand behind it.

The CMO Survey found that rather than investing in deeper customer insights, most marketers focus on developing stronger performance tracking as the primary way to demonstrate value. Tracking nobody trusts is a bigger invoice line and little else. Reading the report you already get is covered in Marketing Agency Reporting: Read It Like an Operator, and the question of which numbers actually predict revenue in this breakdown.

Put the calibration period in writing

Week one is not evidence, and everyone involved knows it. The platform documentation is specific enough to borrow from. Google states that it can take a few conversion cycles (1-2 typically) for the bid strategy to calibrate to the new objective, and that its algorithms continue to learn even when the bidding status no longer shows 'Learning'.

That supports a two-stage SLA. During a calibration window with a written end date, the commitments are process: named deliverables, reporting cadence, a decision log, creative volume shipped. After that date, the outcome clause switches on against the metric you defined above.

Be honest about what the documentation covers. It describes platform behaviour after a change, not when to fire a partner. Any give-it-90-days rule is your judgment as the person signing the cheque, and the agreement should say so rather than borrowing authority from a help page. The month six fight is where that judgment usually gets made.

A worked example you can re-run on your own numbers

Assumptions first, because this is a model and not a result. One platform, one conversion action, $15,000 a month in spend, a $150 target CPA, and the Google default 30-day click-through window.

At target, $15,000 divided by $150 is 100 conversions for the month. Now add the window. Spend runs days 1 to 30, and a click on day 30 can still record a conversion as late as day 60, so the month-one cohort is not final until day 60. Pull the report on day 31 and you are grading a partial month against a clause written for a whole one.

Stack calibration on top of that. If the bid strategy needs one to two conversion cycles to settle after a change, and your conversion cycle runs close to the full 30 days, calibration can absorb days 1 to 60. The first cohort that is both post-calibration and fully closed spends across days 61 to 90 and finishes reporting near day 120. That is where the first genuinely testable outcome review belongs on the calendar.

Now re-run the same spend with a 7-day window. Conversions landing on day 12 after the click are never recorded, the counted conversions drop, and the reported CPA rises while the business performs identically. Then change the inputs to your own. A two-week sales cycle under a 90-day window produces a completely different review schedule, which is the reason to do this arithmetic instead of copying a quarterly cadence off a template.

Media quality belongs in the reporting SLA

Most reporting SLAs stop at spend, results and a paragraph of commentary. The ANA's programmatic benchmark suggests the money worth arguing about sits upstream of all three. It found that higher-performing advertisers converted 54.0 percent of programmatic spend into qualified impressions in Q1 2026, while the lower-performing cohort converted just 32.1 percent, and that the weaker cohort lost 38.4 percent of spend to media quality issues, against 19.0 percent for the higher-performing cohort. Where the gap came from matters more than its size: transaction costs differed by just 2.4 percentage points between cohorts, while media productivity losses differed by 19.4 percentage points.

That benchmark covers large advertisers buying programmatic, so it is no yardstick for a mid-market Meta and Google account. The clause it justifies travels anyway. Ask that the monthly report name what was lost before the auction did any work, with viewability, invalid traffic and made-for-advertising placements as their own lines rather than absorbed into a blended CPA. If a partner buying display or programmatic for you cannot produce those lines, you have learned something useful about which half of the spend is buying attention. The same logic applies to creative: platforms document what a clean test looks like, and we walked through that in the ad creative testing framework.

Exit terms: name the owner of every account on day one

Exit clauses usually get written about notice and money, which is the easy half. The enforceable half is asset control, and on the platforms it is mechanical. Google documents that a manager that creates a new account automatically becomes the owner of that account and that a client account can only have one owner. The counterweight sits on the same page: the client account still owns its data and has the ability to remove ownership access by unlinking.

So settle the ownership questions on day one rather than during a breakup. Who creates each ad account. Which manager account or business manager holds it. Who holds admin on the pixel, the tag container and the CRM. What the unlinking sequence looks like, written as steps with an owner and a date against each one.

Terms like these get tested when money is tight, and money is tight. The CMO Survey has marketing budgets down to 9.0% of revenues, with overall marketing spending up just 1.7% over the prior 12 months, and reports that when profits fall short of expectations, 53.1% of company executives focus on cutting expenses rather than investing in revenue growth. Write clauses you would be willing to exercise in a quarter where the number comes in soft, because that is the quarter they exist for.

Here is the hour-long version of all this. Print your current agreement, circle every number in it, and ask for each one whether a stranger with access to the named report could compute it and land on your figure. Where the answer is no, the repair is usually a single sentence naming the event, the window and the report. Then check that the exit section names an owner for every account instead of a notice period alone. If you are still choosing a partner, the vetting-question version of this sits in 15 questions worth asking, and if you would rather work through the terms against your own setup, start here.