The KPIs that keep your marketing agency honest

Why "the ROAS looks fine" is how accounts die

Most agency relationships don't explode; they drift. The monthly deck says ROAS is stable, everyone nods, and eighteen months later the founder discovers that new-customer acquisition has been shrinking for two quarters while remarketing and brand-term spend propped up the blended number. Nobody exactly lied. The KPIs just measured the wrong things, and the reporting was built to be nodded at rather than acted on.

We see this drift story constantly in audits of inherited accounts, and the founder's first question is always the same: how did we not catch this earlier? The honest answer is that nothing in the reporting was designed to catch it.

The fix is boring and works: a KPI contract. A small set of numbers with definitions agreed in writing at kickoff, targets attached, reported on a fixed cadence, reviewed in meetings that end in decisions. This post is the set we'd hold any agency to — including ourselves. It fits on one page, and the one page is the point: what's written down can't be quietly redefined the month the numbers dip.

The north-star three

Three numbers outrank everything else, and all three live downstream of platform dashboards — which is exactly why they're hard to game:

• Blended MER (marketing efficiency ratio): total revenue divided by total marketing spend, every channel included. It can't be inflated by attribution settings, which is precisely why agencies rarely lead with it.
• New-customer CAC: acquisition spend divided by first-time customers. Blended CAC hides an agency milking your existing base with remarketing; the new-customer cut exposes it in one line.
• Contribution margin after marketing: revenue minus COGS, shipping, returns, RTO and ad spend. For COD-heavy Indian D2C this is the only honest number — platform ROAS and reality part ways badly once returns bite.

Set targets for these three with the agency, in writing, with the measurement definition and data source attached to each. Every dispute you will ever have with an agency traces back to an undefined metric.

Channel KPIs that catch the games early

Under the north-star three, a handful of channel-level cuts make the common tricks visible months before they'd otherwise surface:

• Meta: new-customer CPA (not blended), creative velocity — new creatives tested per week — and frequency on core audiences. Rising frequency plus sagging CTR is fatigue the agency should be ahead of, not reacting to.
• Google: brand versus non-brand split — spend, revenue and ROAS reported separately. A gorgeous brand-term ROAS is mostly harvesting demand you already own; it should never hide inside a blended figure.
• Marketplaces: TACoS (total ad cost of sale) rather than ACoS alone, so ad spend is judged against total marketplace revenue including organic.
• Retention channels: email and WhatsApp share of revenue, so "growth" isn't simply paid spend rising.

One line each in the monthly report is enough. The goal is not more data — it's the four or five specific cuts that make the classic games impossible to run quietly.

Leading indicators: what predicts next month

The north-star three tell you what already happened. Four smaller numbers tell you what's about to:

• Creative testing rate — variants launched per week. When testing stalls, performance follows within three to six weeks, every time.
• CTR trend on top spenders — decay here precedes CPA inflation, and it's visible weeks earlier.
• Repeat rate by cohort — if 60-day repeat is sliding, CAC math that looks fine today is already broken; you just haven't received the invoice yet.
• Spend concentration — share of spend in the top three campaigns. Over-concentration means one fatigue event can take out the month.

Ask for all four as a standing block in the monthly report — they take an agency minutes to pull and save you quarters. The tell worth watching: agencies that surface these unprompted are managing the account. Agencies that only ever report trailing ROAS are narrating it.

The cadence: what each review actually decides

Reviews without decisions are theatre, so tie each meeting to the decisions it exists to make:

• Weekly, 30 minutes: execution versus plan — spend pacing, test results, creative pipeline, blockers. Attended by the people doing the work, not account management alone.
• Monthly, 60–90 minutes: the KPI contract review — north-star three against targets, channel cuts, leading indicators, and what changes next month as a result. The senior agency lead attends, no substitutions.
• Quarterly: strategy — channel-mix shifts, incrementality tests, budget re-forecast, and an honest what-did-we-get-wrong section from both sides of the table.

Then apply the simplest audit there is: read the last three months of meeting minutes and count decisions changed by the numbers. If the answer is zero, the reporting is decoration and the drift has already begun. Keep the deck to ten slides and the decision log to one page — volume is where accountability goes to hide.

When the KPIs say fix — and when they say fire

Be fair about time first. Performance marketing needs runway, and judging an agency before 90 days mostly measures noise — here's the realistic timeline. Two bad months accompanied by an honest diagnosis and a changed plan is an agency doing its job through a rough patch, and the KPI contract is what lets you see that clearly instead of panicking at a red cell.

The fire signals are different in kind, not degree: numbers that won't reconcile with your order data, metric definitions that shift when performance dips, targets renegotiated after the fact, or a refusal to report the north-star three at all. Those are integrity failures, and they don't improve with time. If that's what you're seeing, read when to fire your agency and the fourteen red flags — then move deliberately, not angrily.