Know which spend actually caused the sale.

A reconciled measurement layer, a model that joins spend to real outcomes, a small set of dashboards people trust, an experiment programme for the causal questions, and a written definition per metric that marketing, sales and finance have all signed. The first item does most of the work: collection you can reconcile is worth more than a sophisticated model built on duplicated events. A good engagement also leaves your team able to run and extend the whole thing themselves.

More than one, deliberately. Last-click stays in the report because it is what most platforms optimise toward; a multi-touch view shows the middle of the journey; pipeline reporting shows what actually closed; blended efficiency shows whether the business as a whole is working. Google's 2025 measurement research makes the point neatly: adoption of MMM, incrementality testing and attribution individually runs at 79%, 86% and 81%, yet only 46% of teams use all three together — and the teams that combine them get the trustworthy answers. Picking one model and defending it is the common mistake.

Neither on its own, and that is expected rather than alarming. Ad platforms count conversions they believe they influenced, using their own windows and modelling, so the totals across platforms will always exceed reality. The sales system knows what closed but little about the touches before the form. We use each for what it is good at — platform data for in-flight optimisation, outcome data for what happened, blended efficiency for the business — and reconcile them on a schedule so the gap becomes a known quantity instead of a monthly surprise.

It is establishing what would have happened without the spend, usually by holding out geographies, pausing a channel in a controlled way, or comparing matched markets. You need it most on the channels attribution flatters: branded search, retargeting, and anything measured by a platform whose own pixel awards its own credit. It is now standard practice rather than an experiment in itself — 52% of US brand and agency marketers already run incrementality tests. We run one when the decision at stake is worth more than the short-term efficiency the test costs, and we say up front what we will do with a flat result.

Once your spend and history are big enough, it is a genuinely useful strategic lens — it works without user-level data, covers offline channels, and answers planning questions attribution cannot. Interest is climbing fast: TransUnion reports nearly half of US brand and agency marketers planning to invest in MMM over the next year. But it is a monthly or quarterly instrument, not a weekly one, and it should be calibrated against experiments rather than trusted on its own. Below roughly a year of clean history we would rather fix collection and run tests first.

A reconciled baseline usually takes four to six weeks: that is the collection audit, the fixes, the metric dictionary and a first set of dashboards. Modelling follows once the inputs agree, and the experiment programme starts as soon as there is a decision worth testing. For long sales cycles the closed-revenue picture necessarily lags, so we report a leading indicator alongside it rather than making you wait a quarter to act. You will see the reconciliation gap shrink week by week in the working session, so trust is built on visible progress rather than a final reveal.

They constrain user-level tracking; they do not prevent good measurement. That is exactly why we lean on first-party collection, server-side conversions, outcome data from your own systems, and experiments — none of which depend on following individuals across the web. Consent is implemented properly, and where consent rates or platform restrictions genuinely reduce the signal we quantify the gap rather than modelling over it silently. Teams that built measurement this way have been comfortably unaffected by each successive browser and platform change.

Yes, and it is often the best split. We tend to own the marketing measurement layer — tags, events, platform APIs, marketing definitions and the experiment programme — while your data team owns the warehouse and governance, with the transformation layer shared and reviewed. Everything we write is documented and version-controlled so it survives handover. Where there is no in-house data capability yet, our analytics team runs the whole layer and trains whoever will inherit it.

The first engagement is normally a fixed-scope audit and reconciliation with a defined deliverable list, priced up front so you approve a number rather than an open-ended retainer. Modelling and the experiment programme follow as a second phase, and ongoing support is a monthly fee scoped to the reviews and tests you actually want. We do not price measurement as a percentage of media spend: the two are unrelated, and linking them creates an incentive we would rather not have.

We can, and many clients prefer it because the loop closes faster when the team reading the evidence is the team changing the campaigns — our paid search and paid social teams work from the same definitions. It is equally fine to buy measurement alone while other agencies buy your media: we report on their performance honestly and share the methodology with them, because a measurement layer only one agency can interpret is not much of a measurement layer.