Numbers you would bet the budget on.

Four things: collection you can trust, definitions everyone agrees on, a data model that joins spend to outcomes, and reporting that ends in decisions. In practice the first two matter most, because a sophisticated attribution model on duplicated events produces confident nonsense. A good engagement leaves you with a validated measurement layer, a warehouse where ad cost and closed revenue sit together, a handful of dashboards people actually use, and a monthly review that changes what you do next. It should also leave your team more capable rather than more dependent.

Neither, on its own. Ad platforms count conversions they believe they influenced, using their own attribution windows and modelling, so totals across platforms will always exceed reality — each claims the same sale. The sales system counts what closed but usually knows little about the touches before the form. The workable answer is to use each for what it is good at: platform data for in-flight optimisation, CRM data for what actually happened, and a blended efficiency number for the business. Then reconcile them on a schedule so the gap is a known quantity rather than a monthly surprise.

GA4 is enough for behavioural analysis and, configured well, for a lot of day-to-day reporting. It is not enough when your revenue truth lives in a sales or billing system, when refunds and cancellations materially change the picture, when sales cycles run for months, or when you need to model spend against contribution. At that point a lightweight warehouse — BigQuery or Snowflake with a modest transformation layer — is the cheaper answer, because it lets you join sources properly instead of reconciling exports by hand. We recommend one only when the joins justify it.

Chrome reversed its deprecation plan in 2025, so the deadline disappeared and the problem did not. As EMARKETER put it, Safari and Firefox already block third-party cookies by default, which makes cookie-based measurement the exception rather than the rule — before counting consent tooling, tracking prevention and app environments. The durable response has not changed: collect first-party data properly, move critical conversions server-side, implement consent correctly, and lean on experiments for the causal questions. Teams that did that work are unaffected by whichever way the browsers jump next.

By pushing outcomes back into the systems that buy media. Offline conversion imports and CRM-stage feedback let the ad platforms optimise toward qualified pipeline instead of raw form fills, which is usually the single highest-value change in a B2B account. We map the stages that predict revenue, assign values to them, and report both the leading indicator and the closed number so nobody has to wait a quarter to act. Where the cycle is long enough that attribution is genuinely unresolvable, we design a holdout test instead of arguing about models.

It is the practice of establishing what would have happened without the spend — typically by holding out a set of geographies, pausing a channel in a controlled way, or comparing matched markets. You need it most on channels attribution flatters: branded search, retargeting, and anything measured by a platform whose own pixel decides the credit. The cost is some short-term efficiency; the return is knowing which line of the budget is genuinely creating demand. We run tests when the decision at stake is bigger than the cost of the test, and not as a default ritual.

Less, in our experience. Most teams are spending days a month reconciling exports, and that work disappears once the joins happen automatically. We also deliberately build fewer dashboards than clients expect: three good views beat fifteen nobody trusts, and every additional chart is a maintenance liability. Alerting takes over the monitoring that people currently do by eye. What we do ask for is an hour a week from someone who owns the numbers, because measurement that nobody inside the company owns decays quickly.

Yes, and it is often the best arrangement. We tend to own the marketing measurement layer — tags, events, platform APIs, marketing definitions — 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 you have no in-house data capability, we run the whole layer and train whoever will inherit it, which is the same reason we keep the stack conventional rather than clever.

The first engagement is usually a fixed-scope audit and rebuild: measurement audit, remediation, metric dictionary and a baseline set of dashboards, priced up front so you can approve a number rather than an open-ended retainer. Warehouse and attribution work follows as a second phase, and ongoing support is a monthly fee scoped to the reviews and testing you want. We do not price analytics as a percentage of media spend — the two have nothing to do with each other, and tying them creates exactly the wrong incentive.

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