Google’s measurement update: recovered conversions are not incremental demand

When a marketing dashboard starts reporting more conversions after a measurement change, the budget conversation can move faster than the evidence. Did the business gain customers, or did the system become better at observing activity? Google’s September update gives marketing leaders a timely reason to separate those questions. The useful response is a clearer decision process: establish what changed, reconcile the outcome and choose evidence that can support the proposed investment.

What Google announced on September 10, 2026

Google introduced the Data Strength Uplift Metric for conversions recovered through first-party data setups. It also announced global general availability for Meridian GeoX and new agentic capabilities in Meridian. These are distinct measurement developments. Read Google’s dated announcement.

The announcement establishes the product direction, not whether a particular account already exposes every feature. Check the account and its current documentation before promising a rollout. Our analysis below concerns the decisions a team should make when its evidence changes; it does not depend on every announced feature being present.

First identify which question the number answers

Recovered measurement asks whether more activity is now represented in a reporting system. Business reconciliation asks whether those reported outcomes correspond to the orders, accepted opportunities or other events the company actually values. Incrementality asks what additional outcome the marketing intervention caused relative to what would have happened without it. Each question needs its own evidence.

That distinction matters when AI participates in campaign decisions. A team should understand the goal and the evidence behind it before approving changes to spend. More observable activity can improve the information available to a decision process without demonstrating that the latest campaign created more demand. Treat the reporting change as something to explain before treating it as a budget recommendation.

Keep a dated measurement change record

Before making a comparison, record the configuration change, implementation date, affected conversion actions and reporting period. Include the business definition of each action, its counting rules and the system that verifies it. Ask the measurement owner whether any other changes occurred around the same time, such as a different import process or a revised qualification definition.

For example, imagine a retailer changes its data connection and subsequently sees more reported purchases. This is an illustrative situation, not a client result. The first task is to compare the affected reporting periods with order records using agreed definitions. Counting the additional reported purchases as additional orders would skip the very question the team needs to answer.

Choose the next check from the observation

Use this editorial decision table in a review meeting. It is a triage method, not a Google product specification. Assign one person to each next check and document the evidence required to close it.

Which evidence should guide the next marketing decision?
Observation	What it supports	Next check before changing the budget
Reported conversions rise after a data change	A change in recorded activity that needs explanation	Reconcile definitions, dates and business records; separate measurement effects from demand.
Reported leads rise but accepted opportunities do not	A possible quality, definition or handoff mismatch	Review acceptance criteria and follow-up outcomes with the commercial owner.
Business outcomes improve while several initiatives change	An observed business improvement with unclear attribution	Identify competing explanations and whether a controlled test is feasible.
A geo experiment estimates an effect with uncertainty	Evidence about a specified intervention in the tested conditions	Review the design, uncertainty and business relevance before applying it elsewhere.
Evidence is incomplete or incompatible across systems	An unresolved decision dependency	Assign the missing check; retain uncertainty instead of filling the gap with an ROI claim.

Use a causal experiment when the decision needs one

Meridian GeoX is Google’s open-source, publisher-agnostic solution for geographic experiments. Its documentation describes standalone use and integration with marketing mix modeling. That makes it relevant to a different question from recovered measurement: the effect of a defined intervention. See the GeoX documentation.

Start with the business decision: for example, whether changing investment in a channel is justified. Specify the intervention, the outcome, the markets and the period before reviewing results. Ask an analyst to assess whether the available regions and data can support the proposed comparison, and whether other changes would make the interpretation unreliable.

Practical constraints matter. A promotion applied only in one region, uneven inventory or a sales process that changes mid-test can complicate the story. The method should reflect the business conditions, not force them into an attractive dashboard. If the setup cannot answer the intended question, repair it or choose a narrower decision. An inconclusive result is a legitimate outcome.

Ask for an interpretation, not just a lift number

Require the analysis owner to explain the estimated effect, uncertainty, included costs and the conditions under which the finding applies. Separate the statistical estimate from the proposed commercial action. A result relevant to one intervention and period does not automatically justify the same allocation across every market or channel.

The official GeoX repository describes controlled geographic testing and its use in marketing measurement. The operational review questions here are our editorial framework, rather than a claim that software resolves every design problem. Review the project and methodology entry points.

Connect measurement to the work someone will change

Once the evidence is understood, name the action and the person authorized to take it. A data discrepancy may call for an instrumentation fix. Weak lead quality may require revised criteria or a commercial handoff review. A credible experiment may support a bounded allocation decision with a scheduled reassessment. These are different actions and should not share an automatic approval rule.

If the proposed change involves an AI workflow, evaluate its operating costs and useful business outcomes separately. Our guide to measuring an AI marketing pilot addresses that evaluation. For the step before a pilot, use the first-workflow selection framework to establish usable inputs and ownership.

Bring a short evidence brief to the next budget meeting

Prepare a one-page brief with five items: the decision being requested, the observed change, the business record used for reconciliation, the remaining uncertainty and the owner of the next check. Label observed facts, estimates and proposals separately. This gives finance, marketing and commercial operations something concrete to review.

The broader task is to turn signals into accountable work, as discussed in revenue intelligence. For this update, the immediate priority is simple: understand what the new evidence measures, then decide what it warrants changing. The presence of a new metric is a reason to improve the conversation, not a substitute for it.