The Measurement Cadence That Survives a CFO Review

Fifty-two percent of US brand and agency marketers now use incrementality testing to measure their campaigns, according to a July 2025 survey from eMarketer and TransUnion, and 36.2% plan to spend more on it over the next twelve months. The same research puts media mix modeling at the top of the reliability rankings. The argument about which measurement method wins is over. The winning answer is all three, arranged in a hierarchy, run on a schedule, and reconciled in one meeting that finance attends.

This is that schedule: which test runs when, who owns the room where the numbers get reconciled, and what a CFO actually signs at the end of the quarter.

Three methods, one sentence each

Precision starts with definitions, because half of all measurement arguments are two people using the same word for different things.

Media mix modeling (MMM) is a statistical model built on aggregate weekly data that estimates each channel's contribution to revenue without tracking any individual person. An incrementality test is a controlled experiment that withholds or varies advertising for a comparable group in order to measure the sales that would have happened anyway. Platform attribution is the conversion credit an ad platform assigns to itself based on the touchpoints it observed or modeled inside its own walls.

The 2026 hierarchy gives each a role. MMM is the backbone: it produces the cross-channel budget allocation. Incrementality tests are the validation layer: they check the model's most consequential claims against experimental ground truth. Platform attribution is demoted to a signal: fine for creative decisions and bid adjustments inside a channel, ineligible for budget decisions across channels. If your team is still framing this as MTA versus MMM, the short version is that the market has already voted for both, in different jobs.

Why the hierarchy settled

Three data points explain the consolidation.

Trust inverted. In a January 2026 survey from Haus reported by eMarketer, 60% of US senior decision-makers said they trust independent incrementality testing most among measurement solutions, against 40% for media mix modeling and 37% for in-platform reporting. Experiments now outrank the dashboards that used to run the meeting.

Money followed. TransUnion found that 47% of marketers plan to increase MMM spend within the year, with 26% citing dissatisfaction with their current measurement technology as a driver. And when the same TransUnion survey population was asked to name the single most reliable measurement approach, MMM took the top spot at 27.6%.

Tooling got cheap. Google open-sourced its Meridian media mix model in early 2025 and Meta maintains the open-source Robyn project, which means a class of model that cost six figures a decade ago now runs in a notebook. The bottleneck moved from modeling to governance: who calibrates the model, who reconciles the numbers, who signs the memo.

For a fuller stat sheet to bring into your own deck, see our roundup of marketing attribution and measurement statistics.

The quarterly calendar

The table below is the entire playbook. Everything else in this article is commentary on it.

Two design principles hold the calendar together. The MMM refresh runs first because the model decides what gets tested: experimental capacity is scarce, so it goes to the channels where the model and the platforms disagree most, since that is where a wrong number costs the most money. Second, at least one experiment is always in flight. "Always-on incrementality" is a simple rule with a precise meaning: no quarter ends with zero completed tests, because a model validated once a year drifts unchecked for eleven months.

Which test runs when

Each experiment type has one job, so define it before scheduling it.

A geo holdout is an experiment that pauses or reduces spend in a set of matched geographic markets while spend continues everywhere else. It produces the strongest evidence available and takes the longest, so reserve it for the largest disputed budget line, run it four to six weeks, and never run two holdouts that overlap in audience or funnel stage. Contaminated cells produce reads you cannot use, and a wasted holdout costs you a full quarter.

A conversion lift study is a platform-run experiment that compares users randomly withheld from your ads against exposed users. It is faster and cheaper than a geo test but scored by the platform being graded, so treat the result as directional and schedule a geo test the following quarter to confirm any number that would move seven figures.

A scale test is a deliberate, stepped budget increase in a single channel to observe where marginal returns flatten. Run one when the MMM's response curve claims a channel is underfunded and you want experimental confirmation before committing the increase permanently.

Seasonality is a scheduling constraint, and pretending otherwise ruins tests. A retailer should not hold out its largest channel in Q4. Either run the decisive tests in Q2 and Q3 and carry those calibration factors through peak, or accept smaller holdout cells in November and wider confidence intervals with them.

The reconciliation meeting

One meeting, sixty minutes, week 11 of every quarter. It has a fixed cast.

The measurement owner (usually the analytics lead) prepares the pre-read and presents. The VP of marketing chairs and makes the calls. The FP&A partner attends with a veto on anything entering the financial plan. Channel leads attend to add context on execution; they do not grade their own channels.

The agenda is four items. First, the calibration readout: where experiments confirmed the model, where they contradicted it, and what multipliers now apply to each platform's self-reported numbers. Second, the disagreement review: the updated ranking of channels by MMM-versus-platform gap, which becomes next quarter's test backlog. Third, budget consequences: reallocations the calibrated model recommends, stress-tested as scenarios before the meeting (a media mix calculator is the fastest way to pressure-test them). Fourth, the decision log: every number that changed, why it changed, and which test or model run justified it.

Two standing rules keep the meeting honest. A platform number within roughly 20% of the calibrated estimate stays in service as a daily signal. A channel where the experiment and the model disagree by more than 2x escalates automatically to CFO review, because one of your two most trusted instruments is broken and you do not yet know which.

If nobody on the team can own the analytics lead seat credibly, that seat is precisely what a data and analytics partner exists to fill; the cadence fails without it.

What finance actually signs

The output of week 12 is a one-page memo. A CFO signs it when it contains five things.

Ranges. Every channel ROI is stated as a confidence interval, never a point estimate. A memo that says "CTV returned 2.1x" invites the question it cannot answer; "1.7–2.5x, geo-validated in Q2" survives an audit.

Pre-registration. Confirmation that hypotheses, cell design, and success thresholds were documented in week 3, before any results existed. This is the single strongest defense against the suspicion that marketing shopped for a favorable read.

Calibration factors. The documented multipliers applied to each platform's self-reported conversions, with the test that produced each one.

A model-free guardrail. Marketing efficiency ratio (MER) is total revenue divided by total marketing spend, and it works as finance's cross-check because it requires no model and no attribution logic at all.

If the calibrated stack claims efficiency improved and MER moved the other way, the memo does not get signed until someone explains the gap.

The decision log entry. What changed, what stayed, and what the next quarter's tests will resolve. Over four quarters this log becomes the most valuable measurement asset the company owns: a running record of claims made, tests run, and money moved, in an order an auditor can follow.

That is the real product of triangulation. The individual methods were never the point. The point is a system where marketing's numbers arrive in finance's language, with error bars attached and receipts behind them, on a date everyone already had on the calendar.

Where the survey figures and the calendar come from

The survey figures are drawn from the sources linked beside them: the incrementality adoption and spending-intent numbers from the July 2025 eMarketer and TransUnion survey, the trust ranking from the January 2026 Haus survey as reported by eMarketer, and the MMM investment and reliability figures from TransUnion's own release. They are shares of surveyed US marketers, so they describe adoption and sentiment; none of them measures whether the methods improved anyone's results. The 12-week calendar and the meeting rules are our own operating template rather than a survey finding: they come from running the cadence with clients, and the stretch-or-compress note in the table is where it bends by sales cycle. On engagements the template is the starting point; the actual test windows, the disagreement thresholds and the memo format are set with the client's FP&A partner in the first quarter and adjusted from the decision log after that.

Where does the quarterly cadence break down?

The calendar assumes a company with enough spend, enough history and a sales cycle short enough to fit a test inside a quarter. Where one of those fails, so does the cadence:

• Sales cycles longer than the test window. A geo holdout of four to six weeks cannot observe pipeline turning into revenue for a considered B2B purchase. What you will see is a null read on revenue and a clean read on leads. The trade-off is to measure lift on a leading indicator with a documented conversion rate to revenue, and to accept that the memo carries that assumption.

• Spend too small or too flat for the model. MMM needs variation in spend to estimate anything. Under a modest budget that has run at the same level for years, the model returns its priors and calls them results. In practice the honest version of triangulation here is tests plus platform data, with the model added once spend varies.

• The same channel winning the disagreement audit every quarter. Brand search will top the gap ranking forever, because it will never calibrate near the platform's claim. The pitfall is spending scarce test capacity on a known answer. Cap re-tests on a channel whose multiplier is stable and move down the list.

• A single test overwriting the model. Feeding one lift result in as a hard prior anchors the MMM on a six-week snapshot. The downside is a model that now agrees with the test in every season. Use the result to narrow the prior rather than set it.

• A new channel with no history. Neither the model nor a test can see it yet; the exception to the demotion rule is that platform attribution is all you have for the first quarters, and the mistake most teams make is pretending otherwise.

That said, the cadence still holds in every case above; it is the method assigned to each slot that changes.

Sources

• eMarketer: FAQ on incrementality (2026)
• eMarketer: Incrementality testing earns marketers' top trust
• TransUnion: Marketers' confidence in measurement has stalled