← All perspectives

Perspective / Search & performance marketing

From attributed revenue to incremental profit: designing a defensible marketing measurement system

Campaign reporting can show where conversions were credited without establishing what advertising caused. A defensible measurement system connects experiments, commercial outcomes and uncertainty before making budget decisions.

10 min readReviewed

Precision balance comparing glass and solid-blue spheres as a metaphor for marketing value.
Conceptual editorial image.

Incremental marketing measurement estimates the business outcomes that would not have happened without an advertising intervention. Attribution allocates credit among observed interactions; it does not, by itself, establish that causal effect. For a commercial board deciding where to invest its next pound, the distinction matters. A useful measurement system combines reliable outcome data, appropriately designed experiments and models whose assumptions are visible. Here we develop a hypothetical regional experiment and show how its findings can inform, but not automatically determine, budget allocation.

The executive view

  • Attributed revenue and incremental commercial value answer different questions.
  • Translate estimated revenue effects into contribution and a visible break-even threshold.
  • Use experiments and models together, with assumptions and uncertainty preserved.
01

Separate three questions that reporting often combines

First ask what happened: spend, leads, qualified opportunities, revenue and contribution. Then ask how platforms assigned credit. Finally ask what changed because of the advertising. These are related questions, but each needs different evidence. A platform’s attributed revenue can rise because measurement improved while the underlying business remained unchanged.

For a B2B service, define the commercial outcome carefully. Signed contract value may overstate immediate economic benefit where delivery costs, cancellation rights or long payment terms vary. An analysis based on expected contribution should document the margin assumptions and distinguish realised sales from weighted pipeline.

Create a shared metric dictionary before introducing a sophisticated model. Finance, sales operations and marketing should agree what counts as a new customer, how renewals are treated and which time period receives revenue. A causal estimate built on inconsistent outcomes can be statistically elegant and commercially misleading.

Define the budget decision before selecting the evaluation method. “Which channel generated the most tracked sales?” differs from “Would removing the next £50,000 of spend reduce contribution?” The latter concerns marginal response, not an average return over historic spending. A channel can have a strong average return and a weak return on the next increment once its most responsive audience has been reached.

Set an analysis grain that matches the decision and the sales process. For long-cycle B2B work, connect opportunity creation, stage changes, contract recognition and cancellations using stable identifiers. Report immature cohorts separately. Otherwise a recently launched campaign can appear inferior simply because its opportunities have not had time to close, while an older campaign receives credit for revenue that was already highly likely.

02

Draw the causal assumptions before fitting a model

Imagine a regional advertising campaign for a business software provider. Advertising, brand demand, competitor activity, sales capacity and seasonality may all influence sales. Strong underlying demand can also cause the team to raise advertising budgets. A simple correlation between spend and revenue therefore mixes several pathways.

A causal diagram helps identify which variables might confound the comparison, which are intermediate steps and which should not be controlled indiscriminately. For example, controlling away every search for the brand could remove part of the effect a campaign was intended to create. The appropriate treatment depends on the question, not a universal dashboard recipe.

Google’s Meridian documentation describes a framework using control variables, business-informed priors and response curves. Those mechanisms make assumptions explicit; they do not make observational data automatically causal. Document the alternative explanations the design can address and those that remain unresolved.

The diagram should make timing explicit. Price changes, stock availability and sales-team capacity can influence both advertising decisions and commercial outcomes. They are not always interchangeable control variables: a service-capacity change caused by the campaign may be part of its effect, whereas a pre-existing capacity restriction may confound the comparison. Record the reasoning with the model specification.

Look for data that cannot identify the effect being asked about. If every channel budget increases by the same proportion each week, there may be insufficient independent movement to separate channel contributions. Adding a more flexible model cannot manufacture that missing variation. The practical next step may be a deliberately varied intervention within acceptable commercial limits, or a broader decision about total media investment rather than unsupported precision about individual channels.

03

Design an experiment around operational constraints

A geo experiment can compare areas receiving an advertising intervention with areas operating under a defined control condition. Before choosing regions, check that media can actually be restricted, sales outcomes can be allocated consistently and customers do not routinely cross boundaries in ways that undermine the comparison.

Use pre-intervention data to assess whether candidate regions behave comparably. Define the intervention, primary outcome, duration, analysis method and stopping rules in advance. Account for the delay between lead generation and a closed sale. A short test might measure qualified pipeline as an early outcome while reserving the revenue conclusion until the sales cycle matures.

Google announced general availability of Meridian GeoX in September 2026, including independent geo experiments and model calibration. Its availability makes the approach more accessible, but the suitability decision still depends on geography, spend, outcome volume and the size of effect the business needs to detect.

Agree the minimum commercially meaningful effect before running the test. A design capable of detecting only a very large uplift may not resolve whether a modest profitable intervention deserves funding. Estimate feasibility using historical outcome variability, available regions and plausible contamination, then decide whether the decision value justifies the required duration and spend.

Randomise where the design permits and document departures. An operational team that reallocates sales attention only to treatment regions creates another intervention. Similarly, a national promotion can change control-region demand during the test. Keep an intervention log, record implementation fidelity and analyse the originally assigned groups under the agreed method. Excluding poorly performing regions after seeing results changes the question and can inflate the reported effect.

04

Work through an illustrative commercial result

Suppose the experiment’s model estimates that the treated regions would have generated £500,000 of revenue without the intervention. Actual recognised revenue is £560,000, with £30,000 of additional advertising spend. The estimated incremental revenue is therefore £60,000, and incremental revenue return on spend is 2.0.

Now apply a hypothetical 40% contribution margin before advertising. Incremental contribution is £24,000. After the additional £30,000 media cost, the estimated short-term contribution change is negative £6,000. An attractive revenue multiple has not produced a positive immediate contribution result.

Illustrative measureAmount
Estimated incremental revenue£60,000
Contribution before advertising, at 40%£24,000
Additional advertising cost£30,000
Net contribution change−£6,000

This simplified calculation excludes future renewals and strategic benefits. Include them only with explicit retention, margin and discount assumptions. Do not quietly substitute lifetime revenue for current contribution to make the investment appear profitable.

Calculate the economic threshold before interpreting the midpoint. With the illustrative £30,000 media increase and a 40% contribution margin, break-even incremental revenue is £30,000 divided by 0.40, or £75,000. If fulfilment requires another £3,000 of genuinely incremental support, the threshold becomes £82,500. This is a decision threshold derived from the assumptions, not an industry benchmark.

Separate a budget test from a whole-channel test. Increasing spend by £30,000 estimates the effect of that intervention under those conditions; it does not directly establish the revenue lost if the entire channel disappeared. Report the actual contrast. Include the treatment period, spend change and outcome window alongside the return figure so an executive cannot accidentally apply the estimate to a substantially different budget decision.

05

Treat uncertainty as part of the decision

The £60,000 estimate is not an observed invoice labelled “caused by advertising”. It is a comparison with a counterfactual. Report the uncertainty interval produced by the chosen method, its assumptions and the range of commercially relevant interpretations.

If a plausible revenue range runs from £10,000 to £110,000, the finance decision is less settled than the midpoint suggests. At a 40% margin, the campaign needs £75,000 of incremental revenue to cover £30,000 of advertising cost before other incremental expenses. That break-even threshold makes the uncertainty actionable.

A statistically inconclusive result does not establish that advertising had no effect. It may mean the experiment was too small, the intervention too weak or natural variation too high. Equally, an impressive point estimate should not justify unlimited scaling. Use the evidence to choose between continuing, redesigning, reducing exposure or gathering additional data.

Present uncertainty in the same units as the commitment. For example, a revenue interval can be translated through the agreed margin assumption into a range of contribution effects, with additional cost assumptions shown separately. A marketing and finance discussion can then focus on the probability and consequence of exceeding the economic threshold rather than debating whether a chart looks positive.

Sensitivity analysis should vary the assumptions that could change the recommendation: margin, delayed recognition, contaminated control demand or customer retention. Do not vary dozens of immaterial inputs merely to create a dense appendix. If a favourable decision relies on retaining customers for three years, identify the cohort evidence behind that assumption and show the decision without it. This exposes whether the proposal is a short-term performance investment or a longer-term strategic bet.

06

Calibrate a broader model without erasing its limitations

A marketing mix model can extend analysis across channels, time periods and spending scenarios. It can represent delayed effects and diminishing returns, then estimate how outcomes might change under a different allocation. Experimental findings can inform model priors or calibration rather than being left in a separate presentation.

Keep the link traceable. Record which experiment informed which channel, what outcome it measured, how recent it is and whether the spending range resembles the proposed scenario. A regional test of one channel should not be treated as conclusive evidence for every audience, product or future budget level.

Model diagnostics matter as much as the optimisation output. Examine fit, sensitivity to assumptions, plausible response curves and whether channels can be distinguished with the available data. If spend always moves together, the model may struggle to separate effects. In that situation, a confident-looking allocation chart can convey precision the evidence does not support.

Use backtesting to assess predictive behaviour, but do not equate a good forecast with a valid causal decomposition. Several channel-effect combinations can predict the same total outcome. Holdout prediction, posterior checks and sensitivity to priors each answer useful questions; none independently proves that every attributed contribution is correct.

A scenario also needs an operating range. An allocation recommendation far beyond observed spending may rely heavily on the assumed shape of a response curve. Constrain proposals to credible ranges and test major movements in stages. Record the business actions needed to implement them: creative supply, sales capacity and conversion operations may limit what a theoretically attractive allocation can achieve. Modelled opportunity has commercial value only if the organisation can execute the intervention being represented.

07

Build a measurement programme proportionate to the decision

Not every organisation needs an enterprise MMM programme. A business with low conversion volume, limited geographic reach or inconsistent historical data may gain more from fixing outcome capture and running a narrow controlled test. Start with the highest-value unresolved decision rather than the most elaborate available tool.

A practical roadmap names an owner for outcome definitions, an owner for experimentation and a finance sponsor for commercial interpretation. Maintain an experiment register, a record of model assumptions and a schedule for reviewing whether previous findings still apply. Revisit evidence after material changes in pricing, product, market or sales execution.

The board should receive a decision range: what the evidence suggests, which assumptions drive it, how much is at risk and what the next measurement step would resolve. That creates a more defensible investment discussion than treating attributed revenue, modelled revenue and incremental profit as interchangeable numbers.

Maintain a decision register connecting evidence to action. For each meaningful budget change, capture the question, method, estimated effect, uncertainty, accepted assumptions and observation plan. Revisit decisions when new evidence arrives rather than retaining only the successful tests in a case-study archive. Negative and inconclusive findings also prevent repeated spending on unresolved hypotheses.

Use a staged maturity path. First make commercial outcomes reliable; next run a focused experiment; then decide whether a broader model would change enough decisions to justify its cost. A small organisation may benefit from a disciplined quarterly test more than a continuously refreshed model that nobody can challenge. For a larger portfolio, the model can coordinate many decisions, but it still needs experiment-informed checks and finance ownership of the meaning of profit.

08

Questions to resolve before the next media increase

Before approving a material increase, ask for the commercial outcome, the intervention being evaluated and the counterfactual. Then ask whether the result concerns average historic return or the proposed next increment. Those distinctions should be visible in the main decision paper rather than buried in methodological notes.

Require a reconciliation between platform reporting and the commercial ledger. Differences can be legitimate because of attribution windows, cancellation treatment or revenue timing, but they need an explanation. Ask which assumptions most strongly influence the recommendation, what exposure is being accepted and how the team will know whether the decision worked.

These questions do not demand perfect measurement before taking action. They make uncertainty an explicit part of capital allocation. A credible performance partner should be able to describe what the available evidence supports, propose a proportionate next test and explain the circumstances in which its own recommendation should change.

Further questions

Questions this raises

What is the difference between attribution and incrementality?

Attribution assigns credit for observed conversions to marketing interactions. Incrementality estimates the difference an intervention caused compared with what would otherwise have happened.

Does every business need marketing mix modelling?

No. Suitable data volume, history and meaningful variation are important. A focused experiment or better CRM outcome capture can be a more useful first investment for a smaller or less measurable business.

Can an inconclusive experiment prove advertising does not work?

No. Inconclusive evidence can reflect limited statistical power, noisy outcomes or a weak intervention. Review the plausible effect range and commercial threshold before deciding what to do next.

References

Sources & further reading

  1. Drive profitable growth with new data and measurement tools 10 September 2026
  2. An introduction to Meridian Updated 9 July 2026

Connected capabilities

Search & performance marketingStrategy & digital transformation

From thinking to doing

What could this mean
for your business?

Your next chapter

What could your
business do better?

Tell us where you want to go, what’s getting in the way, or what you need to deliver. We’ll start there.

Start a conversation

Start a conversation

Tell us what
you have in mind.

Interactive preview. Your information stays in this page and is not sent.