Costs
Part of Reading campaign measurement with a sceptical eye
How to confirm your campaign measurement setup won't mislead you
How to measure campaign measurement starts with a decision, fixed boundary, outcome, comparison, tracking test, quality gate, economics, report, and action log.
What to take away
- Write the objective and comparison before campaign delivery.
- Test the complete record path from exposure through mature business outcome.
- Report uncertainty and data coverage with every decision-critical result.
How to measure campaign measurement is a real question: teams must test whether the measurement process itself is fit for a business decision. Use this sequence to turn a brief into auditable evidence.
Use a ten-step sequence
Step / Required work
- 1
- Name the decision, owner, and deadline
- 2
- Freeze the campaign boundary and version
- 3
- Define the mature outcome and value
- 4
- Choose the comparison and evidence design
- 5
- Write metric and source contracts
- 6
- Instrument and reconcile test records
- 7
- Set power, duration, quality, and guardrail rules
- 8
- Monitor delivery without changing the primary test
- 9
- Analyze absolute, relative, economic, and uncertainty results
- 10
- Log the decision and later validation
NIST's engineering statistics handbook advises teams to state experimental objectives before choosing a design. The page addresses experiments generally, but its focus on clear objectives, responses, factors, and constraints supports disciplined campaign planning. Objectives only make sense once the marketing KPIs behind the campaign are agreed.
Ten-Step Measurement Sequence
- Name decision, owner, deadline
- Freeze campaign boundary and version
- Define mature outcome and value
- Choose comparison and evidence design
- Write metric and source contracts
- Instrument and reconcile test records
- Set power, duration, quality, guardrails
- Monitor delivery without changing primary test
- Analyze absolute, relative, economic, uncertainty
- Log decision and later validation
Validate the comparison
Check assignment, eligibility, contamination, sample loss, preperiod balance, concurrent activity, and whether the outcome can mature within the analysis window. Do not extend or segment a weak result after seeing it unless the change is labeled exploratory.
Google's explanation of Conversion Lift reporting separates incremental conversions from standard attributed conversions and describes modeled projections, intervals, incremental cost, iCPA, and iROAS in that product. Those fields depend on study setup and eligibility; they are not universal campaign formulas.
- Preserve the prespecified primary outcome
- Show treatment and comparison counts
- Report absolute and relative effect
- Include interval or uncertainty information
- Align incremental value and cost to the same population
- State platform, product, and study limitations
Finish with a decision record. Capture what the team knew, the choice it made, expected effect, owner, and review date. When mature outcomes arrive, test whether the original interpretation held. Judging whether a result is normal also needs marketing KPIs benchmarks built on the same population.
Run a controlled handoff
Assign each step to a named role and require an observable finish condition. The analyst should be able to reproduce the result, while the decision owner can explain the action and stop rule. Include one broken-data case and one revoked-access case. Record the repair, the time required, and any vendor help.
The GOV.UK technology selection guidance recommends choices that can change over time, preserve data control, address security risk, and include ownership cost. Those public-service rules become useful buying questions, but they are not private-sector mandates.
CISA's software acquisition fact sheet points buyers toward development practice, supply-chain exposure, deployment, and vulnerability management. Add those questions to the tooling review. The government context does not replace a buyer's own testing or contract review.
Keep the evidence record beside the decision so a reviewer can reproduce the reasoning. The common campaign measurement questions cover the objections a reviewer usually raises about the design.
Common questions
Can measurement begin after launch?
Basic reporting can, but causal and baseline options may already be lost. Document the limitation and prepare the next campaign before delivery.
What if the test is inconclusive?
Report that the study did not distinguish the planned effect at the stated uncertainty. Do not convert absence of evidence into proof of no effect.
Should teams optimize during a lift test?
Protect the prespecified design. Limit operational changes to safety or delivery rules, and document any change that could affect interpretation.
