Reviews
Part of Reading campaign measurement with a sceptical eye
Straight answers to the campaign measurement questions
Common campaign measurement questions cover scope, success, attribution, lift, timing, channel reports, offline outcomes, ownership, and evidence limits.
What to take away
- Start with a boundarynamed audiences, tactics, offers, dates and versions, and nothing else.
- Delivery, attribution and incrementality answer different questions, so label which one produced any number you quote.
- Put each answer in a decision record: owner, population, period, method, result, limitation, action.
- Reconcile the campaign figure with the system of record and report the gap in units.
- Set one review date and one named trigger that reopens the answer.
Fix the campaign boundary first
A campaign boundary is a written list you can defend line by line. It fixes who is eligible, what counts, and when counting stops. Teams that skip it argue about definitions instead of results, the failure mode covered in campaign measurement.
Campaign boundary
- Audiences
- Existing trial accounts in the US and Canada; lookalike prospecting excluded
- Tactics
- Paid search, paid social, two emails, one webinar
- Offers
- 20 percent first-year discount; no free tier
- Dates
- March 4 to April 30; reporting closes July 15
- Versions
- Creative v1 to March 17; v2 from March 18
Fix the boundary
- Write those five rows into one file the whole team can open.
- Name an owner for each row.
- Exclude spend that cannot be tagged to a row.
- Re-check the file when pricing, creative or dates change.
Four scope questions sit on top of the table: what counts as the campaign, what counts as success, who owns the result, and when reporting closes.
Campaign measurement scope questions
- What counts as the campaign?
- What counts as success?
- Who owns the result?
- When does reporting close?
Match the method to the question
Attribution assigns credit among observed interactions under a stated rule. Incrementality estimates what changed against what would likely have happened without the campaign. Use both if you like, but attributed volume is not incremental impact. The rules and their limits sit in marketing attribution.
Match method to question
| Question | Method | Output |
|---|---|---|
| Which touches preceded a conversion? | Multi-touch attribution | Credit split under a stated rule |
| Did the campaign change behaviour? | Holdout or geo test | Lift against a baseline |
| What did a channel contribute over a quarter? | Marketing mix modeling | Aggregate contribution with an interval |
Channel reports need care. Keep raw delivery beside any deduplicated total, and state which people, devices and offline events fall outside. Google Analytics 4 and a warehouse export rarely agree, so fix one identity and lookback policy, then label modeled fields where they appear.
The Media Rating Council's cross-media standard defines deduplicated video audience measurement across television and online; read its scope before borrowing a field from it.
Fill in the decision record
The record is the answer. Use the same fields every time, and date the file.
Decision record
- Decision
- Move budget between paid social and paid search
- Owner
- VP Growth
- Population
- US trial accounts created from 4 March
- Period
- To 30 April; closed 9 May
- Method
- Last non-direct click, 30-day lookback, 10 percent holdout
- Result
- 412 attributed upgrades; 1,180 orders in the system of record
- Limitation
- Non-US accounts unread
- Action
- Shift 15 percent of paid social to search; review 6 June
Settling one shared outcome across channels belongs to marketing KPIs, not to one campaign report. If the same question returns, compare records before changing the metric. A changed population or source explains most drift.
Example: reconciling campaign outcomes with the order table
Report the gap in units, not in adjectives.
Reconciling outcomes
| Line | Count | Treatment |
|---|---|---|
| Orders in the system of record | 1,180 | Starting total |
| Orders carrying the campaign flag | 640 | Eligible population |
| Matched to a campaign touch by key | 412 | Reported as attributed |
| Eligible but unmatched | 228 | Reported as an open gap |
| Flagged to a different campaign | 180 | Excluded from this record |
| Refunds after the close date | 36 | Subtracted from the total |
Write '228 eligible orders unmatched', not 'some coverage loss'. Offline sales follow the same pattern: eligible sale, matching key, timing window, coverage and privacy basis. In Canada, matched purchase data needs a documented consent basis under PIPEDA and, in Quebec, Law 25. A team rewriting definitions should start with how to measure a campaign.
Borrow the principle, not the rulebook
Public standards transfer one idea each. The CDC's 2024 program evaluation framework ties evaluation questions to designs, indicators and metrics. The GAO evaluation design guide shows how evidence needs shape the design. The W3C privacy principles warn against shifting privacy work onto individuals. None of them certifies a marketing result.
