Building a Measurement Plan Before the Campaign Runs

Most measurement problems are created in the planning stage and discovered at the reporting stage, when nothing can be done about them. For teams coordinating several accounts, this practical guide offers a related example of structuring work across multiple clients.

The campaign launched alongside a price promotion, so neither can be isolated. No holdout was kept, so there is no comparison. The tracking for one payment method was broken the whole time. Nobody agreed what success meant, so the debate about whether it worked is a debate about definitions.

All of these are prevented by one page written before launch, and the page takes twenty minutes.

What goes on it

The decision this campaign informs

Not the objective — the decision. "Whether to continue this channel next quarter." "Whether to increase this budget." "Whether this creative approach replaces the current one."

If no decision depends on the result, that is worth knowing before spending on measurement. Some campaigns run for reasons that are not measurable, and saying so up front is better than manufacturing a metric afterwards.

The primary metric

One. Named, defined precisely, with the source system stated.

Not three, and not a list of things to look at. Multiple primary metrics is how a null result becomes a positive one after the fact. See A/B tests.

Guardrail metrics

Things that must not get worse. Refund rate, margin, unsubscribe rate, support contacts, page performance.

A campaign that raises orders and refunds equally has not worked, and without a guardrail nobody checks.

The comparison

The most important line, and the one most often absent.

What are we comparing against? A holdout group, matched regions, a pre-period, a switchback pattern.

If the answer is "the previous period," say so explicitly and note what else differs between the periods. That is a weak comparison and it is sometimes all that is available — recording it as weak is what stops it being over-read later. See answering "did the campaign work" honestly.

The detectable effect

Given the expected volume, what is the smallest effect this could distinguish from noise?

If the answer is larger than the effect you expect, the measurement cannot inform the decision. That is worth knowing before spending, and it changes the plan — a longer run, a bigger sample, or an acceptance that the decision will be made on judgement.

The decision rule

What result leads to what action, written before the result exists.

"If incremental revenue exceeds spend by 20%, continue at this budget. If it is between break-even and 20%, continue at reduced budget and retest. If below break-even, stop."

This single line prevents the most common failure in marketing measurement: a result arrives, everyone interprets it in the direction they preferred beforehand, and the discussion is about methodology rather than about the decision.

The tracking check

What must be working, and who confirms it before launch.

A test conversion completed on each payment method and device. The campaign tags validated against the convention. The holdout confirmed excluded from every channel, not just the tested one.

This is fifteen minutes and it prevents the most expensive failure, which is discovering at the end of a campaign that half its conversions were never recorded. See data quality.

What else is changing

Everything else happening in the window: promotions, price changes, product launches, site deployments, other campaigns, consent banner changes, seasonal events.

If two things launch together, neither can be measured. Naming them beforehand is the moment when someone can still move one by a fortnight, and that fortnight is what makes both interpretable.

The reporting date and audience

When the result is delivered, to whom, and in what form. Agreed now, so nobody is waiting for a number that was never scheduled.

The one-page template

Campaign:
Decision this informs:
Primary metric (definition, source):
Guardrails:
Comparison group / method:
Minimum detectable effect:
Decision rule:
  If X → we do A
  If Y → we do B
Pre-launch tracking checks (owner):
Other changes in the window:
Report due (date, audience):

Ten lines. It is not a project document and it should never grow into one.

Who writes it and when

Written by whoever will report on it, because they are the person who will discover the gaps later.

Agreed at campaign planning, alongside budget and creative — not afterwards.

Signed off by whoever will make the decision, which is the step that gives it force. A measurement plan the decision-maker has not seen is a document the analyst wrote for themselves.

The conversation this makes possible

The plan changes what happens when the result is disappointing.

Without it: the number is low, the methodology gets questioned, the comparison period gets renegotiated, and a decision is eventually made on the basis of who argued more persistently.

With it: the rule was agreed, the result is checked against it, and the decision follows. Somebody may still dispute it, and now they are disputing a rule they agreed to rather than a number the analyst produced.

This is the main value of the document and it is not really about measurement. It is about removing the incentive to relitigate.

When there is no time

The abbreviated version, which is better than nothing:

Three questions before launch. What decision does this inform? What are we comparing against? What result would make us stop?

Two minutes, verbally, and written in a message afterwards so it exists.

Most of the benefit comes from the comparison question, because that is the one that has to be arranged in advance and cannot be recovered later.

The summary

Measurement problems are created before launch and discovered after. One page beforehand prevents most of them.

The comparison group is the line that matters most — it cannot be added retrospectively.

Write the decision rule before the result exists, and the argument about methodology stops happening.

And check tracking before launch, because a campaign that ran with broken tracking cannot be measured at any price afterwards. A concise public-sector template is available in the GOV.UK measurement-plan guidance.