Measurement
Part of Programmatic campaign optimisation
Comparing inventory segments at a useful sample size
Judge whether inventory-segment delivery and outcomes can support a budget decision while keeping small samples and buying differences visible.
An inventory segment has a useful sample when its observed delivery and outcomes can inform the allocation decision you plan to make. There is no universal impression minimum: enough impressions to assess delivery may still yield too few conversions to judge effectiveness. Define the outcome and the smallest difference worth acting on before reading the comparison.
Make the rows comparable
Define each segment precisely, such as a deal, supply source, format or placement group. Use the same event, attribution setting, cost definition and reporting cut-off across rows.
Show market, device, creative and audience differences rather than hiding them in a blended rate. Video and display buys, for example, may have different exposure measures and opportunities.
For each row, record its ID, dates, spend, bought impressions, relevant outcome count and the denominator behind any rate. Keep material eligibility differences visible, and treat any availability or forecast figure as a planning estimate rather than a record of campaign wins.
Let the decision set the sample
Ask how large a difference would justify moving budget and how easily another outcome could change the apparent ranking. Suppose a hypothetical segment has 2 actions from 20 eligible visits: its observed action rate is 10%, or 15% with one more action from the same 20 visits.
A segment with 20 actions from 200 visits moves from 10% to 10.5% with one more action. This arithmetic illustrates sensitivity; neither example is a campaign benchmark or a statistical test.
Use evidence suited to the decision. Requests, submitted bids and won impressions can help diagnose deliverability.
A conversion-led reallocation needs completed outcomes; impressions alone cannot substitute for them. Compare periods with similar time for actions to appear.
An extract taken early may not reflect later updates. An update cadence does not tell you when this campaign's customer actions are mature.
Decide what the comparison supports
| Evidence | Defensible response |
|---|---|
| One segment has little bought delivery | Diagnose eligibility or observe longer; do not call its outcome rate inferior. |
| Both deliver, but outcomes are sparse | Keep allocations bounded and set the next review point. |
| A decision-relevant difference persists under comparable definitions | Consider a limited shift and monitor what the added spend buys. |
| Buying conditions differ materially | Describe the performance of those buys without claiming the segment label caused the difference. |
An ordinary segment report is an operating comparison. A controlled experiment is a separate task: test one variable between arms where possible, and expect limited reach to produce wide confidence intervals. Keep a routine allocation change reversible and state what remains uncertain.
