Measurement Design
Part of Media measurement planning
Recording what cannot be attributed reliably
Keep an attribution limits register that states the missing link, affected result, reporting treatment and effect on decisions.
Record an attribution limit beside the result it affects. State what is missing or uncertain, why channel credit cannot be assigned confidently, which channels or dates may be affected, and what conclusion needs a qualification. ‘Tracking is imperfect’ tells a reviewer too little.
Keep a limits register
Give each issue a row:
| Field | What to record |
|---|---|
| Outcome affected | The sale, lead, booking or other event |
| Missing link | The exposure, click, identity, event or business record that cannot be connected |
| Scope | Channels, devices, regions and dates affected, if known |
| Reporting treatment | Excluded, unattributed, estimated or included under a stated rule |
| Decision effect | Which comparison or conclusion needs a qualification |
| Owner and next check | Who will investigate and when the entry will be reviewed |
Use ‘unknown’ when the affected volume cannot be measured. Do not turn an unknown amount into a percentage merely to make a tidy report.
Distinguish the gaps
No observed path. A customer may buy after encountering several placements, but available records may not connect those encounters to the order. Keep the order in a reliable business total. Do not assign it to a channel solely because that channel reported nearby activity.
Different credit rules. Suppliers can use different events, interaction types or windows. Keep each rule with its figure; their credited actions cannot simply be summed into unique customers.
Modelled rather than directly observed data. Eligible Google Analytics 4 reports can include modelled key events where direct observation is incomplete. Describe the report according to its method rather than presenting every included event as individually observed. Modelling is not necessarily present for every property or event.
Incomplete business outcome. A form submission may be visible while later qualification or an offline sale cannot be connected. Report the form as a response measure; do not rename it a qualified lead or sale.
Say what the evidence supports
A limitation does not make every measure useless. Delivery can still show whether bought activity ran.
A sufficiently complete business record can show orders during the period. Channel-attributed actions can help diagnose response under their stated rules. None of these alone measures the extra orders caused by the campaign.
For a hypothetical campaign, a useful note might read: ‘Confirmed orders are counted in the business total. Phone orders cannot currently be joined to ad interactions, so the channel-attributed sales figures do not describe that path. Its contribution by channel is unknown.’
This identifies the gap without guessing its size or direction.
Review the register when collection or reporting rules change. Keep the earlier entry and its dates so improved observability is not mistaken for a sudden change in media performance.
Reporting principles for uncertain attribution
- Confirmed orders
- Counted in business total — reliable even if unattributed
- Channel-attributed actions
- Help diagnose response under stated rules — but not causal impact
- Phone orders
- Cannot be joined to ad interactions — contribution by channel is unknown
- Hypothetical campaign example
- ‘Contribution by channel is unknown’ — clear, evidence-based statement
