Attribution is a rule, not an observation
The most useful thing a clinic can understand about attribution is that it is a modelling choice. When somebody sees a post, searches for the clinic a week later, asks a friend, walks past the door and telephones on a Tuesday, no system observed the cause. A rule was applied, and the rule decided which of those five events received the credit.
Different rules produce different answers from identical behaviour. That is not a defect of any particular platform; it is what happens when a single event is assigned to one of several possible antecedents. It becomes a problem only when nobody says which rule is being used, which is the normal situation.
Two consequences follow. First, comparing figures produced under different rules is meaningless, including comparing a platform's own reported figure with the clinic's enquiry count. Second, any change of rule produces a change in apparent performance with no change in reality, which is why criterion eight asks for a change log.
Attribution also depends on capture, which is why an unresolved enquiry response problem shows up here as an attribution problem. The most common distortion in clinic reporting is the treatment of unknowns. When a substantial proportion of enquiries have no identifiable source, and that proportion is quietly distributed across known channels, every channel appears to perform better than it did. Showing unknowns as their own line is uncomfortable, accurate, and the single highest-value change most clinics can make to their reporting.
How to score this checklist
Take last month's report. Add up the enquiries attributed to each channel and compare the total with the number of enquiries the clinic actually received. If the two differ and the difference is not explained on the report, score criterion six 0.
Look for a line labelled unknown, direct, or unattributed. If none exists, ask where those enquiries went. In most reports they have been distributed, and nobody decided that they should be.
Ask whoever produces the report which model it uses and what the window is. If the answer takes more than a sentence or invokes the platform's defaults without knowing what they are, score criteria one and seven low.
Ask the person who makes spending decisions to explain what the numbers credit. Criterion ten is scored by their answer, not by the report's sophistication. A simple model everybody understands is worth more than a defensible one nobody does.
For criterion nine, attempt to reproduce one channel's reported enquiry count from your own records. If you cannot, the number is unverifiable regardless of who supplied it.
Common scoring errors
Accepting a platform's self-reported conversions as the clinic's enquiry count. They count different things, usually with different windows.
Allocating unknowns proportionally. This is a common default and it manufactures precision from absence.
Excluding offline enquiries because they are hard to capture. Excluding the largest category to preserve tidiness produces a tidy report about a minority of the business.
Treating self-reported source as equivalent to tracked source. Both are useful and they have different failure modes. Report them separately.
Changing model to resolve a disagreement. If a channel looks weak, changing the rule until it looks strong is a way of stopping the argument rather than settling it.
What to do when attribution cannot answer the question
Attribution is frequently asked to answer a question it cannot: whether spending on a channel produced the enquiries credited to it. The only method that answers that question is an experiment, and clinics can run simple ones.
Stop a channel for a defined period and observe total enquiries, not channel enquiries. If total enquiries do not move, the channel was receiving credit for demand that existed anyway. Run the period long enough to cover the decision cycle, which for aesthetic procedures is often weeks rather than days, and account for seasonality.
This is blunt, it costs something, and it produces a clearer answer than any model. It also has the advantage of being intelligible to everybody at the clinic.
Where an experiment is not practical, hold the attribution model steady and watch direction rather than level. A stable model applied consistently will show real changes even if the absolute credit assigned to each channel is arguable.
Above all, keep the unknown line visible. A clinic that knows a third of its enquiries have no identifiable source is better informed than one whose report accounts for everything. Pair this with the measurement maturity assessment, which addresses the capture problems that produce unknowns in the first place.