attribution

Attribution without false precision

A useful attribution model is an honest description of what the data can and cannot show.

Published 2026-09-17; topic coverage 2025-01

Illustrative scenario: not a client case study

Name the question

Decide whether the team is allocating budget, improving journeys, forecasting demand or auditing channels. Each question needs a different view; one model should not pretend to answer all of them.

Protect source capture

Use consistent campaign naming, tagged links, referral exclusions and a documented definition of conversion. Preserve the original source alongside later interactions where the system permits.

Connect online and offline carefully

Use a consented lead key to join CRM stages to marketing touchpoints. Document missing stages, duplicate records, delayed imports and sales interventions. Never turn an unobserved sale into a confidently attributed one.

Compare models as lenses

First touch, last touch, position based and data driven views answer different questions. Compare them with a simple holdout or budget change observation when feasible, and explain the assumptions.

Make decisions with ranges

Report direction, quality and uncertainty. A channel that creates few visible leads may support later demand, while a channel with many forms may need quality controls. Collect sales feedback before reallocating materially.

Choose the attribution question

A finance review may ask which investment deserves protection, while a channel owner may ask where a journey breaks. Write one question per report and reject fields that do not help answer it. For a Dubai service with phone enquiries and an Abu Dhabi B2B sale, a single last click table can conceal different buying paths.

Keep source history usable

Create a naming convention for campaigns, landing pages and offline imports before joining records. Preserve the first known source, later interactions and the human owner who changed a stage. Test duplicate leads and delayed sales updates with a small sample; otherwise a clean looking dashboard can turn missing data into false credit.

Use attribution as a decision aid

Compare several views, then state what would make you change budget or creative. If one channel appears weak because its influence happens before the measurable conversion, describe that as uncertainty rather than assigning invented value. Use controlled budget changes or carefully documented comparisons when feasible, and record what the method cannot observe.

Explain the missing middle

Many B2B journeys include private recommendations, forwarded documents, meetings, and untagged return visits that analytics cannot connect. Ask sales which influences they regularly hear about, but label that as reported evidence rather than measured credit. A leadership report becomes more useful when it shows the visible path, the plausible unseen influences, and the decision that remains safe despite the gap.

Make the operating decision explicit

A lead may read an article, return through search, speak to a colleague and finally complete a form on a phone. Calling one touchpoint “the source” makes a tidy report, not necessarily a truthful one. Name the person who owns the next decision, the evidence they are allowed to use, and the smallest reversible action available. For a UAE campaign this can include checking service coverage, language support, response capacity, consent configuration, or the distinction between an office and a delivery area. Write the assumption beside the decision instead of hiding it in a dashboard. When the assumption changes, record whether the campaign, page, audience, or measurement definition should change with it.

Interpret movement with care

A lead may read an article, return through search, speak to a colleague and finally complete a form on a phone. Calling one touchpoint “the source” makes a tidy report, not necessarily a truthful one. A performance signal is a prompt for investigation, not a conclusion by itself. Compare the relevant period, traffic mix, service availability, response process, creative version, and tracking status before explaining a rise or fall. Separate what the platform reports from what the business has verified downstream. If the sample is small or the journey is partly offline, describe the observation as directional and choose a check that could disprove the first explanation.

Leave a useful record

A lead may read an article, return through search, speak to a colleague and finally complete a form on a phone. Calling one touchpoint “the source” makes a tidy report, not necessarily a truthful one. Close the review with the change made, the reason, the owner, the date to revisit it, and the evidence that would support keeping or reversing it. Include rejected alternatives and unresolved uncertainty. This record helps a team serving Dubai, Abu Dhabi, or both avoid repeating a campaign decision simply because a new person cannot see its context. It also makes future Arabic and English updates easier to align without turning a translation into a new commercial promise.

What to take away

Start with the decision the model supports.
Use consistent naming and conversion definitions.
Join CRM data only with lawful, documented keys.
Show uncertainty instead of manufacturing exact credit.

Frequently asked questions

Which attribution model is correct?

None is universally correct. Choose the lens that fits the decision and disclose its assumptions and blind spots.

Can a CRM solve attribution?

A CRM can improve stage and revenue evidence, but it cannot observe every influence or repair inconsistent source capture.

Should last click be discarded?

It can remain a useful operational view, provided it is not presented as the whole customer journey.