Which attribution model should an ecommerce store use?

An attribution model takes the touchpoints a system observed and applies a rule for splitting the order between them. Last click gives everything to the final one. A data-driven approach compares converting and non-converting paths to estimate each touchpoint's contribution. Both are arithmetic performed on an incomplete record.

The record is incomplete for reasons no model can fix. Consent choices remove sessions. Browser restrictions shorten the life of identifiers. People switch devices, read an email on a phone and buy on a laptop, see a video with no click at all, or hear about you from a friend. None of that is in the data, so none of it can be credited.

That is why changing model changes the report and changes nothing about the business. Moving from last click to a data-driven model will make upper-funnel channels look better and search look worse, and no revenue moved.

So the question is not which model is right. It is which model you will use consistently, and which decisions you will refuse to make with it.

In analytics, the practical choice is between a data-driven model and last click. Data-driven spreads credit across the observed path, which is closer to how buying actually works and makes assisted channels visible. Last click is simpler, more stable, and easier to explain to people who are not going to read a methodology note.

Use whichever you choose everywhere, including in the reports your team looks at weekly. The cost of switching between models depending on the argument being made is that nobody trusts any of the numbers.

In the advertising platforms, you get each platform's own view of its own advertising, with its own windows and its own inclusion of view-based credit. Those numbers are built to steer bidding, and they do that job well. They are not comparable across platforms and they will always sum to more than your actual revenue.

In your store, you have the truth about orders and revenue and nothing about what caused them. That is the ledger, and it is where profit decisions belong.

Three layers, three purposes. Platform data steers campaign-level decisions: which creative, which product group, which target. It is the signal the bidding uses, so aligning it to your accounting usually makes the bidding worse.

Analytics steers behavioural decisions: which landing pages work, where people drop out, how channels assist each other, what the path to purchase looks like. This is where a data-driven model earns its place, because it shows the contribution of channels that rarely close the sale.

Your store plus a test steers budget. The question of whether to spend more on a channel is a causal question, and the only honest answer comes from withholding the spend somewhere and comparing. Everything else is a description of what was recorded.

Sitting over all of it, marketing efficiency ratio gives you a single check: total revenue against total marketing spend. When channel reports improve and that ratio does not, credit has moved between reports rather than revenue arriving in the business.

Most attribution disputes are really data quality disputes. Missing campaign parameters on email and social links, a payment provider appearing as a referral source, a checkout on another domain without cross-domain measurement, or two purchase tags firing: each of these distorts every model equally.

So tag every outbound link you control with consistent campaign parameters, agree a naming convention and enforce it. This single piece of housekeeping usually changes the reports more than a model change does.

Set the attribution windows deliberately and keep them the same across systems where you can. A comparison between a platform using one window and analytics using another is not a comparison at all.

Then write down which number governs which decision, and put it where the reports live. The point of the stack is not to make the numbers agree. It is to stop people using one number for a decision it cannot answer.