Add up the conversions each platform claims and you will comfortably exceed the number of sales you actually made. Nobody is lying. They are each answering a slightly different question.
Key takeaways
- Every channel reports the conversions it touched, so totals across platforms always overcount.
- Last-click systematically overvalues the closing channel and undervalues everything that created the demand.
- No attribution model is correct. Pick one, use it consistently, and judge changes rather than absolutes.
- When budgets are large enough, a holdout test beats every attribution model because it measures cause instead of correlation.
On this page
Why the numbers never reconcile
Each platform counts conversions it can claim a role in, using its own rules and its own window. A single purchase can therefore appear in three reports simultaneously, and each report is internally consistent.
Consider a customer who sees a social ad on Monday, searches your brand on Wednesday, clicks an email on Friday and buys. The social platform claims it because of the view. Search claims it because of the click. Email claims it because it was last. Your accounts record one sale.
Privacy changes made this worse rather than better. Shorter cookie lifetimes, tracking prevention and consent refusals all mean more journeys are partly invisible, and platforms increasingly fill the gaps with modelled estimates rather than observed events.
The models, and what each one distorts
Every model is a rule for splitting credit, and every rule flatters some channels at the expense of others. Knowing which distortion you have chosen matters more than choosing the theoretically best one.
| Model | Gives credit to | What it distorts |
|---|---|---|
| Last click | The final touch | Overrates brand search and email |
| First click | The initial touch | Overrates awareness channels |
| Linear | Every touch equally | Treats a banner view as a sales call |
| Time decay | Recent touches more | Still favours the closing channel |
| Position based | First and last most | Arbitrary weighting, but a defensible one |
| Data driven | Modelled contribution | Opaque, and needs high volume to work |
Last click remains the default in most setups and is the worst offender in practice. It tends to conclude that brand search is your best channel, which is a little like concluding that the cashier sold the product.
Choosing one and living with it
Pick the model that matches the decision you are making, then leave it alone. Switching models mid-campaign makes performance appear to change when only the accounting did.
- Short, simple journeys and a small budget. Last click is fine. The distortion is real but small when most people convert in one or two visits.
- Longer journeys, several active channels. Position based is the pragmatic default, it credits the discovery and the close without pretending a view equals a click.
- High volume, one platform doing most of the spend. Data driven is worth using, provided you accept you cannot audit it.
- Considered purchases over weeks or months. No click-based model will cope. Ask new customers how they found you; the answers are messy but they are causal.
Whatever you choose, record the date you chose it. Attribution changes are the single most common explanation for a channel that appears to have collapsed or doubled overnight.
The question attribution cannot answer
Attribution tells you which channels were present at a sale. It cannot tell you which sales would not have happened otherwise. That second question, incrementality, is the one that should decide budgets.
The classic illustration is branded search. It converts superbly under every model, because people searching your name were already going to buy. Some of that spend is genuinely defensive; much of it is paying for traffic you had already earned.
The test is simple and uncomfortable: turn a channel off in one region, or for a fixed period, and watch total sales rather than that channel’s reported sales. If overall revenue holds steady, the channel was taking credit rather than creating demand.
Attribution is a filing system for credit. Incrementality is a question about cause. Only one of them should set a budget.
Serhii Yelbaiev, Well Web Marketing
A workable approach for a small budget
Below roughly a hundred conversions a month, sophisticated attribution is unaffordable noise. Something much simpler will serve you better.
- Track every channel in one analytics property with consistent campaign tagging, so at least the comparison is like for like.
- Add a “how did you hear about us?” field to your main form. Self-reported data is imprecise but it captures word of mouth and offline sources that no platform can see.
- Judge each channel against its own history, not against other channels.
- Change one substantial thing at a time and give it long enough to clear the noise.
- Sanity-check everything against total revenue. If channel reports improve and revenue does not, the reports are wrong.
Frequently asked questions
None of them. They are competing conventions for dividing credit, not measurements. The useful question is which distortion you can live with given how your customers actually buy.
Ad platforms count view-through conversions and use their own attribution window, while analytics typically credits the last click within the session. Both are internally consistent and will never agree.
Rarely below significant spend. The tools are good, but they need volume to say anything reliable, and at low volume they add confidence without adding accuracy.
Ask. A single self-reported source field on your enquiry form routinely reveals that the largest channel was never in any dashboard.
Sources and method
Model definitions follow standard digital marketing attribution conventions. Observations on cross-platform overcounting and holdout testing come from our own client reporting. Last reviewed February 2026.