Ecommerce Analytics · focused work
Ecommerce attribution reporting
Every advertising platform reports the sales it believes it caused, and adding those reports together produces more revenue than the store actually took. Attribution reporting is the work of building one view the business argues from, deciding which number answers which question, and testing the claims that carry enough budget to be worth testing.
This is the right work if
- Adding up the conversions each platform claims produces more orders than the store received
- Budget discussions run on different numbers depending on who prepared the slide
- Nobody has agreed which system is the authority on revenue
- Lookback windows differ across platforms and no report accounts for the difference
- New customer revenue is not separated from repeat revenue anywhere
What it is
The three views and what each is for
The store's own order data is the fixed total. It is the only number that is not a model, and it is the denominator everything else is compared against. Platform reported conversions describe what each channel claims using its own lookback window, its own rules about which interactions count and its own modelling for the gaps. A cross-channel analytics model gives one consistent set of rules across every channel, which makes channels comparable to each other while agreeing with none of them. None of the three is correct in isolation, and the mistake most stores make is choosing one and treating the others as errors.
The reporting work defines which view is used for which decision, aligns windows and definitions so differences stop being mysterious, and builds a report that shows all three next to spend. Around that sit the measures that do not depend on tracking at all: blended efficiency across total revenue and total marketing cost, contribution margin after advertising, the split between first orders and repeat orders, a self-reported question after checkout, and holdout tests for the channels large enough to justify one. The output is not a perfect model. It is a small set of numbers, an agreed cadence, and a rule for what each one triggers.
How it is done
The work, in order
What changes
- The business plans from one report instead of several conflicting ones
- Every claimed figure sits beside the store's own revenue total
- Acquisition cost is measured against new customers rather than all orders
- Channel weight is corrected by tests instead of by self-reporting
Fix the total before anything else
Take the store's own orders, net of cancellations and refunds, as the authority on revenue for the period. Every channel claim is then expressed as a share of a known total rather than as a free-standing figure nobody can place.
Document each platform's rules
Record the lookback window, the interaction types counted and the modelling each platform applies, and put that summary at the top of the report. Most arguments about attribution end quickly once everyone can see that two numbers were measuring different things.
Build one report with all three views
Weekly, showing store revenue, platform-claimed conversions, the analytics model, spend per channel, and blended efficiency for the business as a whole. Same definitions, same period boundaries and the same currency treatment every week.
Separate first orders from repeat
Split revenue and orders by whether the customer had bought before, so acquisition spend is measured against acquisitions. A channel that looks efficient on total revenue can be selling almost entirely to existing customers, which is a different business decision.
Ask the customer at checkout
Add a short question about how they first heard of the store. The answers are imprecise and directionally useful, and they are the only visibility most stores get into podcasts, word of mouth, print and anything else that leaves no click behind.
Test the channels worth testing
For the largest spend, run a geographic holdout or a scheduled pause long enough to cover the store's normal purchase cycle, then compare total revenue rather than campaign revenue. Use what the test shows to adjust how much weight the routine reports carry.
Which number should the business treat as the truth?
The store's own revenue for totals, and the analytics model for comparing channels against each other, with platform figures used to manage inside each platform. Naming the role of each one, in writing, prevents most of the disagreement that attribution debates are actually about.
Is a data-driven model better than last click?
It usually distributes credit more sensibly across a longer journey, and it is harder to explain and impossible to audit. Many stores run both, using the modelled view for planning and last click as a stable reference that behaves predictably when the modelled view moves for reasons nobody can see.
Are post-purchase survey answers reliable enough to use?
Not as precise numbers, and yes as a directional signal. People misremember and skip the question, so treat the results as trends over time rather than percentages of truth. Their real value is revealing channels no tracking captures, which no amount of tagging work would have surfaced.
How often should the attribution setup be reviewed?
The report runs weekly, and the underlying assumptions deserve a review each quarter or after any platform changes its measurement. Windows, models and consent behaviour all shift, and a report built on last year's rules will keep producing confident numbers that quietly stopped being comparable.
Can better tracking solve attribution on its own?
No. Tracking improves the data available; it cannot tell you what would have happened without the ad, which is the actual question behind every budget decision. That answer only comes from holding a channel back and measuring the difference, which is why testing sits inside this work rather than beside it.
Part of
Related work and answers
Find the leak.
A free Growth Analysis ranks what your store should fix first, by revenue at stake.