First click, last click, data-driven: every attribution model is wrong differently. A practical measurement stack for small ecommerce that still makes good decisions.
A customer sees an Instagram ad on Tuesday, googles the brand on Thursday, clicks a search ad, leaves, comes back Saturday through an abandonment email and buys. Which channel gets the sale? Every one of the marketing attribution models answers differently, every answer is defensible, and none is true. Attribution isn't measurement; it's an opinion about credit, formalised.
This sits under the GA4 pillar and underwrites the budget logic in the paid media budget piece.
The models and their biases
| Model | Credit goes to | Systematic bias |
|---|---|---|
| Last click | Final touchpoint | Flatters closers: brand search, email, retargeting |
| First click | First touchpoint | Flatters openers: prospecting social, content |
| Linear | Split evenly | Flatters being present, rewards channel sprawl |
| Position-based | Ends weighted, middle shared | A compromise with two thumbs on the scale |
| Data-driven | Algorithmic, from conversion path patterns | Opaque, and limited to the touches the platform can see |
That last clause is the modern catch. Since iOS App Tracking Transparency arrived in 2021 and browsers tightened cookies, no platform sees the whole journey. Data-driven attribution is a sophisticated model of an incomplete dataset, and each ad platform additionally grades its own homework: add up the conversions Meta and Google each claim and you'll routinely exceed your actual order count.
The practical stack for a small store
Three layers, used for different decisions:
- Truth layer: MER and contribution margin after ad spend. Total revenue from your store platform against total marketing cost from your invoices. No model, no claims, can't be gamed. This decides whether overall spend rises or falls, as set out in the budget piece.
- Direction layer: platform metrics, read comparatively. Use Meta's numbers to compare Meta campaigns against each other, Google's against Google's. Never across platforms, and never as gospel revenue.
- Honesty layer: cheap incrementality checks. Post-purchase survey ("how did you hear about us", one question, surprisingly clarifying), geo or time-based holdouts (pause a channel in one region or fortnight and watch the truth layer), and a sense check of branded search volume against prospecting spend.
A worked example of the stack catching a lie: a retargeting campaign reports a spectacular return at the platform layer. A two-week pause shows total revenue barely moves. The campaign was claiming credit for customers already coming back. The truth layer ruled; the platform layer was redecorating.
Choosing a default model anyway
Reports still need a setting. GA4 defaults to data-driven attribution and that's fine as the comparative lens, with one discipline: pick it and stop switching. Changing models between months makes trends unreadable, and the trend is the only part of attribution that's reliably informative.
What to read next
- GA4 ecommerce tracking: the pillar this all sits under
- Paid media budget: deciding what overall spend should be
- Server-side tracking: recovering the signal platforms lost
Frequently asked questions
Which attribution model should a small business use?
Data-driven or position-based as the default reporting lens, held constant. But govern budget decisions with MER and margin, and settle disputes with holdout tests, not model swaps.
Why do Meta and Google both claim the same sales?
Each platform attributes any conversion it touched within its window. Both touched the journey, both claim it, and the sum exceeds reality. That's why platform numbers compare campaigns, not channels.
What is incrementality testing?
Measuring what happens to total revenue when you remove or add a channel, via geo splits or pauses. It's the closest thing to ground truth a small business can run, and it's nearly free.
How Qwrki fits
Qwrki is the operating layer that runs ecommerce delivery for small businesses, so attribution stops being a monthly argument and becomes a standing read. We wire the truth layer, the platform layer and the holdout checks into one place, then hold the model constant so the trend stays legible. Book a call and we'll walk through your current numbers and where the credit is being double-counted.
You may also like
- analytics
GA4 ecommerce tracking: the events that matter and the discrepancies that don't.
The GA4 ecommerce events that actually matter, the setup checklist, and why GA4 never matches Shopify. A practical guide for store owners.
4 min - analytics
Server-side tracking: what it fixes and what it doesn't.
Server-side tracking and the Conversions API in plain English: what signal it recovers, what it can't, the real costs, and who actually needs it.
4 min