An attribution model is the rule that decides which marketing touchpoint gets credit for a sale when a customer interacted with several before buying. Choose a different attribution model and the same sale is credited to a different channel, which is why the choice changes where budgets go.
Attribution models exist because of one awkward fact: add up the conversions your platforms report and you will usually find they sold more than you did. This is not fraud. It is what happens when several systems each claim credit for the same customer.
Understanding why takes ten minutes and saves a lot of bad decisions.
What each attribution model does
A rule for assigning credit when a customer touched several things before buying. Someone sees an Instagram ad, reads a blog post, searches your brand, clicks a Google ad, then buys. Who gets the sale?
- Last click: all credit to the final touch. Systematically overrates brand search and retargeting.
- First click: all credit to the first. Overrates awareness channels.
- Linear: split evenly. Simple, and treats a decisive touch the same as an incidental one.
- Time decay: more credit nearer the sale. Reasonable default for short cycles.
- Data-driven: the platform decides using its own model. Usually the best available and unavoidably self-interested.
The same sale under each attribution model
Say a customer sees an Instagram ad, later reads one of your blog posts from organic search, then clicks a Google brand-search ad and buys for 2,000. Here is how much of that 2,000 each model gives to the Instagram ad (social), the organic blog visit and the brand-search click:
| Model | Social | Blog | Search |
|---|---|---|---|
| Last click | 0 | 0 | 2,000 |
| First click | 2,000 | 0 | 0 |
| Linear | 667 | 667 | 667 |
| Time decay | ≈400 | ≈600 | ≈1,000 |
Under last click the Instagram campaign looks useless and brand search looks brilliant, so budget drifts away from the ad that found the customer. That is the most common way an attribution model quietly shrinks a business.
In practice the choice is narrower than the table suggests. Since 2023, Google Analytics 4 and Google Ads offer only data-driven and last-click attribution, and Meta reports on its own click and view windows. For most businesses the real question is how far to trust each platform's version.
Why every platform over-claims
Each platform only sees its own touchpoints, and each applies its own window. Meta counts a conversion within its window; Google counts the same one within its; your email tool counts it too. Nobody is lying, and the total is still wrong.
Add view-through conversions — credited when someone saw but did not click — and the overlap grows further.
Platform numbers are for deciding what to do inside that platform. They are not for deciding how much to spend overall.
What small businesses should actually do
1. Use blended for budget decisions
Total marketing spend divided by new customers, monthly. It is impossible to game because it does not rely on any platform's claim. If blended cost per customer is under your ceiling, you are fine — regardless of what the dashboards argue among themselves.
2. Use platform numbers only within a platform
Comparing two Meta campaigns using Meta's numbers is valid. Comparing a Meta campaign against a Google campaign using each platform's own numbers is not.
3. Ask the customer
One optional question on your form or after purchase: how did you hear about us? Messy, unscientific, and consistently more informative than any model. It also catches the channels no pixel can see — word of mouth, a WhatsApp forward, an AI assistant's recommendation.
4. Run holdouts when it matters
The only genuinely reliable test available to a small business: turn a channel off for two weeks and see whether total revenue moves. Crude, slightly expensive, and it settles arguments that models cannot.
Improve the input before debating the model
Most attribution problems are data problems wearing a theory costume:
- Server-side conversions recover a meaningful share of events browser tracking loses.
- UTM parameters on every link, consistently, so your own records can tell you what the platforms cannot.
- Send qualified-lead status back so platforms optimise towards customers rather than form-fillers.
That is most of what tracking work is, and it beats picking a cleverer model every time.
The short version
Blended for how much to spend. Platform numbers for what to do inside a platform. A survey question for what the numbers miss. A holdout when it really matters. That is a complete attribution stack for a business without a data team. If you would rather have it set up and read for you, it is part of how our performance marketing team runs every account.
To see how much the platforms are over-claiming on your account, use the blended ROAS and MER calculator.
Questions people ask
What is an attribution model?
A rule for deciding which touchpoint gets credit for a sale. Last-click gives it all to the final interaction, first-click to the first, and data-driven models split it across the path. None of them is true — each is a different way of dividing the same conversion.
Why do my platforms report more sales than I made?
Because each one claims any conversion it can connect to, using its own rules, without knowing about the others. Two platforms touching the same customer will both count that customer. The overlap is normal and grows as you add channels.
Which attribution model should a small business use?
Spend less time choosing one and more time on a blended number: total revenue or total leads against total spend. It cannot be gamed by any platform and it matches what the bank account did.
How do I settle an argument about attribution?
Turn a channel off for a defined period and watch what happens to total business, not to the platform’s own number. An incrementality test answers in a fortnight what a model debate will not answer at all.
What is the best attribution model?
For most small businesses: the platform’s data-driven model for decisions inside that platform, and blended cost per customer from your own records for budget decisions. No single model is accurate on its own.
Which attribution model does GA4 use?
Data-driven attribution by default, with last click as the alternative. Google removed the first-click, linear, time-decay and position-based models from GA4 and Google Ads in 2023.
Want help setting it up? Tell us what you are running.