The Ecomm Analyst

Attribution tools are accurate enough to rank your channels correctly and not accurate enough to give you a true revenue number for any single one of them. That is the short answer. A well configured tool on a typical Shopify store will capture somewhere in the region of 80 to 95 percent of orders and will assign them to sources with real but bounded confidence. Anyone promising more than that is selling something.

The more useful framing is that accuracy is not one property. It splits into two questions that get confused constantly, and most disappointment with attribution tools comes from measuring one and complaining about the other.

What does accuracy actually mean for an attribution tool?

The first question is coverage. Of all the orders your store took last month, how many did the tool see and successfully tie to some source? This is measurable. Pull your order count from Shopify, pull the attributed order count from the tool, and compare. If a tool is showing you 780 orders and Shopify says 840, you have 93 percent coverage and a known gap.

The second question is assignment. Of the orders it did see, did it credit the right source? This is not measurable in the same way, because there is no answer key. Nobody knows with certainty which touchpoint caused a purchase. Every tool is applying a model, and the models differ.

Coverage problems are fixable. Assignment differences are structural. When someone tells me their attribution tool is inaccurate, I ask which one they mean, and about half the time they have not separated them.

Why do attribution tools disagree with Meta?

Because they are counting different things on purpose. Meta counts a conversion when someone who saw or clicked an ad later buys, within a window Meta chose, using data Meta holds, across devices Meta can link. Your attribution tool counts an order in your store and works backward using the signals that survived the trip to your site.

These are not competing measurements of the same quantity. Meta also has an obvious incentive in how it draws the boundaries, which is not a conspiracy so much as a fact about who built the ruler. A gap of 20 to 40 percent between platform reported revenue and store side attributed revenue is normal on Meta. It is not evidence that either number is broken.

What should worry you is a gap that changes shape suddenly without a corresponding change in your setup or spend. Stable disagreement is a modeling difference. Moving disagreement is usually a tracking problem.

How close should two attribution tools be?

Run two tools side by side and they will not match. If they rank your top four channels in the same order and their channel level revenue figures land within roughly 10 to 15 percent of each other, both are working correctly and the difference is modeling.

If one tool says email is your second best channel and the other puts it fifth, something is wrong, and it is usually a configuration difference in how discount codes or post purchase surveys are being credited rather than a defect in either product.

Can you test accuracy yourself?

You can test coverage directly, and you should, in the first week. Reconcile attributed orders against Shopify order counts for a full month. Anything under 90 percent is worth investigating before you trust a single report.

You cannot test assignment directly, but you can test it indirectly with a holdout. Turn a channel off in one geography for two weeks and watch what happens to total revenue there against a comparable region. That is the only method on this list that produces evidence rather than a model output, and it is why incrementality testing keeps coming back into fashion. I walked through the reconciliation process in more detail in what I check before I trust a new attribution tool’s numbers.

For context on what this looks like in practice, tools built specifically for store side attribution, including ThoughtMetric, which sponsors this blog, generally report against your actual order data rather than platform reported conversions, which is what makes the coverage check possible in the first place.

Frequently asked questions

Is server side tracking more accurate?

It improves coverage, sometimes substantially, by surviving ad blockers and browser restrictions that break client side scripts. It does not improve assignment. You will see more orders. You will not necessarily credit them better.

Do attribution tools get less accurate over time?

Coverage degrades if nobody maintains the setup. UTMs drift, new channels launch untagged, theme updates break scripts. The tool does not decay, the implementation does. A quarterly audit catches most of it.

Which attribution model is most accurate?

None of them, in the sense of being true. First touch overcredits discovery, last touch overcredits closing, and multi touch distributes credit according to assumptions somebody wrote down. Pick one, keep it consistent, and use holdouts when a decision is expensive enough to justify the test.