The Ecomm Analyst

I look at a lot of attribution setups on stores in the one to ten million range, and the same handful of mistakes come up again and again. None of them require a data team to fix. Most are about discipline and mindset more than tooling.

Treating platform ROAS as truth

The most common one is reading the number Meta or TikTok reports as if it were real revenue. It is the platform grading its own homework on its own attribution window, and it is almost always inflated. Use it to compare campaigns inside one account, not to decide what the business actually made. The day you stop quoting platform ROAS as your real return is the day your planning gets more honest.

Living in last click

Plenty of brands still run on last-click attribution, often without realizing it, because that is what the default report in their analytics shows. Last click hands all the credit to the final touch, usually a branded search or a retargeting ad, and starves everything that created the demand in the first place. It makes your top-of-funnel look worthless and your bottom-of-funnel look like genius, and people end up cutting the exact spend that was feeding the machine.

Over-instrumenting before there is volume

Smaller stores often build elaborate tracking, server-side tagging, multi-touch dashboards, cohort reports, before they have the order volume to read any of it. At a few hundred orders a month the patterns are mostly noise, and a detailed dashboard just gives you more confident ways to be wrong. Get the volume first, then add resolution.

Confusing dashboard correlation with incrementality

A channel showing up next to conversions in your reports does not mean it caused them. Branded search and retargeting are the classic offenders, they sit close to the purchase and soak up credit for demand other channels created. The only way to know what a channel actually adds is to test it, turn it off in a region or for a period and watch what happens to total orders. Correlation in a dashboard is a hint, not proof.

Ignoring the new versus returning split

A comfortable blended return can completely hide weak acquisition when repeat customers are carrying the number. I have seen brands celebrate a healthy blended figure while their actual cost to acquire a new customer had quietly gone underwater. Always look at new customer revenue separately. That is the part marketing spend is supposed to be buying.

Changing the model mid-analysis

Switching attribution windows or models partway through a comparison is a quiet way to fool yourself. A seven-day window in one report and a twenty-eight-day window in another are not comparable, and a shift that looks like a performance change is often just the measurement moving under your feet. Pick a model, hold it constant, and only change it deliberately and with a note to yourself about when and why.

Reconciling to the penny

The last one is chasing a single attribution number that ties perfectly to Shopify revenue. It does not exist, and the week you spend trying to build it is wasted. The platforms inflate, your store under-attributes, and the truth sits in a range. Accept the range, anchor planning to real Shopify revenue, and use the platform numbers for steering. Operators who make peace with imperfect attribution make faster and better decisions than the ones still trying to force every dashboard to agree.

Fixing these is mostly about restraint. Quote the honest number, hold your models steady, test before you believe, and do not build more measurement than your volume can support. The brands that get attribution right at this size are not the ones with the fanciest stack, they are the ones who stopped lying to themselves about what the numbers mean.