Shopee Ads · decision
How to tell if a Shopee ad is working — and why ROAS is not enough.
Shopee Ads reported more attributed orders than the valid orders recorded in my operational data. The question was not which figure looked better, but whether those numbers could be compared at all.
The discrepancy was recorded in September 2026. Differences in attribution, time window and order status needed to be understood before drawing conclusions about advertising returns.
Why the same-looking numbers can mean different things
Ad reporting mainly asks: what results does the advertising platform attribute to these campaigns?
The order view asks something else: which orders were recorded by the store, within a particular period and under particular status rules?
Those questions sound alike but can use different definitions. Attribution can follow a different time window from order creation. Orders may be filtered by status. Different kinds of ad attribution may also be grouped together.
These are possible explanations, not an established root cause of the discrepancy I observed. That would require a proper reconciliation of definitions, windows and records.
Where ROAS fits in
ROAS means return on ad spend:
ROAS = sales attributed to ads ÷ ad spend
If the platform shows ROAS 2, it attributes two units of sales to each unit spent on ads. It does not mean one unit of profit.
Material, production, fees, discounts, shipping and other expenses still matter. And ROAS alone cannot answer how many purchases would have happened without the ad.
In a 3D-printing studio there is an additional constraint: more orders may mean more material, printer hours and packaging work.
I stopped the conclusion, not the campaign
With two datasets that did not line up, I could have chosen whichever number supported a more attractive recommendation.
Instead, the documented decision was not to draw an economic conclusion until the definitions and periods could be reconciled.
There is no documented ad pause, budget increase or recovered profit from this specific episode. What I have is a real interpretation problem and a decision not to pretend the data were more conclusive than they were.
How to tell whether an ad is working
The experience gave me a more disciplined way to investigate a campaign:
- Decide what you are asking. Did the ad generate orders? Were those orders profitable? Did the ad create sales that would not otherwise exist? Those are different questions.
- Check what each dataset measures. Align dates, attribution types and order statuses before comparing figures.
- Account for costs. The full sale price is not the amount available to pay for advertising after platform fees, production and other expenses.
- Check physical capacity. Growing demand is not automatically valuable if stock or production capacity cannot support it.
- Keep uncertainty visible. Without comparable records or a credible incremental-sales test, you cannot confidently attribute all observed sales to advertising.
Groks, the AI operator I use with my store, can help gather signals and suggest investigations. It should not fill data gaps with plausible guesses.
What the episode actually established
I did not find a secret number that replaces ROAS. Nor did I prove Shopee's attribution was wrong. I found that combining reports does not automatically make them comparable.
Before changing ad spend on the strength of ROAS, first establish what the reports measure, what the sale is worth after costs, and how much confidence the evidence actually supports.
The discrepancy between ad attribution and synchronized orders was observed, but its exact cause was not established. The documented decision was to withhold economic conclusions from incomparable measures, not a proven increase in sales.