Quick answer: how do you measure results from AI advertising?
Measure AI advertising on profit after ads, not on the return on ad spend the platform reports. Remove tax, undelivered orders and returns from reported revenue, subtract product and delivery costs, then subtract ad spend. Track that number weekly, next to a blended return across all channels and the cost of each new customer.
Platform ROAS grades the ad. Profit grades the business
Return on ad spend (ROAS) is reported revenue divided by ad spend. It is a fine way to compare two ads on the same platform. It is a poor way to decide whether advertising is working.
The reason is simple. The platform only knows the order value your store sent it. It does not know:
- how much of that value is tax,
- which orders were never delivered,
- which came back as returns,
- what the product and the delivery cost you.
AI campaigns make this gap more important, not less. An automated campaign optimises hard towards the number it is given. If that number is inflated, the system gets very good at finding orders that look valuable and are not.
This article covers the measurement side only. For the full set of inputs a founder controls, see our guide to advertising and AI.
The 4 numbers to track instead
You do not need a data team. You need four numbers, worked out the same way every week.
| Number | How to work it out | What it tells you |
|---|---|---|
| Break-even ROAS | Reported revenue per order ÷ profit per placed order before ads | The platform ROAS below which you lose money |
| Profit after ads | Delivered, tax-exclusive revenue minus costs minus ad spend | Whether advertising made money this week |
| Blended return | Total store revenue ÷ total ad spend, all channels | Whether the whole mix is healthy, with no double counting |
| Cost per new customer | Total ad spend ÷ first-time buyers | Whether growth is getting cheaper or dearer |
The first number is a target. The other three are results. Set the target once a month, and read the results once a week.
Blended return matters most when you run several platforms. Each one claims the same customers, as we explain in AI advertising across channels.
From platform ROAS to profit: 5 corrections in order
Apply these corrections in this order. Each one removes money that was never yours.
- Tax. Remove GST in India, VAT in the UK, EU and Gulf, or sales tax in the US from reported revenue.
- Undelivered orders. Remove orders that were cancelled or refused. With cash on delivery (COD), refused parcels come back as return to origin (RTO).
- Returns. Remove orders that were delivered and then sent back.
- Costs. Subtract product cost, shipping both ways, and payment or COD fees.
- Ad spend. Subtract what you paid the platforms, including tax on the ad bill if you cannot claim it back.
What is left is profit after ads. It will be far smaller than the number in the dashboard. That is normal. The aim is not a big number. The aim is a true one.
See the corrections on a real-sized account
The example below uses round figures so you can follow each step with a calculator.
Worked example: a brand spending ₹60 lakh a month
Step 1: Start from what the platforms report
Example: ad spend is ₹60,00,000 and the dashboards show ₹2,10,00,000 of revenue. That is a ROAS of 3.5x, which looks healthy.
Step 2: Remove tax, undelivered orders and returns
Without 18% GST, revenue is ₹1,77,96,610. 82% of orders are delivered, which leaves ₹1,45,93,220. 4% of delivered orders are returned, which leaves ₹1,40,09,492.
Real revenue for each ₹1 the platform reports
(1 ÷ 1.18) × 0.82 × 0.96 = ₹0.667Step 3: Subtract costs and ad spend
After product cost, shipping and fees, the brand keeps 45% of real revenue, which is ₹63,04,271. Subtract ₹60,00,000 of ad spend.
Profit after ads
₹63,04,271 - ₹60,00,000 = ₹3,04,271Step 4: Find the break-even point
Each ₹1 of reported revenue is worth ₹0.667 × 45% = ₹0.30 of margin. So the account breaks even at 1 ÷ 0.30 = 3.33x. At 3.0x, the same account loses about ₹5,96,000 a month.
Our rule of thumb: set the target 15-20% above break-even. For this brand, that is about 3.8x to 4.0x, not the 3x many teams aim for by habit.
Is the AI adding sales, or claiming them?
Profit after ads tells you whether the month made money. It does not tell you how much of that the ads caused. Some buyers would have ordered anyway.
AI campaigns are good at finding people who are about to buy, including people who already know you. That raises reported ROAS without adding many new sales.
Meta itself says this. In a post on ad measurement in March 2026, it called incrementality experiments, such as its Conversion Lift test, the gold standard. A lift test holds ads back from a random group and compares what they buy with everyone else.
If you cannot run a formal test, three simple checks help:
- New customer share. If most ad-driven orders come from past buyers, the campaign is harvesting, not growing.
- A regional pause. Stop ads in one state or city for two weeks and compare its store sales with similar regions.
- Brand search. If brand searches fall when you cut video and social ads, those ads were doing more than the report showed.
More methods are in AI in advertising measurement after cookies and iOS.
Why this year's numbers do not match last year's
Before you compare periods, check that both were counted the same way. In 2026, Meta changed the counting twice.
- From 12 January 2026, Meta's Ads Insights API no longer returns the 7-day view and 28-day view attribution windows. Reports and dashboards that used them now show fewer conversions.
- In March 2026, Meta said click-through attribution would count link clicks only. Conversions after a like, share or save are now reported separately as engage-through.
Neither change altered what customers bought. Both changed what the report credits to ads. A team that missed them may think performance fell, cut budget, and hurt real sales.
When reported conversions move sharply but store orders do not, look at attribution settings first. Store orders are the number nobody recounts.
Clean signal matters here too. If events reach the platform late or twice, every report is off. The checks are in what AI needs from your pixel, CAPI and store.
What most guides get wrong about measuring AI ad results
Most advice still treats the platform's report as the result. These are the habits to drop.
- "A ROAS of 3x or 4x is good." There is no universal good ROAS. It depends on your margin, tax, delivery rate and returns.
- "The AI campaign beat the manual one, so it works." Platform averages, such as the 9% lower cost per conversion Meta reports for Advantage+ sales campaigns, compare ads with ads. They say nothing about your profit.
- "Add up revenue from every platform." The total will exceed your real sales. Use the store's own total.
- "Daily ROAS tells you what to change." Daily numbers are noisy, and delivery data arrives days later. Decide weekly.
- "Attribution tools solve this." They help split credit between ads. They cannot know your costs or your refused parcels.
A weekly profit review that takes 20 minutes
Hold one short review a week, on the same day, with the same sheet. It needs five lines.
- Store revenue, delivered and without tax, for the last full week.
- Total ad spend across all platforms.
- Profit after ads, using your cost percentages.
- Blended return and cost per new customer, next to the last four weeks.
- One decision: raise, hold or cut, and on which channel.
Use last week's delivered orders, not this week's placed orders. In markets with COD, an order is not revenue until it is accepted at the door. During Diwali, Ramadan or Black Friday, expect delivery times and return rates to shift, and wait an extra week before you judge the sale.
Change budgets only when profit after ads has moved the same way for two weeks in a row. One good or bad week is usually noise.
Pulling these five lines by hand is the slow part. ScaleWin connects the ad accounts and the store, and reports profit for each ad after product cost, shipping, tax and returns.
The daily and monthly checks around this review are in our Meta ads optimization checklist.
Read next in this series
Key takeaways
- Platform ROAS compares ads with each other. It does not tell you whether the business made money.
- Four numbers are enough: break-even ROAS, profit after ads, blended return and cost per new customer.
- Correct reported revenue in a fixed order: tax, undelivered orders, returns, costs, then ad spend.
- In our example, 3.5x on the dashboard is only ₹3 lakh of profit on ₹60 lakh of spend, and 3.0x is a loss of about ₹6 lakh.
- Meta changed how it counts conversions twice in 2026. Check attribution settings before you compare periods.
Advertising and ai: frequently asked questions
What is a good ROAS for AI advertising?
How do I calculate profit after ads?
What is blended return, and why use it?
Why did my Meta ROAS drop in 2026?
How can I tell if AI campaigns are bringing new customers?
Should I include tax when I work out ROAS?
Sources
- Meta for Business: Simplifying ad measurement for a social-first world (3 March 2026)
- Meta for Developers: Ads Insights API metric availability updates (16 October 2025)
- Meta Business Help Centre: About Advantage+ sales campaigns
Let ScaleWin read your ad data every day
ScaleWin checks your Meta and Google Ads daily, judges every ad on profit and tells you exactly what to change.
Start free, no card



