Quick answer: what matters when advertising runs on AI?
In advertising and AI, the platform now decides who sees your ad, where and at what bid. What still decides results is what you feed it: a profitable offer, clean conversion data, a steady supply of new creative, honest measurement and clear limits. A founder who manages those five inputs gets more from AI than one who manages campaigns.
What AI already decides in your ad account
On Meta and Google, AI now makes most of the choices a media buyer used to make by hand. You set a goal and a budget. The system picks the people, the placements and the bids, and it increasingly edits the ad itself.
| Product | What the AI decides | What you still provide |
|---|---|---|
| Meta Advantage+ sales campaigns | Audience, placement and budget split | Conversion data, creative, a goal and a spend limit |
| Google Performance Max | Bids and placements across YouTube, Display, Search, Discover, Gmail and Maps | Assets, product feed, conversion goals and values |
| Google AI Max for Search | Which searches to match, the ad text and the landing page | Keywords, brand rules, pages worth sending people to |
Meta's help page on Advantage+ sales campaigns says its most advanced AI optimisations are enabled when a campaign uses Advantage+ audience, placement and budget together. Meta's own figure is a 9% average improvement in cost per conversion for advertisers who used these campaigns.
Google describes Performance Max as a goal-based campaign that reaches all Google Ads inventory from a single campaign. You supply assets, feeds and bidding goals, and Google AI bids in real time and combines your assets into ads.
Search is going the same way. In April 2026, Google announced that AI Max for Search was out of beta. It also said that from September, campaigns using Dynamic Search Ads, automatically created assets or campaign-level broad match would be upgraded to AI Max automatically.
If you ran Dynamic Search Ads or campaign-level broad match, look at your Search campaigns now. An automatic upgrade can change which searches you pay for and which pages people land on.
None of this is bad news. It means the place where a founder adds value has moved.
The 5 inputs that still decide results
When the platform controls the settings, results depend on what goes in. There are five inputs, and they are all business decisions, not ad-tech ones.
- Unit economics. What an order is really worth after cost, shipping, tax and returns. The AI cannot see this.
- Conversion data. The purchases and leads you report back. The AI learns only from these.
- Creative supply. New ads, in enough variety, every month. Creative is how you now steer who the system finds.
- Measurement. A way to know what ads truly caused, beyond the platform's own report.
- Guardrails. Budget limits, brand rules and legal duties the machine does not set for itself.
The rest of this guide takes them one at a time. If you want the industry view first, read what AI changes for brands, agencies and platforms.
A simple test shows whether a team has made the shift. Ask what they changed in the account last week. If the answer is a list of bid and audience edits, they are still doing the machine's job. If it is "two new creative angles, a fixed purchase event and a corrected profit target", they are doing yours.
Start with unit economics: the number AI cannot see
The AI optimises towards the order value you send it. It does not know your product cost, your shipping bill, the tax inside the price or the orders that come back. So the first job is to work out the return on ad spend (ROAS) at which you actually break even.
Three things make platform revenue higher than real revenue:
- Tax. Order value often includes GST in India, VAT in the UK, EU and Gulf, or sales tax in the US.
- Refused orders. With cash on delivery (COD), common in India, the Gulf and parts of South-East Asia, some parcels are refused at the door. These return-to-origin (RTO) orders were already counted as sales.
- Returns. A returned order still taught the system that this kind of buyer is good.
Turn your costs into a break-even ROAS
You need five numbers: order value with and without tax, product cost, shipping and fees, the share of orders delivered, and the cost of each failed delivery.
Worked example: a brand spending ₹40 lakh a month
Step 1: Profit per placed order, before ads
Example: the average order is ₹1,770 including 18% GST, which is ₹1,500 without tax. Product cost is ₹450, shipping ₹120, and payment and COD fees ₹30. That leaves ₹900 on each delivered order.
85% of orders are delivered. Each refused order costs ₹200 in two-way shipping.
Profit per placed order
(0.85 × ₹900) - (0.15 × ₹200) = ₹765 - ₹30 = ₹735Step 2: Break-even ROAS on the number the platform shows
₹735 is the most you can pay for an order without losing money. The platform reports ₹1,770 of revenue per order.
Break-even ROAS
₹1,770 ÷ ₹735 = 2.41xStep 3: Read last month honestly
The account spent ₹40,00,000 and the platform showed 2.8x, which is ₹1,12,00,000 of reported revenue and about 6,330 orders. At ₹735 each, that is about ₹46,50,000 before ads and roughly ₹6,50,000 of profit after them.
At 2.3x, a number many teams would call healthy, the same sum gives a loss of about ₹1,80,000.
Our rule of thumb: set the target ROAS 15-20% above break-even, so a normal bad week does not push you into loss. Here that means a target near 2.8x to 2.9x.
This is the idea ScaleWin is built on. It reads the store next to the ad accounts and judges every ad on profit after product cost, shipping, tax and returns.
Feed the machine: data and creative are the new targeting
With broad, automated targeting, two inputs tell the AI whom to look for: the conversions you send back and the ads you give it. Weak data makes it learn slowly. Narrow creative makes it find the same people again and again.
Data: make every sale count
Meta's Conversions API best practices ask you to send the same events from the browser pixel and from your server, so sales the browser misses still reach the system. On Shopify this is a data sharing setting, not a development project.
Volume matters as much as accuracy. Meta says an ad set usually leaves the learning phase after about 50 results in the week after its last significant edit. If your weekly purchases are spread over too many ad sets, none of them learns properly.
We cover the full set of checks in what AI needs from your pixel, CAPI and store.
Creative: variety is the lever
Meta's advice for Advantage+ sales campaigns is to provide a wide variety of diverse creative assets. In practice, variety means different reasons to buy, not ten versions of one image.
A workable monthly supply for an account spending ₹10 lakh to ₹1 crore:
- 3-4 new angles, such as a problem, a proof point, an offer and a founder story.
- 2-3 formats for each angle: a short video, a static image and a carousel.
- One new product or bundle story, so the system has something fresh for past buyers.
AI tools can now draft much of this. They still need your product truth and your brand's voice, as we explain in AI and advertising creativity.
Measure what AI cannot grade itself on
The platform's report is the AI marking its own homework. It is useful for comparing ads inside one platform. It is not proof of what advertising added to your business.
Two changes in 2026 make this more important:
- From 12 January 2026, Meta's Ads Insights API, which feeds most dashboards, stopped returning the 7-day view and 28-day view attribution windows. Dashboards built on them now show less.
- On 3 March 2026, Meta said click-through attribution would count link clicks only. Conversions after a like, share or save moved to a separate bucket called engage-through.
So a fall in reported conversions this year may be a change in counting, not in sales. Before you compare this quarter with last year, check that both periods use the same attribution setting.
In the same March post, Meta called incrementality experiments, such as its Conversion Lift test, the gold standard in measurement. A lift test holds ads back from a random group and compares their purchases with everyone else's.
For a founder, three numbers outside the platform are enough:
- Blended return: total store revenue divided by total ad spend, across all channels.
- Profit after ads: delivered, tax-exclusive revenue minus costs and ad spend.
- New customer share: how much of the month's revenue came from first-time buyers.
More on attribution in AI in advertising measurement after cookies and iOS.
Guardrails: budget, brand and the law
AI will spend whatever you allow, wherever it predicts a result. Guardrails are the limits you set so that speed does not become waste or risk.
- Budget. Use campaign spend limits and a monthly cap per platform. Raise budgets in steps, and decide the next step from profit, not from platform ROAS.
- Brand. Give the system rules: words you never use, claims you cannot make, pages it should not send traffic to. Google's AI Max has brand and location controls for this reason.
- Starved campaigns. Google notes that AI Max will not be effective in campaigns limited by budget. A system with no room to explore cannot learn.
- Law. Since 2026, AI-made people and scenes in ads carry disclosure duties in the EU, India and New York, and Meta and Google now label AI ads.
The legal side changes quickly. Our guide to AI advertising ethics and disclosure rules has a 4-question test for when to label an ad.
Change one big input at a time: budget, creative or structure. If you change all three in a week, nobody can tell which one moved the result, including the AI.
What most guides get wrong about advertising and AI
Most articles on this topic are either tool lists or predictions. They skip the part a founder needs: what to do differently on Monday. These are the claims to be careful with.
- "AI replaces your marketer or agency." It replaces bid and audience edits. Someone still has to own margin, creative, data and measurement.
- "Switch everything to automatic and results follow." Automation magnifies its inputs. With wrong order values or thin creative, it finds the wrong buyers faster.
- "A high platform ROAS means the AI is working." Not if the revenue includes tax and refused orders. Judge it against your break-even number.
- "More campaigns give more control." They split your results into pieces too small to learn from. Fewer, better-fed campaigns usually do more.
- "Conversions fell, so performance fell." In 2026 the counting rules changed twice. Check attribution settings first.
- "The tool decides, so the tool is responsible." The advertiser answers for claims, targeting and disclosure, whatever software wrote the ad.
A 30-day plan for a founder spending ₹10 lakh to ₹1 crore a month
You do not need a new tool to start. You need four weeks of putting the five inputs in order, then a short weekly habit.
- Week 1: numbers. Work out profit per placed order and break-even ROAS, as in the example above. Share the target with whoever runs the account.
- Week 2: data. Confirm purchases arrive from both browser and server, without duplicates. Compare one day of platform purchases with store orders.
- Week 3: structure and creative. Merge ad sets that get fewer than about 50 results a week. Brief 3-4 new creative angles.
- Week 4: measurement and limits. Set up the three founder numbers. Write down budget caps, brand rules and who approves AI-made creative.
After that, hold one 30-minute review a week with five fixed questions:
- What was profit after ads, and was it above target?
- Did store orders and platform purchases move together?
- Which new creative angles went live, and which earned more spend?
- Was anything edited that reset learning, and why?
- What is the one change for next week?
Festive periods such as Diwali, Ramadan and Black Friday raise ad costs for everyone. Plan budgets and creative for them a month ahead, and avoid daily edits once the sale starts.
For the day-to-day version of this routine, use our daily, weekly and monthly Meta ads checklist.
Read next in this series
Key takeaways
- Meta and Google now automate audience, placement, bidding and parts of the creative. Your control has moved from settings to inputs.
- Five inputs decide results: unit economics, conversion data, creative supply, measurement and guardrails.
- Work out break-even ROAS on delivered, tax-exclusive revenue before you judge any AI campaign. In our example it is 2.41x, not the 2x many teams assume.
- Platform reports changed twice in 2026, so compare periods carefully and run a lift test before big budget moves.
- A 30-minute weekly review with five fixed questions is enough for a founder. Daily edits usually hurt more than they help.
Advertising and ai: frequently asked questions
How is AI used in advertising today?
Do I still need a marketing team or agency if the platforms use AI?
What should a founder look at each week?
Why did my Meta conversions drop in 2026 when sales did not?
How much budget does AI advertising need to work?
Is Advantage+ or Performance Max better than manual campaigns?
What is the biggest risk when advertising runs on AI?
Sources
- Meta Business Help Centre: About Advantage+ sales campaigns
- Google Ads Help: About Performance Max campaigns
- Google: Dynamic Search Ads are upgrading to AI Max (April 2026)
- Google Ads Help: How AI Max for Search campaigns works
- 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 the learning phase
- Meta Business Help Centre: Best practices for Conversions API
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


