AI advertising · Founder guide

Advertising and AI: A Founder's Guide to What Actually Matters

The platforms now make most of the small decisions in your ad account. This guide shows the five decisions that are still yours, and how to make them well.

Ajay Kumar • •11 min read
Illustration of an ad next to a results chart

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.

ProductWhat the AI decidesWhat you still provide
Meta Advantage+ sales campaignsAudience, placement and budget splitConversion data, creative, a goal and a spend limit
Google Performance MaxBids and placements across YouTube, Display, Search, Discover, Gmail and MapsAssets, product feed, conversion goals and values
Google AI Max for SearchWhich searches to match, the ad text and the landing pageKeywords, 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.

Check what was switched on for you

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.

  1. Unit economics. What an order is really worth after cost, shipping, tax and returns. The AI cannot see this.
  2. Conversion data. The purchases and leads you report back. The AI learns only from these.
  3. Creative supply. New ads, in enough variety, every month. Creative is how you now steer who the system finds.
  4. Measurement. A way to know what ads truly caused, beyond the platform's own report.
  5. 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:

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 = ₹735
Step 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 = 2.41x
Step 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:

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:

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:

  1. Blended return: total store revenue divided by total ad spend, across all channels.
  2. Profit after ads: delivered, tax-exclusive revenue minus costs and ad spend.
  3. 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.

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.

Our rule of thumb

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.

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.

  1. 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.
  2. Week 2: data. Confirm purchases arrive from both browser and server, without duplicates. Compare one day of platform purchases with store orders.
  3. Week 3: structure and creative. Merge ad sets that get fewer than about 50 results a week. Brief 3-4 new creative angles.
  4. 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:

  1. What was profit after ads, and was it above target?
  2. Did store orders and platform purchases move together?
  3. Which new creative angles went live, and which earned more spend?
  4. Was anything edited that reset learning, and why?
  5. 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.

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?

AI now handles most delivery decisions on the big platforms. Meta Advantage+ picks audiences, placements and budget splits. Google Performance Max bids and places ads across all Google inventory, and AI Max for Search matches searches and adapts ad text. Advertisers supply goals, conversion data, creative and limits.

Do I still need a marketing team or agency if the platforms use AI?

Yes, but for different work. The platform makes bid and audience decisions. People are still needed to set profit targets, keep conversion data clean, produce new creative every month, check results outside the platform and make sure ads follow the law.

What should a founder look at each week?

Five things: profit after ads against target, whether store orders and platform purchases agree, which new creative went live, whether any edit reset the learning phase, and the one change planned for next week. This takes about 30 minutes and does not require opening every campaign.

Why did my Meta conversions drop in 2026 when sales did not?

Two reporting changes can explain it. From 12 January 2026 Meta stopped returning 7-day view and 28-day view attribution data through its reporting interface. In March 2026 it said click-through attribution would count link clicks only. Compare periods on the same attribution setting before judging performance.

How much budget does AI advertising need to work?

Enough for each ad set to learn. Meta says an ad set usually leaves the learning phase after about 50 results in the week after its last significant edit. Multiply your cost per result by 50 to get a weekly minimum for each ad set, then fund fewer ad sets properly.

Is Advantage+ or Performance Max better than manual campaigns?

Meta reports a 9% average improvement in cost per conversion for advertisers using Advantage+ sales campaigns, and Google reports gains for AI Max. These are the platforms' own averages. Test automation against your current setup, and judge it on profit after ads, not on platform ROAS alone.

What is the biggest risk when advertising runs on AI?

Optimising towards the wrong number. The system chases the order value you send, which often includes tax and orders that are later refused or returned. Without a break-even ROAS based on delivered, tax-exclusive revenue, a campaign can look strong while losing money.

Sources

Ajay Kumar
Founder, ScaleWin · 12+ years in performance marketing

Ajay is the founder of ScaleWin. He works hands-on with Meta Ads, Google Ads and SEO for Indian D2C and service brands, and built ScaleWin so that the daily work of a performance marketer - reading numbers, pausing losers, scaling winners - happens automatically.

Drafted with AI assistance. How these articles are written

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

Get marketing tips that grow your business

One practical email a week on ads, SEO and growth. No spam, no fluff. Unsubscribe anytime.