AI Advertising · Case Studies

Artificial Intelligence Advertising Case Studies: What Brands Actually Got

Every famous AI ad result, sorted by how strong its evidence is. You also get a method to re-run the claims in your own account, adjusted for tax, COD and returns.

Ajay Kumar • •10 min read
Magnifying glass examining a stack of graded advertising case-study cards next to a balance scale

Quick answer: what results have brands actually got from AI advertising?

Real AI advertising results fall into three groups. Platform averages show modest gains: Meta's Help Center is widely cited as reporting about 9% lower cost per conversion, and Google reports 14% more conversions from AI Max. Brand stunts report reach or production savings. Vendor cases report large lifts without control groups. Treat each as a hypothesis and test it in your own account.

Most famous AI ad case studies measure the wrong thing for your budget

The best-known AI advertising cases are brand stunts. They are measured in reach or production savings, not ROAS (return on ad spend) or CPA (cost per acquisition). If you spend ₹10 lakh to ₹1 crore a month on Meta and Google, they tell you little about your next purchase campaign.

A useful split is to sort AI results into three kinds. Generation wins show up as lower cost and faster production. Idea-led wins show up as reach. Optimisation wins show up as ROAS, CPA and CTR (click-through rate). Only the third kind maps directly to your profit and loss.

The second problem is who is reporting the number. Most roundups mix platform data, brand disclosures and vendor sales material without saying which is which. In this article, every number carries one of four grades.

GradeWho reported itHow much to trust it
A: Platform averageMeta or Google, from their own dataA fair guide for an average account, not a promise for yours
B: Brand or agency disclosedThe brand or its lead agencyReal, but rarely measured against a control group
C: Vendor caseA company selling the AI toolChosen because it won. Treat it as a hypothesis
D: Practitioner anecdoteOne agency, one accountA test idea, nothing more

What Meta and Google officially claim, and the conditions attached

Platform-native AI campaigns are where most ₹10 lakh+ budgets actually go. They also have the best-sourced numbers, all grade A. The gains are modest averages with conditions that case-study pages usually leave out.

Meta Advantage+ sales campaigns

The condition: these campaigns run with Advantage+ audience, placements and campaign budget switched on. Several guides citing Meta's Help Center say Advantage+ switches off if a campaign has multiple ad sets without campaign budget. We could not check the Meta page directly. So some "we tried Advantage+ and it failed" stories were never really Advantage+ tests.

Google AI Max for Search

Google's AI Max help page says advertisers that activate it typically see 14% more conversions or conversion value at a similar CPA or ROAS. That figure comes from Google internal data from 2025, for non-retail advertisers. Think with Google quotes 27% more conversions, but against campaigns that mostly use exact and phrase match keywords.

Same product, two baselines. If you already run broad match with Smart Bidding, expect something closer to the smaller figure, or less.

Conditions that change the result

Google says AI Max won't be effective if campaigns are limited by budget, and its Search Term Matching feature does not work with manual CPC bidding. Two accounts using the same tool can get opposite results for these reasons alone.

For how these systems make decisions, see our explainer on how artificial intelligence in advertising works.

Creative and optimisation cases: what brands actually got

The famous brand cases prove that AI makes creative cheaper and faster to produce. None of them is a D2C purchase campaign with a holdout group. Here they are, graded.

CaseReported resultGrade
KalshiA TV spot made with Google's Veo 3, reported to have aired during Game 3 of the 2025 NBA FinalsB
CadburyMore than 130,000 individual video ads for local stores, per OgilvyB
Heinz "A.I. Ketchup"Over 1.15 billion earned impressions and social engagement 38% above benchmark, per award entriesB
JPMorgan Chase + PersadoMachine copy lifted CTR by over 450% versus human copy in some tested variants, per secondary reports of the Persado caseC
Vallo Media + StackAdapt DCO60% CTR lift. 30% of ad-attributed revenue from 12% of budgetC
PMax, sporting goods brand15% cost cut after adding "free" and "used" as negative keywordsD

Read the fine print. JPMorgan's 450% applies to "some variants", not the average. Vallo used DCO (dynamic creative optimisation) to retarget cart abandoners, a group that converts well anyway.

What carries over to your account is variety. Advantage+ and Performance Max (PMax) need many distinct creative concepts to test against each other. The PMax anecdote is a useful test idea: in that one account, a small manual control cut cost by 15%.

What most guides get wrong about AI ad case studies

Most roundups treat a 2019 or 2023 case as if the tools still work the same way. They don't. The rules changed in 2025-26, so an older case was run under different controls.

Some lists labelled "2026" are built on 2016-2019 cases. Those campaigns predate Advantage+, PMax and AI Max entirely.

How to run your own case study before moving more budget

The only case study that should move your budget is one run in your own account, against a control. Both platforms now make this practical.

  1. Fix tracking first. Confirm the Meta Pixel and Conversions API (CAPI) send deduplicated purchase events. On Google, check that conversion values match your store orders.
  2. Google: use the built-in experiment. AI Max offers a 1-click 50/50 split test between your base Search campaign and an AI Max arm. Use Smart Bidding, not manual CPC, and keep both arms off "limited by budget".
  3. Meta: run campaigns side by side. Put 20-30% of prospecting budget into an Advantage+ sales campaign and keep your manual campaign on the rest. Give both the same creative pool.
  4. Hold budgets steady. Avoid testing across Diwali, Ramadan or BFCM (Black Friday to Cyber Monday). CPMs (cost per 1,000 impressions) swing too much in those weeks.
  5. Separate existing customers and brand search. AI campaigns can lean towards people who would have bought anyway.
  6. Judge on store profit. Use delivered, pre-tax revenue from Shopify or WooCommerce, not platform ROAS.
Our rule of thumb

Run for at least 4 weeks and wait for at least 50 purchases per arm. Move more budget only if the AI arm beats the control by 10% or more on delivered pre-tax profit per unit of ad spend. At 50 purchases per arm, even a 10-20% gap can still be noise, so the more purchases each arm collects, the more you can trust the result.

ScaleWin does this comparison daily. It checks platform-reported results against store orders and break-even ROAS before suggesting a change.

Adjust any claimed ROAS for tax, COD and returns

A headline ROAS can overstate profit. Platform purchase values may include tax and orders that are never delivered. Adjust for both before you compare any case with your own numbers.

Tax on ad invoices and inside revenue

In India, Google lists IGST on ads at 18%. Tax guides say invoices from Google India or Meta India with a GSTIN are normal forward-charge supplies, while for foreign entities you self-pay under the reverse charge mechanism (RCM). Input tax credit (ITC) is usually available in both cases, but confirm your position with your accountant. If you claim it, judge ROAS on pre-tax spend.

In the UK, tax guides say B2B advertisers with a valid VAT number usually apply the reverse charge when Meta supplies from Ireland; check HMRC guidance or your accountant. Australia's GST is 10%, and sales tax rules in the US vary by state. On the revenue side, tax-inclusive order values make ROAS look better than it is, whether the tax is GST, VAT or sales tax.

COD and RTO in India, the Gulf and South-East Asia

COD (cash on delivery) orders count as purchases when they are placed. RTO (return to origin) orders are refused or undeliverable, so they never pay. Rates vary by category and region, so use your own courier data.

Worked example: a ₹30 lakh-a-month Meta account

Step 1: Remove tax from reported revenue

Example: Meta reports ₹1,20,00,000 purchase value on ₹30,00,000 spend, a ROAS of 4.0. Assume the values include 18% GST.

Pre-tax revenue
₹1,20,00,000 ÷ 1.18 = about ₹1,01,70,000
Step 2: Remove undelivered COD orders

Assume 50% of revenue is COD and 20% of COD orders end in RTO. That means 10% of revenue never arrives.

Delivered pre-tax ROAS
(₹1,01,70,000 × 0.90) ÷ ₹30,00,000 = ₹91,53,000 ÷ ₹30,00,000 = about 3.05
Step 3: Compare with break-even

The "4.0" is really about 3.05. If you cannot claim ITC, measure against the ₹35,40,000 invoice instead, which gives about 2.59. Against a break-even ROAS of 2.5, the margin is thin. A 9% average gain from an AI campaign matters, but it will not rescue a weak unit economy.

A scorecard for judging any AI ad case study

Score every claim before it changes your budget. Count one point for each "yes".

Our rule of thumb: a claim scoring 6 or more is worth testing in your account. A score of 3-5 is an idea for your test list. Below 3, ignore it.

For the measurement side, see our guide to attribution after cookies and iOS. For why a high ROAS can still lose money, read why a 3x ROAS campaign can be unprofitable.

Key takeaways

  • The most famous AI ad cases (Heinz, Cadbury, Kalshi) measure reach or production cost, not ROAS or CPA.
  • Official platform claims are modest averages with conditions attached: about 9% lower cost per conversion on Meta (Help Center figure as cited by guides), 14% more conversions from Google AI Max.
  • Label every claim by who reported it. Vendor and agency cases are test ideas, not evidence.
  • Many case studies predate the 2025-26 rule changes to Advantage+, Performance Max and AI Max.
  • Run your own controlled test and judge it on delivered, pre-tax profit, not platform ROAS.

Artificial intelligence advertising: frequently asked questions

What are real examples of AI in advertising with measurable results?

The best-sourced numbers are platform averages. Meta's Help Center, as cited in 2026 guides, reports a 9% average improvement in cost per conversion from Advantage+ sales campaigns. Google reports 14% more conversions or value from AI Max at a similar CPA or ROAS. Brand cases such as Cadbury's 130,000+ local video ads (as reported by Ogilvy) measure scale and production, not profit.

Is Advantage+ shopping the same as Advantage+ sales campaigns?

Yes. Meta renamed Advantage+ shopping campaigns to Advantage+ sales campaigns, a change industry guides date to early 2025. Guides that still say ASC are usually describing older rules, including the old existing customer budget cap.

How much better is AI Max for Search than normal Search campaigns?

Google quotes 14% more conversions or conversion value at a similar CPA or ROAS, based on 2025 internal data for non-retail advertisers. A separate 27% figure compares against campaigns that mostly use exact and phrase match. Your result depends on your starting setup, so test it with Google's built-in 50/50 experiment.

Can I add negative keywords to Performance Max?

Yes. Campaign-level negative keywords reached all Performance Max campaigns in 2025 and apply to Search and Shopping inventory. Negative keyword lists followed in September 2025. Google's Help Center says the limit is now 10,000 negative keywords per campaign, up from 100.

How do I test whether AI campaigns work in my own ad account?

Fix tracking first, then run a controlled split. On Google, use the AI Max 1-click 50/50 experiment. On Meta, run an Advantage+ sales campaign alongside your manual campaign. Keep budgets steady for at least 4 weeks and judge the result on delivered, pre-tax profit.

Do I pay GST on Meta and Google ads in India, and how does it affect ROAS?

Yes, 18% IGST applies. Tax guides describe it as forward charge on Indian-entity invoices or RCM on foreign ones, with ITC usually available in both cases, so confirm with your accountant. If you cannot claim it, a ROAS of 4.0 on stated spend becomes about 3.39 against the invoice.

Which AI advertising case studies are vendor claims?

Results published by a company selling the AI tool are vendor claims. Examples include DCO lift figures from ad-tech providers and copy-testing results from AI copy vendors. Treat them as hypotheses to test, not as evidence for your account.

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.