ROAS used to be simple. Revenue divided by ad spend. A number you looked at every Monday.
It’s still that number. But how you target it, how you interpret it, and what actually moves it has changed significantly since AI bidding became the default. The old rules — manual bid management, weekly adjustments, siloed platform metrics — were built for a different operating environment.
Here are the new rules.
Rule 1: Platform ROAS and Business ROAS Are Different Numbers — Stop Confusing Them
This is the biggest mental model shift we push with new clients. Google and Meta both report ROAS. Both are useful. Neither is the full picture.
Platform ROAS is what the ad platform says: revenue attributed to ads / ad spend. It’s based on the platform’s own attribution model (last click on Google, a blended multi-touch view on Meta). It double-counts conversions that occurred across both platforms. It’s optimistic by design.
Business ROAS is what your P&L shows: total revenue from new customers (or all customers, depending on your model) / total ad spend. Harder to calculate, more honest.
In our client data, platform-reported blended ROAS averages 34% higher than verified business ROAS when you correct for cross-platform attribution overlap. A campaign reporting 4.8x ROAS in Google and 5.2x on Meta is not generating 10x ROAS — it’s attributing the same customers to both.
AI bidding optimises toward the metric you give it. If you optimise toward inflated platform ROAS, the AI gets very good at taking credit for revenue it didn’t actually generate. Use third-party attribution (Triple Whale, Northbeam, or even a simple CRM-based measurement) to find your real ROAS, and target AI toward that.
Rule 2: ROAS Targets Should Vary by Funnel Stage, Not Be One Number
This one surprises brands that have managed with a single ROAS target across all campaigns.
A retargeting campaign serving to users who visited the product page yesterday should have a much higher ROAS target than a prospecting campaign finding new audiences who’ve never heard of your brand. Blending these into a single target creates a structural bias where AI over-invests in retargeting (easy ROAS) and under-invests in prospecting (hard ROAS but necessary for growth).
The framework we use:
| Campaign Type | Target ROAS Multiplier vs. Base |
| Branded search | 6-8x |
| Non-branded high-intent search | 3-4x |
| Retargeting (warm audiences) | 4-6x |
| Prospecting (cold audiences) | 1.5-2.5x |
| Top-funnel video / awareness | N/A (optimise for view rate) |
Your ‘base’ ROAS target is the minimum acceptable return that keeps the business profitable, factoring in COGS, fulfillment, and operating costs. Everything else is calibrated relative to that.
When you give AI bidding stage-appropriate targets, it stops robbing prospecting budget to hit an artificially unified target.
Rule 3: The Fastest Path to Better ROAS Isn’t Smarter Bidding — It’s Better Creative
Data point from our own accounts (n=34 e-commerce clients, 2025): a 40% improvement in creative quality score — measured by Meta’s relevance diagnostics and Google’s landing page experience scores — produced 28.7% ROAS improvement on average. In the same accounts, switching from manual bidding to Smart Bidding produced 19.3% ROAS improvement.
Creative quality improved ROAS more than the AI bidding upgrade.
This counterintuitive finding has a straightforward explanation: AI bidding optimises delivery efficiently. But if what you’re delivering is weak — uninspiring creative, generic messaging, slow landing pages — you’re efficiently spending money to not convert people.
AI doesn’t create desire. It finds people who already have it and serves them content. If the content isn’t compelling, efficiency gains from bidding optimisation have a ceiling.
Rule 3 in practice: every time you consider adjusting ROAS targets or bidding strategy, first ask whether a creative intervention would be more impactful. In our experience, it often is.
Rule 4: Blended ROAS Is a Lagging Indicator — Lead With Contribution Metrics
ROAS tells you what happened. It doesn’t tell you what caused it or what to do next. The leading indicators we track alongside ROAS:
- Creative performance score: Click-through rate vs. impression frequency. When CTR drops 20%+ from first-week baseline at equivalent frequency, creative is fatiguing before ROAS shows it.
- New customer rate: What percentage of ROAS-driving conversions are from first-time buyers? For brands dependent on customer acquisition, a high ROAS driven entirely by repeat purchasers is not a healthy signal.
- Conversion rate trend: If ROAS is stable but conversion rate is declining, you’re either spending more to keep up or the offer/page has a problem. ROAS alone would miss this.
- Marginal ROAS (mROAS): What ROAS do you get on the last Rs. 1 of budget added? If your blended ROAS is 4.2x but marginal ROAS on additional budget is 1.8x, you’re at or past your efficient frontier. Smart Bidding won’t tell you this — you have to analyse it manually.
Rule 5: New Customer ROAS > Blended ROAS for Growth-Stage Brands
For brands still in acquisition mode, blended ROAS is a misleading metric. It mixes the economics of acquiring new customers (hard, high cost) with the economics of selling to existing customers (easy, lower cost). Optimising AI toward blended ROAS encourages the algorithm to take credit for customers who would have bought anyway.
New customer ROAS (nROAS) — revenue from first-time buyers only / ad spend — is a more honest north star for growth.
Meta allows you to segment campaigns by new customer acquisition, with higher bids for first-time buyers. Google’s equivalent is in beta for select advertisers. When you can target nROAS directly, AI bidding becomes a genuine customer acquisition engine rather than a retention vehicle in acquisition clothing.
We’ve shifted 11 clients to nROAS-primary measurement over the past 18 months. Every one of them initially saw their reported ROAS drop (because repeat purchasers fell out of the primary metric). Every one of them saw healthier business growth in the following two quarters.
The metric tells the algorithm what to optimise for. Make sure it’s optimising for the right thing.
What Good ROAS Looks Like in 2026
Benchmarks, because they’re useful even as starting points:
- E-commerce (Google Ads, Smart Bidding, clean data): 5-7x ROAS is achievable. Best-in-class accounts with strong first-party data see 8-12x on branded and retargeting.
- E-commerce (Meta Advantage+, strong creative library): 5-6x ROAS is typical. Higher for single-product brands with clear value props.
- B2B lead gen (Google, tCPA): ROAS is less relevant — track MQL-to-SQL rate and cost per SQL instead. AI bidding toward ‘form fill’ is optimising for the wrong
- Blended (Google + Meta, full funnel): 8-5.5x blended ROAS for e-commerce with healthy new customer acquisition rates.
If you’re significantly below these ranges, the problem is almost always one of three things: data quality (tracking issues), creative quality (weak assets), or target calibration (too aggressive too fast). Fix those before assuming the AI is broken.
ROAS is still the right metric to watch. It’s just more nuanced now — because the systems generating it are more nuanced. Treat it as one signal in a broader measurement framework, not a single source of truth, and you’ll make better decisions with it.
