Is ROAS Enough When AI Shapes the Customer Journey?
As shoppers ask ChatGPT what to buy before they ever click an ad, marketers are learning that ROAS was never built to capture the moment demand is actually created.
Topics
What if your highest-converting marketing channel never appeared in your dashboard?
No impressions. No clicks. No attributed revenue. No ROAS. Yet every day it influences thousands of purchase decisions. That is exactly what is happening as consumers increasingly turn to AI assistants like ChatGPT before they ever visit a brand’s website.
While marketing teams celebrate improving ROAS, customers are making buying decisions in places traditional attribution cannot see. That’s not a reporting gap; that is a growth gap, and it is getting wider each quarter.
The Sale that Happened Before the Click
Imagine you are looking for a moisturiser for acne-prone skin. Ten years ago, you would have searched Google, clicked through a few sponsored results, read reviews, visited Sephora or Nykaa, and eventually made a purchase.
Today, the journey looks very different. You ask ChatGPT: “What’s the best moisturiser for acne-prone skin that’s under £25?”
Within seconds, it compares ingredients, explains why certain products work, filters out the ones with fragrance, and recommends two brands. You don’t click on an advertisement. You don’t browse ten websites. You simply open Sephora, Nykaa, or Amazon and buy the AI-recommended product.
From the retailer’s perspective, a sale happened. From the brand’s perspective, revenue increased, but your ad dashboard saw nothing. This is what I call the dark sale: a purchase whose decision was made inside an AI conversation but whose influence never appears in your attribution model.
Paid search campaigns are reporting a ROAS of 2.3x — slightly lower than last year. Naturally, the instinct is to optimise the campaigns, adjust budgets, or question the agency. But when you step back, something does not make sense.
Overall sales across the portfolio are growing YoY, and are growing for both nominated and those receiving little or no paid media.
Where did that demand come from?
Traditional attribution has no answer because it assumes demand begins when someone clicks on an advertisement (the last-touch model). AI has changed that assumption. Customers are increasingly deciding what to buy before they decide where to buy it. By the time they arrive on Amazon, Sephora, or Nykaa, the recommendation has already been made.
ROAS faithfully measures the final click. It has no visibility into the conversation that created the demand in the first place. So the question is not whether your ROAS has improved. The question is whether ROAS is still measuring the part of the customer journey that actually drives growth.
For a growing number of brands, the honest answer is no.
The Structural Limitations of ROAS in the AI Era
ROAS was created for a world where customer journeys were relatively linear: a customer saw an advertisement, clicked it, and made a purchase. Measure the revenue generated from that advertising spend, optimise the campaigns, and repeat.
And post the launch of AI, ROAS is still doing exactly what it was designed to do. The problem is that marketers are asking it to answer questions it was never built to answer.
1. ROAS is limited to the last touchpoint
ROAS measures the revenue attributed to advertising. It does not measure everything that influenced the purchase. If a customer asks ChatGPT which moisturiser is best for acne-prone skin, receives a recommendation, and later purchases that product on Amazon or Sephora after clicking a sponsored ad, ROAS attributes the sale to the advertisement.
But was it the advertisement that persuaded the customer to buy? Or was the decision already made inside the AI conversation? ROAS cannot distinguish between demand that advertising created and demand that advertising simply captured.
2. ROAS measures what has already happened, not what is about to happen
ROAS is a lagging indicator. It tells you how efficiently advertising converted existing demand into revenue. It cannot tell you whether your brand is becoming more discoverable in AI, whether customers are increasingly choosing competitors before reaching your website, or whether your future demand is quietly eroding.
By the time ROAS begins to deteriorate, customer behaviour has often shifted months earlier. The dashboard reports the outcome. It does not predict the direction.
3. ROAS reduces marketing to a spend-to-revenue equation
At its core, ROAS asks one question: for every £1 we spent on advertising, how much revenue did we generate? That is a useful measure of media efficiency. It is not a complete measure of business growth.
It says nothing about whether you acquired valuable customers, increased market share, strengthened brand preference, improved customer lifetime value, or became the brand AI recommends most often.
In the age of AI commerce, those outcomes increasingly determine future revenue. ROAS simply was not designed to measure them. That is why the question for marketers is no longer, “How do we improve ROAS?” It is what we should measure alongside ROAS to understand real growth.
If ROAS has so Many Limitations, Why is Management still Obsessed with it?
If most senior marketers understand the limitations of ROAS, why does it continue to dominate boardrooms? The answer is not ignorance. It is that ROAS solves several business problems remarkably well — even if it no longer solves the marketing one.
1. It is easy to understand
Marketing has no shortage of metrics. Customer Lifetime Value. Incrementality. Brand Equity. Share of Search. AI Visibility. Each provides valuable insight, but none is as universally understood as ROAS.
For every £1 spent, how much revenue did we generate? It is simple, intuitive, and easy to explain to a CFO or CEO in under a minute. When executives need to make quick decisions amid dozens of competing priorities, simplicity often trumps completeness.
2. It supports short reporting cycles
Businesses do not operate on twelve-month learning cycles. They operate on a monthly trading update schedule, quarterly board meetings, and annual budgets. ROAS fits neatly into those reporting rhythms because it responds quickly to changes in advertising performance.
Metrics like customer lifetime value, incrementality, or AI-driven demand often take months to become measurable. That does not make them less important. It simply makes them harder to use when organisations expect answers every month.
3. It reduces uncertainty
Marketing has always struggled to prove cause and effect. ROAS provides a number that creates confidence, even when customer behaviour is far more complex. It gives organisations a common language for comparing campaigns, channels, agencies, and investments. In an increasingly uncertain market, businesses naturally gravitate towards metrics that appear objective and actionable.
4. Organisations are naturally risk-averse
Changing a KPI is rarely just a reporting decision. Budgets, targets, incentives, agency contracts, and board expectations are often built around existing metrics. Replacing ROAS does not simply require a better metric — it requires organisations to change how they define success.
That is a much bigger challenge. Which is why most businesses will not stop using ROAS. Nor should they. The mistake is not using ROAS. The mistake is using it as the primary measure of growth in a customer journey that has fundamentally changed.
ROAS remains an excellent measure of media efficiency. It is no longer sufficient as a measure of business growth. So what does sufficient look like? That is exactly what the Trojan Horse Model was built to answer.
The Trojan Horse Model
By this point, most marketers do not need convincing that ROAS has limitations. The challenge is not recognising the problem. It is knowing how to change the conversation — and influence the people at the top who control the budget.
The Trojan Horse Model was built to solve exactly that problem.
What is the Trojan Horse Model?
The Trojan Horse Model takes its name from one of history’s most famous strategic victories — and its logic from the same playbook.
The Greeks did not conquer Troy by attacking its walls head-on. They entered the city disguised as something familiar. Only once the horse was inside did it reveal what had been hidden all along.
That is exactly how the framework approaches ROAS. It does not begin by challenging it. It begins by acknowledging it. ROAS is the outer shell of the horse — the metric leadership already trusts, the number that appears on every board report, and the KPI that gets everyone comfortable.
Then the horse opens. And what emerges changes the conversation permanently. Inside are three layers of measurement that ROAS alone can never reveal.
Soldier Metric 1 — Incrementality
ROAS tells you how your campaign performed. Incrementality tells you whether your marketing created growth. Those are not the same thing.
A customer who was already planning to buy may click your advertisement before making a purchase. ROAS records a successful campaign. Incrementality asks a different question:
Would that sale have happened if the advertisement had never existed?
That’s the difference between measuring activity and measuring impact. ROAS is an operational metric. It helps optimise campaigns, channels, and creative. Incrementality is a growth metric. It tells you how much new demand your marketing actually created.
Soldier Metric 2 — Customer Value
ROAS measures what a customer spent today. Customer Value measures what they are worth tomorrow. The objective is no longer simply acquiring customers. It is acquiring customers worth growing.
For monthly reporting, start with CAC (Customer Acquisition Cost) and end with AOV (Average Order Value).
- If it costs £45 to acquire a customer who spends £38, you have bought revenue — not growth.
- If it costs £28 to acquire a customer who spends £67, you have created value from day one.
For longer-term reporting, measure Customer Wallet Share.
Instead of asking how much this customer spends with us, ask how much of their total category spend belongs to us.
The formula is simple. Your brand sales divided by total category sales. If the category is worth £100K and you generate £20K of that, your wallet share is 20 per cent.
That number matters for three reasons.
- It tells you your real market position. A brand growing revenue 10 per cent while the category grows 20 per cent is losing ground, despite the headline number looking healthy. Wallet share exposes that. ROAS never would.
- It informs your targeting and content strategy. If a competitor holds 45 per cent of category spend and you hold 20 per cent — that gap is your brief. What are they owning in the customer’s mind that you are not? The answer shapes your content, your GEO investment, and your media priorities.
- It is the strategic metric top management actually needs. Every brand wants to be the market leader. Wallet share is the roadmap that shows whether you are moving toward it or away from it. It is not a monthly number. It is the annual view that connects daily marketing decisions to long-term market position.
Growth is not just winning customers. It is winning a greater share of the total category — and understanding precisely how much of that share you are leaving on the table.
Soldier Metric 3 — Share of Model
Before a customer visits a brand’s website or clicks on a single advertisement, they are increasingly asking AI what to buy. Share of Model measures how often your brand appears in those answers — across ChatGPT, Gemini, or any other LLM.
Think of it as the AI-era equivalent of Share of Voice. Instead of measuring how often people see your brand in paid placements, it measures how often AI recommends you when your category gets asked about. If competitors are consistently recommended instead of you, you have already lost part of the customer journey before your marketing has even begun.
How to use it?
AI agents and specialist tools can measure a brand’s Share of Model at scale across multiple platforms. Categorise the submitted queries into three buckets: customer popular, brand recommended, and competitor recommended. The gap between the second and third is your content brief.
Every query where a competitor appears and you do not becomes a content title and a GEO (Genitive Engine Optimisation) priority. This is the foundational brick of any GEO strategy. Because in the AI era, if you are not recommended, you are unlikely to be considered.
Scenario: A Monday Morning Board Meeting
Imagine it is 9 am on Monday. You are presenting the quarterly marketing review.
Slide One — ROAS
You open with the familiar number. ROAS 4 to 1. The business spent £200K on media and generated £800K in attributed revenue.
Everyone is comfortable. You do not challenge it. You do not add caveats. You let the room settle into the number they recognise. That is intentional. The horse needs to enter Troy looking familiar.
Slide Two — Incrementality
Only 20% of that £800K, i.e., £160K, was genuinely created by marketing. The remaining 80% would have happened anyway through existing demand, direct traffic, and organic search. You did not criticise ROAS. You gave it context.
Slide Three — Customer Value
Was this customer actually worth acquiring?
How much the brand spent to acquire a unique customer — CAC (customer acquisition cost). And how much that customer spent on purchase — AOV (average order value).
- If CAC exceeds AOV, you have bought revenue, not growth.
- If AOV significantly exceeds CAC, you have created value from day one.
For the longer reporting window, you go one level deeper. Customer Wallet Share — your brand’s revenue as a percentage of total product category revenue. Not just what the customer spent with you, but how much of the total available category spend you are actually winning.
The conversation has moved from ‘did our campaigns perform?’ to ‘are we acquiring and retaining the right customers?’ That is a fundamentally more valuable discussion.
Slide Four — Share of Model
ChatGPT, Gemini, Claude, and other LLMs are actively influencing purchase decisions before a customer visits a single website or sees a single advert.
Your Share of Model is 23%. Your closest competitor is 62%.
That gap means one thing. For the most popular queries in your category, your brand is not being recommended. Customers are being directed elsewhere before your marketing has even begun.
This is not an advertising problem. It is a discoverability problem.
That gap defines your content strategy and your GEO priorities. Every query where a competitor appears, and you do not, is a content title waiting to be written and a digital footprint waiting to be built.
The meeting has changed. Not because you attacked ROAS. Because you walked the room — slide by slide — to a place ROAS could never take them.
That is the Trojan Horse Model in action.
Customer behaviour changed. The metric did not. ROAS kept doing what it was designed to do 20 years ago—and it did so well. But efficiency is no longer the same as growth. The marketers who continue to optimise only for ROAS will become exceptionally efficient at capturing demand that already exists.
The marketers who measure incrementality, customer value, and AI visibility will be the ones creating new demand. Because the real question was never whether ROAS is enough. It is this — when AI is influencing what customers buy before they ever click, are you still measuring growth, or simply measuring the last step of a journey that has already been decided?
ALSO READ: Technology Changes Rapidly, But Culture Determines Trends
