Measure What Grows Revenue When Clicks Disappear
Step 7 of 7 in The Definitive Guide to Marketing in the Age of AI (in Seven Steps). Originally published February 24, 2025. Rewritten for the AI era October 26, 2026.
To measure marketing when clicks disappear, use four instruments, each for its own job. Mix modeling decides how to split the budget. Incrementality experiments prove what a channel adds. A/B tests settle smaller choices on pages and messages. AI-visibility checks give you a leading indicator. Then write the decision rule before you read the result. An insight without a decision rule is just another report.
Why does click-based attribution break in the age of AI?
Because the click is no longer the first thing that happens. A buyer asks an AI assistant and reads a summary that names three vendors. She hears your name again on a podcast. Two weeks later, she types your company name into a browser. Last-click attribution gives all the credit to that last search. Everything that made it happen is invisible.
When Google shows an AI summary, people click a regular result about half as often (Pew; chapter 6 has the numbers). Attribution has less to count, while buyers keep learning from the answer.
Marketers feel it. Measuring marketing ROI was the top challenge in HubSpot’s 2026 survey, cited by 33% (HubSpot). In CMI’s research, 33% of B2B marketers named measuring content effectiveness as a top challenge (CMI). And with budgets flat, you have to defend every dollar.
What’s the difference between mix modeling, incrementality and attribution?
They answer different questions. Use each for its own job.
| Instrument | The question it answers | Best for | Watch out for |
|---|---|---|---|
| Marketing mix modeling (MMM) | How much does each channel contribute to revenue, and where should the next dollar go? | Budget splits across channels, including ones with no clicks | Needs enough history and spend variation; answers come in weeks, not days |
| Incrementality experiment | What happens if we turn this channel off, or up, for a test group? | Proving a channel’s real effect before a big budget move | Needs a clean test and control design |
| A/B test | Which version of a page, email or message works better? | Headlines, offers, formats, calls to action | Tells you about small choices, not budget splits |
| Multi-touch or last-click attribution | Which tracked touches came before a conversion? | Day-to-day optimization inside digital channels | Blind to AI answers, word of mouth and offline influence |
What did mix modeling find at a single Harrah’s property?
That more than $30 million a year in promotional spend could go without hurting results. In 2012, working in partnership with Polaris Research, we used marketing mix modeling to optimize the promotional campaigns at a single Harrah’s property. The work cut more than $30 million a year from marketing spend, with performance maintained.
The lesson isn’t the size of the number. It’s the order of operations: model first, then cut. Let the evidence show what is and isn’t contributing, and let the budget follow it. AI makes that habit worth more, because AI makes it easy to fund whatever is easiest to count.
Disclosure: Paul is EVP, Business Development at Polaris Research, the modeling partner on the Harrah’s work.
Why do so few companies act on mix modeling?
Because they treat it as a report instead of a decision tool. In an HBR survey sponsored by Google, 73% of respondents said they’re under growing pressure to show what their media channels do. Only 22% were “leaders,” very effective at using mix modeling to act on insights. Among leaders, 55% connect model outputs to real marketing decisions, compared with 20% of laggards (EMARKETER).
The gap is decisions, not data. Which brings us to the most useful habit in this guide.
What is a decision rule, and why write it first?
A decision rule says what you’ll do with a result before you see it. It turns measurement into action and stops anyone from explaining away an inconvenient answer.
| If the read shows… | Then we will… | Owner | By when |
|---|---|---|---|
| A channel’s modeled return is below our threshold for two reads in a row | Cut its budget by a set share and run an incrementality test | CMO | Next budget cycle |
| A test group with the channel off shows no drop in qualified conversations | Move that budget to the best-performing channel | Head of growth | Within 30 days |
| We’re named in fewer than one of five target questions after 90 days | Shift effort from owned pages to earned placements | Content lead | Next quarter |
The rows are illustrations. Your thresholds and owners are yours to set. The point is that they’re written down before the data arrives.
Is A/B testing still worth doing?
Yes, for the decisions it fits. Test headlines, offers, page structure and calls to action. Test one thing at a time. Decide in advance what “better” means. Run it long enough to trust. What A/B testing can’t do is tell you whether a channel deserves its budget. That’s a job for models and experiments.
Models can make even small choices pay off. For Cross, in work through PMB from 2010 to 2017, a predictive targeting model more than doubled the revenue from each campaign. At RAPP (1991–2005), the consumer programs I led generated over $1.5 billion in annual revenue for Dell. We also used predictive modeling there to cut acquisition costs for clients such as AT&T and American Express. Testing tells you which message wins. Modeling tells you who to send it to.
How do you measure AI visibility when there’s no click?
With a fixed set of questions, checked the same way every month. AI visibility is a leading indicator. It moves before revenue does.
- Pick five questions your buyers ask before they’d hire you, and keep them fixed.
- Check four engines monthly: ChatGPT, Gemini, Perplexity and Google AI Mode.
- Log three things: whether you’re named, who’s named instead and which sources are cited.
- Watch impressions and clicks separately in Search Console. Seer Interactive found cases where impressions more than doubled while clicks held flat, so click-through rate alone can mislead (Seer Interactive).
- Ask every enquiry where they first heard of you, and record answers that mention AI tools.
Checklist: measure what grows revenue
- Each budget question matched to the right instrument (model, experiment, A/B test or attribution)
- Mix modeling or a lighter-weight model in place for channel splits
- At least one incrementality test planned for your biggest channel
- Decision rules written before each read, with owners and dates
- A fixed five-question AI spot-check, run monthly in four engines
- An enquiry form or intake call that asks where people first heard of you
Where to go next
Measurement closes the loop and starts the next one. Take what you learned back to chapter 1. For the same open loop in content, read Thought Leadership and Curriculum Fail the Same Way. Want an outside read on which spend grows revenue and whether AI answers name you? Ask about the measurement and AI-visibility audit: work with me.
The click was always a proxy. Now that it’s fading, measure the thing it stood for.
