TL;DR: The numbers in marketing reports and the real growth of a business are starting to tell different stories. We call the space between them the proof gap, and it has become a problem for leadership teams, not just for marketers.
Closing it means three changes: measuring the sales your marketing caused rather than the sales it was linked to, checking your data before you trust it, and judging search by how visible your brand is, not only by how many clicks it gets. AI will change how all of this is measured, though no one knows exactly how yet.
For most of the last decade, online retailers have judged their marketing on the figures their ad platforms and analytics tools report. The main ones are platform ROAS, last-click revenue and organic sessions. Everyone knew these numbers were not perfect, but they were treated as good enough, and budgets were set around them. That is changing.
The same questions keep coming up across ecommerce. Which method of crediting sales to our marketing should we believe? Is paying for ads on our own brand name adding anything? Does the revenue in our analytics match the orders we actually took? Is traffic from Google decreasing, or just changing? And where does AI fit in?
They all come back to one question for leadership:
How confident are we that our marketing is creating growth, rather than simply recording it?
The gap between those two things is what we call the proof gap. In our view, closing it is one of the biggest commercial challenges leadership teams face.
Every business starts from a different place. Some have been running controlled tests and statistical models for years. Others still judge success mainly on the ROAS figure an ad platform shows them. Most sit somewhere on this scale:
Reported: the figures from ad platforms and analytics tools are taken at face value.
Reconciled: the revenue shown in analytics is checked against the orders the business actually recorded.
Tested: the biggest channels are tested to see how many extra sales they really create.
Modelled: media mix modelling, checked against those tests, is used to decide where the budget goes.
The three changes below apply at every stage. What differs is the next step each business needs to take.
Are you measuring the sales your marketing touched, or the sales it caused?
Attribution answers the question “what marketing did this customer see before buying?”. Leadership needs the answer to a different question: “would this sale have happened if we had not spent the money?” And incremental sales are only half the picture. A sale still has to make money once product costs, returns and the cost of winning the customer are taken off, so the better measure is incremental profit, with new and returning customers measured separately.
Paying for ads on your own brand name is where this shows most clearly. These ads usually report the best return in the whole account, because the people clicking them were already searching for you and were likely to buy anyway. Whether the spend is worth it depends on what else appears on that search page.
If competitors or other retailers are bidding on your name, the ads may be protecting sales you would otherwise lose. If no one else is there, you may be paying for visits you would have got for free. For brands that also sell through other retailers, the answer is sometimes not to change the bid at all, but to change the agreement with those retailers about bidding on your brand name.
How you test this matters as much as whether you test it. Switching ads off for a few weeks and comparing before and after is easily thrown off by seasonal trends and promotions. A more reliable method is to switch ads off in some regions while keeping them running in similar regions, then compare the two. Google and Meta also offer built-in lift studies, which run this kind of comparison for you.
More broadly, the figures from ad platforms, Google Analytics (GA4), media mix models and tests rarely agree, because each one answers a different question. Businesses with the most developed approach use each for what it does well:
- Ad platform and analytics data for day-to-day decisions, such as which ads to keep running.
- Controlled tests to find out how many extra sales a channel really creates.
- Media mix modelling, checked against those tests, to decide how to split budget between channels.
When the numbers disagree, it does not mean the measurement has failed. It shows you where to test next.
Can you trust the data your decisions are built on?
Every marketing decision is only as good as the data behind it, and that is easy to lose sight of at leadership level.
Quick-pay options at checkout are a good example. Apple Pay, Google Pay and PayPal often process a purchase differently from a standard checkout. If your tracking misses purchases made this way, the money still reaches the business, but it never shows up in the reports. That quietly skews the picture of which channels are working, the results of website tests and decisions about where to spend, all at once.
It also affects how website tests should be read. Say a retailer removes the quick-pay buttons from its checkout to make room for customer reviews. The page may look more convincing but be harder to buy from. If quick-pay purchases were not being tracked properly in the first place, that test could even look like a success.
Accurate data has to come first, and it is an ongoing job rather than a one-off project. The revenue in analytics should be checked against actual orders, split by payment method and device, before it is used to judge performance. Ad platforms need the same fix: tools such as Google’s enhanced conversions and Meta’s Conversions API send sales data straight from the business’s own systems, so fewer purchases go missing.
Is search traffic falling, or just changing?
Google now shows AI-written summaries (AI Overviews and AI Mode) and more ads at the top of search results, which leaves less room for the normal listings. It is tempting to read this as the end of SEO. We see it differently: the way people find brands is changing, but being found in search matters as much as ever.
AI answers let people make up their minds before they click on anything, and sometimes they never click at all. But those answers are still built from website content that search engines can easily read, and from what other websites say about a brand. The basics of good SEO still apply. What has changed is how success should be measured.
For leadership teams, that means tracking how often your brand appears and is mentioned in AI answers, not only how many visits you get. It means bringing SEO, content and PR teams closer together. And it means treating a fall in visits from search separately from a fall in sales from search, because they are different problems.
The other side of this change is advertising inside AI tools. Early reports from advertisers are mixed. Our view is that it is worth learning about but not yet worth spending heavily on: set a capped test budget and agree in advance what success looks like.
What should leadership teams do differently?
No single marketing team can make these changes alone. Four habits help:
- One agreed revenue figure: marketing, ecommerce and finance agree on a single revenue number, allowing for a small, agreed difference between their systems.
- A permanent test budget: money set aside so tests run all year round, rather than arguing once a year about whether marketing works.
- A higher bar for big decisions: large budget changes need a test or model to back them up, not just a dashboard.
- Asking how we know: leaders ask “how do we know?” as often as “how did we do?”.
How will AI change marketing measurement?
AI will change how marketing is measured. How much, and how quickly, is still unclear. We are watching three areas:
- Automated bidding. Google’s Performance Max and Meta’s Advantage+ campaigns use AI to decide where ads appear and how much to bid, based on the results they are told to aim for. If those results include sales that would have happened anyway, the AI will chase them and the proof gap could widen. The practical fix is to give these tools better targets, such as the profit on each sale rather than its revenue, and goals focused on winning new customers.
- Less of the customer journey can be tracked. As more people make decisions while reading AI answers, tools that track clicks see less of the path to purchase. For some channels, tests may become the main way to measure what is working.
- AI doing the measuring. AI is making data analysis and modelling faster and easier. Whether that makes businesses more confident in their numbers, or just gives them more numbers to argue over, remains to be seen.
We do not have all the answers yet, and we would be wary of anyone who says they do. What we do believe is that businesses with reliable data and a habit of testing will be best placed to find out.
How do you close the proof gap?
It is healthy to question figures you have relied on for years. The next step is to make that questioning a regular habit. The brands that pull ahead will not necessarily have the biggest budgets or the most advanced technology. They will be the ones that can show, with confidence, which parts of their marketing make a real difference to sales and profit.
A dashboard can tell you what happened. Only proper testing can tell you what your marketing caused.
Where is the proof gap widest in your business right now? We would love to hear how other teams are approaching it.
