The setup
A capable team, a healthy funnel, and a plateau nobody could explain. They had tested headlines, button colours, and hero images for months. Every test won a fraction of a percent and the overall number would not move. The problem was not their execution. It was where they were looking.
On paper, everything looked healthy. Traffic was strong and growing. The cost per click was reasonable. The landing page was clean, modern, and fast. The team was diligent — they ran tests, kept a backlog of ideas, and shipped a new variant almost every week. By every measure of activity, they were doing CRO correctly. And yet the conversion rate sat flat as a table, month after month, immune to everything they threw at it. That is the frustrating part of a plateau: it does not feel like failure, because you are clearly working. It feels like a mystery.
They came to us with a reasonable request: help us test faster and find bigger wins. But faster testing on the wrong step just finds the wrong answer sooner. Before touching a single variant, we asked a question they had not: where in this funnel is the most money actually being lost — and why? The answer was nowhere near the button they had been recolouring.
The misdiagnosis
The team assumed the leak was on the landing page, because that is where the traffic landed and where CRO tools point you. Months of on-page tests reinforced the belief — and quietly wasted the effort. It is an easy trap, because the landing page is the most visible step, and visibility feels like importance. But the most visible step and the most costly step are rarely the same one.
- They optimized the step with the most traffic, not the most lost value.
- They tested variants without a hypothesis about why people actually left.
- They never spoke to a single visitor who did not convert — so every test was a guess dressed as data.
This is the quiet flaw in how most teams practice optimization: they test what is easy to test rather than what is expensive to leave broken. A/B tools make it trivial to swap a headline and measure the result, so that is what gets tested — endlessly. Meanwhile the real leak, which requires actually mapping the journey and talking to humans, goes untouched because it does not fit neatly into the tool. The measurement drives the strategy, and the strategy quietly wastes the year.
Finding the real constraint
When we mapped the full journey, the leak was not on the page at all. It was a mismatch between the promise in the ad and the first line of the page — a gap of a few words that cooled every hot lead on arrival. The ad promised one specific outcome; the page opened by talking about something adjacent but different. Visitors arrived warm, felt a flicker of “wait, is this the right place?”, and left. No button colour on earth fixes a broken promise.
- We mapped the whole path in Funnel OS, not just the landing page — ad, click, first impression, form, follow-up.
- We measured lost value per step, not traffic per step, and found the biggest drop was at the ad-to-page seam.
- We interviewed people who bounced — and heard the same confusion, in the same words, over and over.
- We changed one promise to match the ad. That was the entire fix.
The numbers
The fix was almost embarrassingly small — a rewrite of the page's first line to echo the exact promise of the ad. No redesign, no new features, no added spend. Here is what a few matched words did.
The reason a one-percent change moved a third of the revenue is that it was not a random one percent. It was the one percent sitting on the constraint — the single point where the most value was leaking. Fix a symptom and you get a symptom-sized result. Fix the constraint and the whole system moves.
The method, reusable
The specific fix does not transfer — your promise mismatch, if you have one, is somewhere else. What transfers is the method. Any time a funnel plateaus despite honest effort, run this before you test another variant.
| Instead of… | Do this | Because |
|---|---|---|
| Testing the highest-traffic step | Measure lost value per step | Traffic ≠ where the money leaks |
| Testing variants blindly | Form a hypothesis first | A test with no belief teaches nothing |
| Guessing why people leave | Interview people who bounced | They will tell you, in their words |
| Redesigning the whole page | Change the one thing on the constraint | Aim beats effort every time |
A different client was convinced their checkout was the problem and spent a quarter simplifying it. Conversion barely moved. When we measured lost value per step, the real leak was a shipping cost revealed only at the final screen — people did not abandon because checkout was hard, they abandoned because the price changed at the last second. Surfacing shipping on the product page recovered 19% of abandoned carts. Same principle: diagnose the constraint, do not assume it.
The lesson
The change touched roughly one percent of the words on the page and moved about a third of revenue, because it fixed the constraint instead of a symptom. The 1% that matters is never found by testing everything — it is found by diagnosing first. Every hour the team had spent recolouring buttons was an hour not spent finding the actual leak, and enthusiasm for testing had become a substitute for thinking about what to test.
Effort is not the scarce resource. Aim is. A small change on the constraint beats a heroic redesign of the wrong step.
Key takeaways
- A plateau is a diagnosis problem, not an effort problem. Working hard on the wrong step keeps the number flat no matter how many tests you run.
- Optimize lost value, not traffic. The most visible step is rarely the most costly one — measure where money actually leaks.
- Talk to people who did not convert. The reason they left is usually sitting in plain language they will hand you if you ask.
- The biggest wins are often upstream. Promise mismatches and hidden costs at the seams beat any on-page tweak.
- Aim, then act. A change to the 1% on the constraint can move a third of revenue; the same effort elsewhere is a rounding error.
Conclusion
The most expensive thing this team did was not a bad test — it was months of good tests aimed at the wrong step. Their diligence was real; their aim was off, and aim is what determines whether effort compounds or evaporates. The moment they stopped asking “what should we test next?” and started asking “where is value actually leaking?”, the plateau ended.
Before your next round of tests, map the whole journey and measure the lost value at every step. Find the single point bleeding the most, form a real hypothesis about why, and fix that. You may find, as they did, that the answer is a handful of words nobody thought to question. Diagnosis is unglamorous and it is the entire game. Map your own funnel in Funnel OS
Frequently asked questions
It is shorthand for the idea that a tiny, precisely-aimed change on the true constraint can move an outsized share of revenue. In this case, rewriting roughly one percent of a page's words — the part sitting on the leak — shifted about a third of revenue. The point is aim, not the literal percentage.
Because the tests were aimed at the highest-traffic step rather than the highest-loss step. The landing page was the most visible part of the funnel, but the value was leaking upstream at the ad-to-page seam. Optimizing the wrong step diligently still leaves the plateau in place.
Map the entire journey rather than one page, measure lost value at each step instead of traffic, and interview people who did not convert. The step with the largest drop in value — often a seam between two stages — is your constraint. That is where a small change pays the most.
No — testing is essential, but it is the last step, not the first. Diagnose the constraint, form a hypothesis about why people hesitate there, then test the change that removes it. Testing without diagnosis is just guessing with extra steps.
When it sits on the constraint, yes. A mismatch between an ad's promise and a page's first line cools every hot lead on arrival — fixing it re-warms all of them at once. The size of the result depends entirely on whether the change addresses the real bottleneck or a cosmetic symptom.