How to Use Chained A/B Tests to Diagnose Your Conversion

When A/B testing is discussed, it's almost always thought of as a standalone experiment: you change a button, compare two versions, and either a winner emerges or it doesn't. However, an A/B test is, above all, a diagnostic tool. And sometimes, a test that "doesn't win" is the one that provides the most valuable clue for the next one.

This post explains exactly what an A/B test is and, using a real case study from Viuty, how four chained tests (three of which did not yield a positive result in the final metric) allowed us to pinpoint the real cart bottleneck and resolve it with an improvement that generated an additional +€10,690 per month.

The Error of Viewing Each A/B Test as an Isolated Experiment

An A/B test compares two versions of the same page or element (the control and the variant), distributing traffic between them to measure which converts better. So far, that's the textbook definition.

The common mistake is to treat each test as a closed question: did it work, yes or no? If the answer is no, the test is closed, the learning is noted, and you move on to the next idea on the list. The problem is that this approach treats each test as if it were unrelated to the previous one, when in reality, every result —whether it won or not— provides information about where your user moves (or gets stuck) in the funnel.

The Case Study: A CTA That "Worked" But Conversion Didn't Rise

Test 1 and 2 — Visible CTA Increases Interest, But Not Sales

The first test in this chain at Viuty involved a sticky CTA that kept the "add to cart" button visible at all times. The Add to Cart result was clearly positive: +11.6%, with nearly 99% probability of winning. At first glance, a success.

However, when looking at the next step in the funnel, the data changed sign: the conversion rate from Add to Cart to Initiate Checkout dropped from 58.9% to 52.3% (−11.2%). The key insight, however, was elsewhere: those who reached the checkout purchased almost identically in both variants (65.5% vs 65.1%). In other words, the sticky CTA generated more intent, but that intent didn't translate into more people reaching checkout. The test report itself highlighted that segment —cart → checkout— as the point to address in the next experiment.

The second test confirmed the same signal through a different path: Add to Cart increased by +4.48%, but Initiate Checkout fell again, and purchases ended up −11.37%.

Test 3 — Confirming the Real Leakage Point

In the third test, a new landing page with a direct shortcut to checkout further confirmed the diagnosis: Initiate Checkout increased by +65%, a huge improvement in that specific funnel step. But, once again, this improvement also did not translate into more purchases.

Three tests, three different results in their final metric —one superficially positive, two without a winner— but all three pointing to exactly the same place: the problem was not in generating interest or driving traffic to checkout. It was in what happened inside the cart, just before the user took the final step.

Test 4 — Addressing the Exact Point of Friction

With that diagnosis already validated by three different experiments, the final test directly redesigned the cart: payment methods and trust seals (delivery, returns) were moved just above the purchase CTA, the price block was unified in a single location, and the button copy changed from "Terminar pedido" (Finish Order) to "Finalizar pedido →" (Complete Order →).

The Result: What Changed and How Much It Impacted the Business

In just 15 days, these were the results:

<table> <thead> <tr> <th>Metric</th> <th>Control</th> <th>Variant</th> <th>Change</th> </tr> </thead> <tbody> <tr> <td>Purchase (main objective)</td> <td>22,12%</td> <td>23,75%</td> <td>+7,37% (90,57% de probabilidad de ganar)</td> </tr> <tr> <td>Track Add Payment Info</td> <td>25,65%</td> <td>27,04%</td> <td>+5,42%</td> </tr> <tr> <td>Revenue Per Visitor (RPV)</td> <td>``</td> <td>``</td> <td>+11,6%</td> </tr> </tbody> </table>

The variant crossed the account's Bayesian threshold (90% probability, expected loss of 2.58%, below the accepted maximum of 3.54%). The estimated economic impact: an additional +€10,690 per month. As in the three previous tests, the effect was more pronounced among returning users (+10.55% in conversion, +15.18% in revenue per visitor), who represent 64% of the account's traffic.

The Methodological Lesson: Reading Tests Collectively, Not Individually

If Viuty had closed the chain after the first test, they would never have reached the final conclusion. And if they had closed it after the second or third —"no winner, next idea on the list"— they would have lost the clue that each of those tests, in its own way, was leaving: the problem resided in the cart → checkout segment, not before or after.

Each individual test answered a small, specific question. The big question (where the leakage truly lies) is only answered by reading the complete series.

How to Set Up Your Own Diagnostic Chain with A/B Tests

You don't need to hit a home run with the first test that's going to "win." What's needed is a clear criterion for interpreting each result, whether it wins or not, as a clue about the funnel:

• Always measure the next step, not just the metric you are directly testing. A test can improve a micro-objective (Add to Cart, clicks, scroll) and worsen the business objective if you don't look at the subsequent step.
• When a test doesn't win, ask yourself what it does confirm. A "no winner" with a consistent pattern across several segments (like Initiate Checkout falling in two consecutive tests) is a signal, not a failure.
• Don't change your hypothesis with each test. Viuty's chain worked because each subsequent test attacked the same friction from a different angle, not by jumping to a completely new idea each time.

If you want to identify which segment of your own funnel is experiencing similar leakage, our free CRO audit is an excellent starting point.