Not every AB test that loses is a bad test. Sometimes the negative result on one metric is exactly what helps you find a better question.

That's what happened with Simbe, the physical intelligence and orchestration layer for retail. Its platform combines autonomous robotics, computer vision, RFID, and edge AI to help store teams understand and act on what is happening in their stores. Tally, its autonomous shelf-scanning robot, identifies out-of-stocks, price label errors, and misplaced products on shelves in real time.

The initial hypothesis for the first test was simple: a brief explanation of what Tally is and what problem it solves would clarify the messaging, help visitors understand the product faster, and increase engagement. However, the result seemed to point the other way: after the copy change, clicks on the main CTA dropped sharply.

Many teams would have discarded the test right away. Simbe's team, however, decided to analyze secondary metrics in detail, which revealed a different reality: the new message explained the product's value so well that it made the top-of-funnel exploration buttons less relevant, a sign of deeper conversion intent.

With that insight, they decided to run a second test. And that's where the story changed.

The initial challenge: explaining the product better

Tally is Simbe’s autonomous shelf-scanning robot, built for retailers. The product page has an important job at first contact: helping visitors quickly understand what the solution does and why it matters. This supports Simbe’s broader website goal: helping more qualified buyers find Simbe, take the right next step, engage with relevant content, and convert.

The original hero used more institutional messaging:

• Title: Tally 4.0
• Subtitle: Meet Tally, the autonomous shelf-scanning robot.

The growth team's hypothesis was that messaging centered on the product name was abstract. In the first 3 seconds, visitors need to confirm they're in the right place to solve their daily operational pain points. So the new version took an answer-first, outcome-oriented approach, describing the category and the robot's direct benefit:

• Title: Meet Tally
• Subtitle: The autonomous inventory robot finding out-of-stocks, pricing errors, and misplaced products in real time.

The first test: when the primary metric pointed to a loss

The primary metric was clicks on the hero CTA. As supporting metrics, the team tracked session duration, clicks on other areas of the page, and visits to deeper funnel pages.

When the consolidated results arrived, the impact on the primary metric was clearly negative:

• -32% in hero CTA clicks
• -19% in average session duration
• 99% probability that the original version was better

Looking only at these numbers, the conclusion would be simple: the new copy didn't work. But the secondary metrics told a different story:

• +110% in visits to the ROI Calculator page
• +86% in engagement with social proof among new visitors
• +35% in demo request intent signals in direct traffic

These data prompted a reflection on the user journey on the page: why did the new copy reduce clicks on the nearest CTA while generating more than double the engagement on mid-funnel elements that were less prominent on the page?

The answer was a paradigm shift: the new copy hadn't failed, it had worked too well.

Reading the new headline, visitors immediately understood what the product did and which problems it solved. They no longer needed to click "Explore Solutions" to discover the value proposition, and the button became inadequate for the decision stage they were in.

The second test: what if the CTAs followed the intent?

The question changed from "does the new copy work?" to "if the new copy already explains the product, are we offering the right next step?"

With this new interpretation in hand, the next experiment was born. The premise was straightforward: if the headline already does the job of explaining the product, the page's buttons need to capture the mid-funnel intent that this copy triggers.

The team kept the new copy and restructured the CTAs at the top of the page:

The results: proof of intent alignment

The hypothesis was confirmed decisively. By aligning the hero message with direct, conversion-focused CTAs, the page delivered strong results:

• +185% in hero CTA click-through rate, by replacing "Explore Solutions" with "Watch a Demo"
• +135% in visits to the ROI Calculator
• More than +5x in visits to the platform demo page

A clearer, outcome-oriented message can reduce the need for exploratory actions. When users already understand what the product does, they may be ready for an action closer to the decision. Keeping a generic CTA can create a mismatch between message and action.

In this example, the gain didn't come only from writing better copy, but from connecting it to the next step of the journey.

At Simbe, we are messaging to multiple audiences across channels at all times. For the website, it’s really important that we support retailers by giving them the information they need for the stage of the journey they are in and guiding them to the next best action. Croct has been a great partner in helping us do this more effectively, strategically, and in a data-driven way. Instead of guessing or making assumptions about what will resonate, we can test it and see.

What we learned from Simbe

Tally's landing page optimization case provides valuable lessons for marketing, product, and CRO teams.

1. An isolated primary metric can hide real user behavior

In the first test, the original copy won, and that was a valid call for the primary metric. But stopping the analysis there would have discarded a highly effective headline.

The increase in ROI Calculator interactions and the behavior observed on other page elements showed that the new headline was changing the user journey. The copy wasn't generating less interest, it was generating a different kind of interest. That distinction matters.

2. Copy creates intent, and CTAs should capture it

It's not enough to ask "which copy generates more clicks?" We also need to ask "after reading this copy, what action makes the most sense for this user?" This shift in perspective can completely change how we interpret an experiment.

3. CRO is a continuous cycle of hypotheses and learnings

The first test rarely tells the whole story. The first AB test wasn't a mistake, it was the input the second experiment needed to succeed. Each test can answer one question and, at the same time, reveal a better one.

The goal of a culture of experimentation isn't just to find out what works, but to understand why it works and use that to continuously refine the experience.

If you also want to turn your website data into product and conversion decisions, create your free account and start testing with Croct today.