The landing page we A/B tested 14 times before we understood what the problem was

The client was a professional certification training company based in Hyderabad. They ran prep courses for finance certifications, good product, reasonable pricing, decent reputation. Their landing page was converting at 1.1 percent on paid traffic and they wanted it at 3 percent. That is a reasonable goal. We thought we could get there in three or four tests.

Fourteen tests later, we were at 1.4 percent and genuinely confused.

We had tested headlines. We had tested the hero image versus no hero image. We had tested testimonial placement, form length, button color, social proof format, FAQ copy, and pricing display. Every change was a reasonable hypothesis. Most of them moved the needle by some amount but nothing moved it enough and nothing moved it consistently.

The moment we stopped trusting the data

Our designer asked if we could just watch some real users navigate the page. Not a formal session, just screen recordings from Hotjar that we had not looked at carefully enough. We spent an afternoon going through them.

What we saw was not in our A/B test hypotheses. Users were scrolling to the "who is this for" section, reading it, and then going back to the top and leaving. The section was accurate, it listed the target audience correctly, but it used certification acronyms that a lot of prospective students did not recognize yet. They were checking if the course was for them and the answer they were getting was written in a language they had not learned yet.

That is not a headline problem. It is not a button color problem. It is a clarity-of-audience problem embedded in language that felt clear to us because we knew the domain.

What we actually fixed

We rewrote the "who is this for" section in plain language. Instead of "ideal for CFA Level 1 candidates seeking structured preparation," we wrote "ideal for working professionals who have decided to pursue the CFA and need a study plan that fits around their job." We added a two-line explanation of what the certification is for people who were researching it for the first time.

Then we ran one more test. Conversion went to 2.9 percent. Not fourteen tests of incremental optimization. One rewrite based on watching actual humans get confused.

The lesson we keep relearning is that A/B testing is only as good as your hypotheses, and hypotheses only come from understanding where users are actually getting stuck. The data tells you that something is wrong. It rarely tells you what.