A payment button, a conversion drop, and the case that found it

The situation

HarukaEdu ran the platform behind Kartu Prakerja, the government training programme. Course enrolment converted at a healthy rate until it did not. The rate fell from above 20% to below 15%, and from the outside that looks like a traffic problem.

I was the only data person in the company, covering product, marketing and enrolment.

Finding it in the funnel

I broke the funnel into steps: landing page, course page, checkout, payment. Acquisition was fine. The drop sat between the course page and the payment form, which is where a product problem hides and a marketing problem gets blamed.

It was not uniform across sessions. On some screens the payment button sat below the fold, and a user who had already decided to buy had to scroll to find the thing that took their money. Users who never scrolled never paid.

The fix

I proposed a floating payment button that stayed in view while the page scrolled. Conversion came back to 20-25%.

The lesson is not that buttons are important. It is that "conversion dropped" is a question about the product, and a funnel report is how you get from the business complaint to the click that caused it.

The pipeline nobody else was going to build

Reporting was rebuilt by hand every week. Someone exported, joined, pasted and formatted, and by Monday afternoon the number was already a week old.

I built the first data pipeline on AWS, which replaced the weekly manual rebuild and gave the team numbers they could query instead of wait for. It was the unglamorous half of the year and the half that outlasted the fix.

Management reporting

I also visualised daily Kartu Prakerja transaction data for management, replacing the manual pull the team did each morning. The programme moved fast enough that yesterday's numbers had opinions attached by lunch.

What I would do differently

Two habits I would keep: opening the page myself before blaming the campaign, and asking what the data cannot see. One thing I would add is a screen-size breakdown in the funnel report from day one, because the answer was sitting in a dimension nobody had requested.

What this proves

A conversion question is usually a product question wearing a data costume. I can run the diagnosis alone, argue for the fix in product language, and then build the pipeline that keeps the number visible afterwards.