University of Michigan Credit Union

Traffic held after launch. People were finding the site and moving through it. What they were not doing, consistently, was converting. It was the risk the research had named, now showing up in live behavior, and I had built the site to be measured so I could see where.

Before launch we set up a GA4 model around meaningful actions instead of pageviews, paired with CrazyEgg heatmaps, scroll-depth maps, and session recordings. The clearest signal came from the clicks that went nowhere. On the balance transfer page, close to a third of all clicks fired on elements that did nothing, and the hero headline and offer copy were pulling more clicks than most of the real buttons. On the rate tables, people clicked rate numbers and term labels the same way, expecting them to lead somewhere. Every one of those clicks was a user telling me what they wanted to do next.

The easy move was to guess and start linking. I decided to ask instead. I specified a one-question survey that fires when someone lingers on a rate page without acting: what are you trying to do next, see a payment estimate, apply, or compare. The answer decides where the rows link, and takes the assumption out of the strategy before a single element gets wired.

That same behavioral read drove the rest. I moved rates out of a buried Resources menu and into the account and loan flows, split the single rates page into individual URLs so each can rank and convert on its own, pulled a buried mobile calculator up toward the hero, and scoped a widget-level personalization framework that shifts CTAs by intent instead of maintaining separate pages.

The launch gets the attention. The version of the site that exists a year later is built from these smaller, evidenced decisions, each one traceable to something a real user did.