A conversion case study: two A/B tests on one of the largest lead funnels on the internet.
Most businesses live with a version of the same problem. Traffic arrives, a multi-step form sits between that traffic and revenue, and somewhere in the middle people quietly give up. Fixing that middle is usually the cheapest growth available, because everyone standing in it has already been paid for.
This is the story of that fix at extreme scale. Two controlled experiments on the form that starts Rocket Mortgage’s application process drove close to $2 billion in incremental loan volume, without adding a dollar of acquisition spend.
The changes themselves were almost embarrassingly simple. What made them bankable was the rigor: each ran as an A/B experiment against the live flow, so every point of lift was measured against a control before anyone claimed a win.
The funnel that was leaking money
Rocket Mortgage begins its mortgage process online. Before a prospect ever speaks with a banker, they work through a questionnaire covering their goal, their property, and their finances. Internally the flow was called Launchpad. Functionally it is a lead form, the same species as the one on almost any website, with a mortgage waiting at the end.
A completed questionnaire becomes a lead. A good lead progresses to a hard credit pull, then underwriting, then a closed loan. Heavy paid spend kept the top of that funnel full. The leak was in the middle: too many people started the questionnaire and never finished it, and every one of them had already been paid for.
High abandonment on a high-intent funnel pointed at two levers. Reduce the friction that made people quit, and guide the ones who stayed toward the option they came for. We ran one experiment on each.
Experiment one: reorder the questions
Watching where people quit told the story. The questionnaire front-loaded its hardest questions. Income, personal financial details, and open-ended answers all appeared early, before a prospect had built any momentum. It asked a stranger for their income almost in the first breath.
Every question in a form is a small ask, and the order of the asks decides how many get answered. Stack the heavy ones up front and people bail before they are invested enough to push through.
The fix sounds too simple to matter: change the order. Easy, low-stakes questions moved to the front to build momentum. The personal and financial asks moved later, once a prospect was engaged. Short explanations were added at the anxious moments, so a sensitive question arrived with a reason attached. Nothing was redesigned and nothing new was built.
Lead completion rose +4% and the loan-closing rate rose +12%, worth roughly $1.5 billion in incremental loan volume. That is what leverage looks like in a high-volume funnel: a resequenced form, built in days, worth more than a billion dollars.
The closing rate moving three times as much as completion is worth pausing on. Reordering changed who finished. Prospects who came through the friendlier flow were more invested, and that showed up downstream as more closed loans.
Experiment two: reorder the loan options
Early in the questionnaire, prospects pick what they came for: buying a home, refinancing an existing loan, or pulling cash out of the equity in their home. The screen presented every option with equal visual weight, leaving each prospect to scan a menu and sort themselves.
Demand had stopped being equal. Interest in cash-out loans was climbing, and a large share of prospects arrived already knowing that was their goal. Making them hunt through a neutral menu added a decision at exactly the wrong moment.
We tested two variants against the control. One added a “Trending” badge to the cash-out option. The other reordered the menu, moving cash-out and refinance to the top and purchase lower down.
The reorder won. Lead completion rose +5% and the loan-closing rate rose +18.8%, worth about $1.1 billion in incremental loan volume. When most of your demand already leans one way, arranging the menu to match removes a decision from the path.
The results
Both experiments ran against live control groups, so the lifts are true incremental gains, and the dollar figures apply those lifts to the funnel’s loan volume. Because both tests improved the same funnel, the combined impact is stated as close to $2 billion rather than a straight sum.
- Question reorder: lead completion +4%, loan-closing rate +12%, roughly $1.5B in incremental loan volume.
- Loan-option reorder: lead completion +5%, loan-closing rate +18.8%, roughly $1.1B in incremental loan volume.
- Close to $2B combined, with no increase in acquisition spend.
- Two experiments, no redesign, no new product. Sequencing and hierarchy did all the work.
If your funnel leaks the same way
You do not need Rocket Mortgage’s volume for this to matter. Any multi-step form, signup, or checkout with drop-off responds to the same levers: open with the easy asks, hold the sensitive ones until people are invested, and arrange choices to match the intent your data already shows.
The discipline is the part that scales. Treat every change as an experiment, measure it against a control, and the wins compound while the losers get killed cheaply. If your traffic looks healthy and the middle of your funnel quietly leaks, let’s talk.