Four and a half years running experiments on one of the largest mortgage funnels in America, and what a disciplined testing engine is actually worth.
Rocket Mortgage runs one of the biggest mortgage funnels in the country. Millions of visitors a year land on a multi-step questionnaire called Launchpad, answer questions about their finances, and either become loans or disappear.
At that scale, a one percent improvement is worth millions. My job was to find those improvements on purpose, week after week, instead of waiting for them to happen by accident.
An experiment a week, every week
The core of this story is a cadence, and the cadence came first. By 2024 the program was launching an experiment a week. In a single quarter we shipped 24 tests, and 42% of them beat their control. The industry norm for win rates sits closer to 25%.
Every test ran against a live control. Every winner was sized by its projected annual impact on four numbers: leads, completed credit applications, closed loans, and revenue. A completed credit application, what we called hard credit, was the strongest signal in the funnel, because a visitor who lets you pull their credit is a visitor who intends to get a loan.
That measurement discipline matters more than any single test. It means every number in this case study comes from a controlled experiment, and every projection is labeled as one.
Experiment: reorder the questions
The questionnaire opened with the questions people find hardest to answer. Income, debts, the state of their credit. Sensitive questions, asked of a stranger, in the first thirty seconds.
We moved the easy, low-stakes questions to the front and held the sensitive ones until people had already invested a few minutes. The psychology is simple: each answered question is a small commitment, and people who have made ten small commitments will make a bigger one.
Leads rose roughly 3%. Completed credit applications rose 12% to 15.6% depending on the flow. In the refinance flow alone, the win projected to about $1.3M in annualized revenue.
Nothing about the questions changed. Only the order did.
Experiment: the screen nobody wanted to touch
Every funnel has a screen the whole company is afraid of. At Rocket it was the communication consent screen, the one that asks permission to contact you. It carried the highest legal risk in the funnel, so by default nobody tested it.
We partnered directly with the legal team and tested it anyway, with compliant, clearer consent patterns. To understand what users actually experienced, we wired session replay into the experiment so we could watch real test sessions, a first for the program.
One earlier consent test on that screen lifted completed credit applications 11%. The clearer consent patterns became the single biggest win of the quarter: a projected $15.4M in annualized revenue and more than 2,300 additional closed loans a year. A second iteration added a projected $2.2M on top.
The screens everyone is afraid to test are usually the ones carrying the most unclaimed money, precisely because nobody has ever tested them.
Experiment: badges and input fields
Two smaller wins show the range of the program.
A trending badge on the loan options, redesigned around momentum, lifted leads 6% and completed credit applications 16.4%. People hesitate less when they can see what others are choosing.
And a plain input field redesign on the highest traffic form projected about $3.5M in annualized revenue and roughly 1,000 additional closed loans a year. Not a new offer, not a new page. Better input fields.
What it added up to
In the single quarter shown above, the winning experiments projected roughly $24M in annualized revenue and about 4,300 additional closed loans, from testing alone, with no added acquisition spend.
Across four and a half years and more than 225 growth opportunities, the program's wins added an estimated 10,000 incremental closed loans, roughly $2 billion in loan volume.
The lesson underneath the numbers: none of this came from a redesign, a rebrand, or a bigger budget. It came from a system that asked a small question every week and measured the answer honestly.
If your funnel leaks the same way
Most multi-step funnels leak in the same three places: sensitive questions asked too early, scary screens nobody has ever tested, and forms that were designed once and never touched again.
You don't need Rocket's traffic to run this playbook. You need a live control, one clear success metric, and the discipline to ship a test a week. The growth audits I run start by finding the step where your funnel leaks, and case studies like YouMail and FX Replay show what happens once you fix it.
If two sets of your numbers disagree about whether marketing is working, get in touch.