One Rocket Mortgage upsell tile, re-tested across five releases in about a month, roughly tripled its output, from an estimated 73 to 223 closed loans a month. No new traffic was bought. The tile simply got tested a second, third, fourth, and fifth time.
When you get preapproved for a mortgage, the lender hands you a letter you can show sellers. A Verified Approval Letter goes a step further: an underwriter has actually checked your income, your assets, and your credit, so the letter tells a seller your financing has been confirmed rather than guessed at. An offer carrying that letter is harder for a seller to pass over. Rocket Mortgage, among the country's largest home lenders, showed preapproved buyers a tile inviting them to upgrade their standard preapproval to the verified version.
The tile was already valuable ground. Around 43,000 people saw it every month, each of them already inside the funnel and already approved, and each upgrade fed a countable pipeline of closed loans. The share of those viewers who tapped the button, labeled Upgrade My Approval, sat near 9%. Real audience, real money behind the click, and a conversion rate with plenty of room in it.
The reflex growth move when a tile is only converting at 9% is to go find another surface: a new email, a new banner, one more page to carry the offer. We left the offer exactly where it was. The question we chased instead was whether that one tile, tested again and then again, could deliver more than any fresh placement would. What follows is five releases on a single tile, and a click rate that walked from 9% to 27% while the traffic behind it never moved.
Test one: four ways to ask
The first test was pure copy. The tile's headline read "Strengthen your next offer on a new home," a serviceable line that named the benefit in the most general way possible. We wrote three challengers, each pulling on a different reason a buyer might care. One leaned on certainty, the confidence a verified letter gives you at the table. One framed the upgrade as the obvious next step right after a preapproval. And one went straight at competition, positioning the verified offer as the one built to beat other bids on the house. Four arms, one live control, the same tile underneath all of them.
One of the three challengers won cleanly. Clicks on Upgrade My Approval rose 75.8% over the control at 99% confidence. Run through the funnel that sits behind the tile, that lift penciled out to an estimated 80 additional closed loans a month, taking the tile from an estimated 73 to 153. Another set of Rocket experiments found some of the program's biggest wins by reordering the application form, work I break down in the form reorder deep dive. This one moved an upsell tile by refusing to leave it alone.
Test two: a checkmark people already trusted
With sharper words banked, the next test changed a symbol instead of a sentence. People recognize a verification badge on sight, the small checkmark on social platforms that means an account is who it claims to be. We added a checkmark badge in that same visual language to the tile and ran it as a multivariate test against the wordmark that was already there. The badge won by 38% more clicks at 98% confidence, on the same 43,000 monthly viewers. Stacked on top of the copy win, it was worth an estimated 70 more closed loans a month, carrying the tile from 153 to 223.
A checkmark borrowed from social media did in half a second what a paragraph of mortgage copy could not.
The release that broke it, and the chart that caught it
Shipping five releases in about a month against a live control means other teams are shipping into the same product, and one of them clipped us. A separate platform project, nothing to do with this tile, went out mid-program and quietly broke the test. The click rate sank on the next reading. The following release found the fault, repaired it, and put the climb back on its line, which is the last step you see in the chart below.
Instrumentation that can catch damage is the same instrumentation that proves lift.
That is worth sitting with. A team reviewing this test only at the end of a quarter would have booked the whole month as flat, never knowing a live release had eaten a working experiment and a later one had won it back.
What compounding looks like
Each release built on the proven version underneath it. A win banked in one test became the baseline the following test had to clear, so the gains sat on top of one another rather than replacing each other.
The tile already had the traffic and the money attached. We tripled its output by testing it a second and third time instead of moving on.
This is the same discipline behind the wider Rocket program that the Rocket Mortgage case study covers, and it is why a growth operating system beats a calendar of redesigns. A system keeps returning to the surfaces that already work and asks them to do more, rather than chasing the next new thing while the proven ones sit at a fraction of their ceiling. When I run a growth audit and find a high-traffic tile or step converting in single digits with revenue behind it, that surface is almost always where the cheapest growth in the whole funnel is waiting.
Nothing here took a redesign, a fresh brand, or a fatter media budget. One tile, the same 43,000 people a month, and the willingness to test it a third and fourth and fifth time carried its output from an estimated 73 closed loans a month to 223. That is what a conversion program looks like when it stops hunting for surfaces and starts compounding on the ones it already owns.
If your funnel has a tile like this, a high-traffic surface with real money behind it and a mediocre click rate, that is precisely the kind of problem I get hired to fix. Let's talk.