Your Test Backlog Already Failed Once. You Just Forgot.
Every conversion team I have seen has the same quiet problem, and almost none of them can see it. They test the same idea more than once, months apart, and treat the second run as a fresh discovery. The person who ran the first version has moved teams, or the result was never written down properly, or it lived in a deck that nobody opens. So the idea comes back around, gets built again, and fails again, or wins again, teaching a lesson the team already paid for.
This is not a discipline problem. It is a memory problem. And memory is fixable.
Testing without a record is not learning
There is a comforting story we tell ourselves: that a testing program accumulates knowledge over time. Run enough tests and you will understand your audience. It is not automatically true. A program only accumulates knowledge if it accumulates a record. Without one, it accumulates activity, which is a different and much less valuable thing.
Think about what actually gets remembered from a test. The headline result, maybe, if it was dramatic. The nuance, the why, the segment where it behaved differently, the reason it was close: all of that evaporates within a quarter. So the next person to have the same idea starts from zero, because zero is all that was preserved.
Build the record before you build the test
The fix is unglamorous. Every concluded test gets written down in a consistent structure, and the record gets consulted before anything new gets built.
At minimum, capture the hypothesis, what changed, the metric that mattered, the result including how close it was, and one or two sentences on what you believe you learned. That last part is the one people skip, and it is the one that pays off later. A number tells you what happened. The learning tells the next person why, so they do not have to reverse-engineer it.
Once you have a body of these, checking a new idea against them takes minutes and saves weeks. Before you commit to building, you search the record for anything close. Three things can happen, and all three are useful.
The three answers the record gives you
Sometimes you find the exact test, already run. If it lost, you just saved a full cycle of traffic and build time. If it won, you have a much stronger case for rolling it out permanently rather than testing it a third time. Either way, you skip the work.
Sometimes you find a near neighbour. Not the same test, but the same lever pulled in a slightly different way. This is the most valuable case, because it lets you weight the new idea. If that lever has won repeatedly for this audience, the new idea inherits confidence. If it has failed repeatedly, you either drop it or design around the reason it kept failing.
Sometimes you find nothing, which is genuinely new territory. Now you know it is new, which is worth knowing, and you can prioritise it as a real discovery rather than an accidental repeat.
Compounding is the whole point
A team with a record gets smarter every quarter, because every result makes the next decision better informed. A team without one runs in place, rediscovering the same handful of lessons on a loop and mistaking the motion for progress.
The unglamorous truth is that the highest-leverage thing most conversion teams could do this year is not a clever test. It is writing down the tests they already ran, in a form they can actually search, and reading it before they build the next one. Your backlog is haunted by its own past. Give it a memory, and the ghosts turn into an advantage.