Playbook/Learning Phase Management

Learning Phase Management

Learning is naturally volatile. Protect signal, fund each concept properly, and intervene only when the test is broken or financially invalid.

1

Protect learning

  • Do not change audience, countries, optimization event, bid strategy, budget, placements, creative, copy, concept, or status.
  • Do not repeatedly pause and reactivate ad sets.
  • Give every concept the same event, starting budget, broad markets, and at least one-third of LTV per day.
  • Compare ad sets only after equivalent time and spend.
2

Do not force equal creative delivery

The platform may concentrate spend on one execution. The unit being tested is the concept ad set, not each isolated ad.

  • Do not pause the highest-spend ad to force distribution.
  • Do not duplicate under-delivering ads inside the test.
  • Do not raise one testing budget because it looks promising.
  • Do not judge an execution only because it received little early spend.
3

When to intervene

SituationAction
Tracking, destination, checkout, or paywall is brokenPause, fix, and relaunch cleanly.
Ads are rejected or the account cannot spendResolve the delivery problem.
Normal delivery with volatile results or uneven ad spendLeave it alone.
Not enough spend to reach the decision thresholdContinue the test.
Enough spend with no credible path to target CACPause the concept.
Trials have not maturedHold the decision.
4

Learning limited

  • Confirm tracking, event volume, and budget.
  • Reduce simultaneous concepts if budget is spread too thin.
  • Move to the next-deepest reliable event when the current signal lacks volume.
  • Do not merge unrelated concepts merely to escape learning limited.
5

Classify the concept

VerdictMeaning
WinnerAcceptable cost with healthy downstream quality.
HoldCredible early signal; revenue cohort is not mature.
LoserHealthy test with no realistic path to target CAC.
InvalidTechnical or delivery failure prevented a fair test.
6

Manage winners

  • Duplicate the full winner into the Winning campaign and pause the Testing original.
  • Preserve concept, audience, creative, and event.
  • Increase a stable winner by no more than 20%.
  • Leave it untouched for three full days before the next decision.
  • Log date, setting, reason, and performance before and after every change.

Takeaway

Patience, sufficient budget, and one controlled change at a time protect the meaning of the data.
Learning Phase Management | UpLyft Playbook