Anatomy of a playable that lost to its own video

The most common playable failure of 2026 is not a broken build or a rejected spec. It is a playable that was commissioned because a video worked, faithfully reproduces that video's mechanic, and then converts worse than the video it copied. The instinctive reading is that the format is wrong for the game. The better reading is that the team asked the playable to do the video's job and then measured it as if it had a job of its own.

The thesis is uncomfortable for playable producers, so it is worth stating plainly: a playable derived from a winning video should not be expected to beat that video on install rate, and a brief that sets that expectation has already failed. The two formats sell different things. The video sells a feeling in fifteen seconds to someone who has not opted in. The playable sells competence to someone who has already stopped scrolling. If you copy the video's beat sheet into the playable, you pay production cost and load time to deliver a slower version of the same promise.

What the video was actually doing

Start by decomposing the winning video honestly. A high-performing mechanic video for a puzzle or merge title usually works on three layers at once: a visible failure or near miss in the first two seconds, a satisfying resolution the viewer did not have to earn, and a cadence of on-screen text that carries the narrative even with sound off.

None of those three layers survives translation into a playable intact. The near miss becomes the player's own mistake, which reads as friction rather than drama. The unearned resolution becomes an earned one, which is slower and depends on the player being competent at a mechanic they saw for the first time four seconds ago. The text cadence collides with input, because a thumb on the screen is not reading captions.

The mechanism here is attention economics rather than design taste. A video spends the viewer's attention for them. A playable asks the viewer to spend it. That is a different transaction with a different conversion curve, and the transaction only pays off when the thing being demonstrated is something the viewer wants to prove to themselves they can do. Reproducing the video's narrative in interactive form removes the proof and keeps the cost.

The second-order effect nobody budgets for

There is a subtler cost to the copy-the-video approach that rarely shows up in the creative review, because it shows up downstream in the cohort rather than in the ad metrics.

When a playable reproduces a video's simplified mechanic, it also reproduces the video's simplified promise. Players who install from the video have a vague expectation and adjust quickly once in the game. Players who install from the playable have spent thirty seconds actually performing a version of the game that does not exist. They arrive with a precise expectation, and the first session either honours it or breaks it. Where the playable oversimplified, day one retention on the playable cohort can run below the video cohort even when the install rate looks comparable.

That is the trade-off most coverage misses. A playable is a retention instrument dressed up as an acquisition instrument. Its value is in filtering and calibrating, not in maximising clicks. Judging it purely on install rate against its parent video guarantees that the honest playable, the one that shows real difficulty, loses to the flattering one, and the flattering one then quietly degrades cohort quality. UA Ledger covered the design side of this in Fake difficulty and real difficulty in playable design. The measurement side is where the decision actually gets made.

An illustrative teardown

Consider an illustrative case with invented but realistic numbers. A merge title has a hero video converting at an install-per-impression rate of around one per cent on a rewarded placement. The team commissions a playable that walks through the same three merges shown in the video, with an end card at the twenty-second mark.

The playable comes back at roughly 0.7 per cent. The review concludes that playables do not work for this title. But the engagement data inside the playable, if anyone looked, would show something more specific: a high interaction rate on the first merge, a sharp drop at the second, and most of the installs coming from players who never reached the end card and tapped the persistent install button early.

Read correctly, that pattern says the playable's opening worked, the second beat introduced a decision the video had made for the viewer, and the install button did the job the end card was supposed to do. The fix is not to abandon the format. It is to cut the second beat, move the end card to eight seconds, and let the persistent button carry the conversion. In this illustrative case the rebuilt playable does not necessarily beat the video on install rate, but the day seven retention on its cohort is where the case for it is made or lost.

A decision rule for the brief

Before commissioning a playable from a video, apply a simple three-question test.

• Is there a moment in the video where the viewer would plausibly say "I could do that"? If not, the video is selling spectacle, and spectacle does not become better when it is interactive.
• Can the core interaction be understood without any on-screen text? If the video needs captions to make sense, the playable will need a tutorial, and tutorials are where interaction rates die.
• Does the game actually feel like the playable in the first session? If the answer is no, decide now whether you are willing to accept a weaker cohort in exchange for a stronger click.

If the video fails the first two questions, do not build the playable from it. Build a playable from a different moment in the game entirely, and let the video keep doing its job at the top of the funnel. The two formats are not competitors for the same slot. They are consecutive stages, and the mistake is running them as a head-to-head test at all.

Measuring the loss correctly

When the playable does lose on the top-line number, the review should produce four figures before anyone calls the format. Interaction rate at first tap. Completion rate to end card. Share of installs from the persistent button versus the end card. And, two weeks later, day seven retention of the playable cohort against the video cohort on the same placement.

A playable with a high first-tap rate and low completion has a pacing problem, which is a production fix. A playable with low first-tap has a hook problem, which is a concept fix. A playable that loses on installs but wins on day seven retention is not losing at all, and the buying team should be told to value that cohort accordingly rather than pausing the asset because a dashboard sorted it to the bottom.

The last figure is the one that changes behaviour. Once a team has seen a playable cohort outperform on retention despite a weaker install rate, the question in the next brief stops being "can the playable beat the video" and becomes "what should the playable filter for". That is the correct question, and it is one the video can never answer.

Related archive reading

These articles provide related context and remain subject to their stated review status.

• AppLovin AI video creative: its own generation tool
• When a winning video needs a playable end card