Predictive Preloading

Pley games learn how they're played, and load ahead of the player. Predictive Preloading uses patterns from earlier play sessions to download content before a player needs it, helping reduce waits when starting a game or reaching the next level, screen, or asset.

How it works

As players move through a game, Pley learns which content they need and uses that knowledge to load ahead in subsequent sessions. The goal is to have content already downloaded on the player's device when the game requests it.

What happens after a new build?

Predictive Preloading needs about a day to learn a game again after a new build ships. Loading times can temporarily increase during this learning period, so allow time for the new build to learn before evaluating its loading performance.

Measured impact on first-time players

Following the October 8, 2026 preloading release, Pley compared loading times for first-time players over 18 days before the release and 10 days after it. The rollout day itself was excluded. The analysis covered Pley's own player traffic and excluded CrazyGames traffic.

The measurement was the time from opening a game until it was ready to play:

• Typical load (P50): the median wait. Half of first-time players loaded faster than this.
• Slow loads (P90): the 90th-percentile wait. Around 10% of first-time players waited longer than this.

Shorter, more consistent loading times

In our largest measured sample, covering approximately 24,900 first-time players, typical loading waits were 18% shorter and slow loading waits were 40% shorter. Slow loads fell from 64 to 38 seconds, a reduction of 26 seconds.

Loading times also became more consistent in that sample. Daily P90 values previously ranged from 51 to 89 seconds; after the release, they stayed between 37 and 39 seconds throughout the measured period.

Across smaller samples, some games saw typical loading waits reduced by up to 65%, and slow loading waits reduced by up to 78%. These are the highest reported improvements in the comparison, not the average across games; smaller samples are less precise.

Results vary by game. A game that already loaded in about nine seconds showed little change. These are observed before-and-after results, not guaranteed improvements.

What to expect for your game

• Games with substantial download waits have more room to benefit. Games already loading in under ten seconds may spend most of that time starting up rather than downloading content.
• Returning players can also benefit, but the measurements above cover first-time players only.
• Evaluate performance after the learning period for each new build, rather than immediately after publishing it.

Measurement notes

The results were reported on October 19, 2026. The before-and-after figures are player-count-weighted averages of daily P50 and P90 values, rather than percentiles calculated over all sessions together. Daily reporting windows ran from 10:00 to 10:00 Barcelona time; October 19 was a partial day. Games with too few first-time players after the release were excluded.

For a breakdown of your build's launch files, runtime bundles, and assets, see Build Analysis.