Canvas app performance: find what's slow and fix it at the source

Slow canvas apps are rarely slow for mysterious reasons — Microsoft's performance documentation names the same handful of patterns again and again. This guide maps symptom to cause to documented fix.

Measure first

Two built-in tools turn "it feels slow" into a diagnosis:

• Monitor logs every data call and key activity as the app runs — including a session attached to a phone, so field-reported slowness can be seen rather than imagined.
• Performance insights and App Checker in Studio flag known anti-patterns; cross-screen references surface under "Inefficient Delay Loading".

Fix what the trace shows, not what a checklist guesses — and when a fix lands, re-measure the same journey so the claim "it's faster now" belongs to evidence someone actually gathered.

Fix the start

App.OnStart is imperative: everything in it must finish, in order, before the first screen appears. Microsoft's guidance:

• Move variable setup to named formulas, which load lazily and don't block the first paint.
• Replace Navigate inside OnStart with App.StartScreen — declarative, and it lets the first screen show before OnStart completes.
• Defer heavy data loading to a non-blocking step after the screen is visible.

// Instead of:
App.OnStart = Collect(OrdersCache, Orders);
If(Param("AdminMode") = "1", Navigate(AdminScreen), Navigate(HomeScreen))

// Declare the start screen and let it paint:
App.StartScreen = If(Param("AdminMode") = "1", AdminScreen, HomeScreen)

Hunt the data patterns

The three documented heavyweights:

• Select N+1: a gallery whose rows each fire a LookUp turns one screen into dozens of calls. Watch Monitor for repeated calls to the same source; fix by batching into one query, using a view that already includes the related columns, or reshaping the data so the join happens at the source. Microsoft's example case cut a five-minute load to under ten seconds by fixing the join at the source.
• Collections as a crutch: ClearCollect of a large table blocks the UI until every row lands. Collections are for small data you compute against repeatedly (a basket, not a backlog); read-only data belongs directly in the gallery's Items, delegated to the source. Delegation itself is its own discipline — our delegation cheat sheet covers which functions push work to the server.
• Power Automate as a data loader: each flow call costs roughly 0.6 seconds just to start, before any work happens. Microsoft is blunt that the worst-performing apps overuse this; query the source directly, or move server-side logic into Dataverse views or low-code plugins that return finished results.

Remove cross-screen references

A label on Screen A reading a control on Screen B forces Power Apps to load both screens before A appears — multiplied across an app, that's the "every screen is slow" signature. The documented cure: hold shared state in variables or collections, pass values through navigation parameters, and keep each screen self-contained. On phones this also fights memory-driven app restarts, which Microsoft ties directly to how the app is authored.

Sources

• Identify and mitigate canvas app performance issues — Microsoft Learn
• Troubleshoot Power Apps canvas app performance issues — Microsoft Learn
• Optimize app or page load for peak performance — Microsoft Learn
• Power Apps coding guidelines: code optimization — Microsoft Learn
• Prevent canvas app restarts in the Power Apps mobile app — Microsoft Learn