Glosarium
Contents

Frontend Development › Web Performance

Real User Monitoring (RUM)

Measuring performance from real users' browsers.

Backend Engineer · SeniorFrontend Engineer · Mid-level

Also known as: real user monitoring, rum, field data

Real User Monitoring (RUM) measures performance from real visits — actual devices, networks, geographies and interactions — via in-page instrumentation reporting paint timings, interaction latencies, layout shifts and custom marks. Where lab tests show what’s possible, RUM shows what happens, including the long tail labs never reproduce.

lab:  flagship phone, cable, warm cache → optimistic best case
RUM:  p75 across real devices/networks → the experience to optimise

Its metrics (percentiles, segmented by device, connection, region, route) direct optimisation at what users actually suffer — usually distant networks, mid-tier devices and specific heavy routes — rather than what developers’ machines show.

The classic mistakes:

  • Lab-only optimisation. Perfect Lighthouse scores alongside terrible field data is the norm, not the exception. Lab for diagnosis, RUM for verdicts.
  • Mean-watching. Averages hide the suffering tail. Track p75/p90+, segmented — the tail is where users churn.
  • Unsegmented data. Global aggregates blend incomparable experiences. Segment by device class, network, geography and route before concluding anything.
  • No interaction metrics. Paint timings without input latency miss responsiveness — often the actual complaint. Measure interactions, not just loads.
  • Sampling blindness. 100% collection on high traffic is expensive; naive 1% sampling misses rare-but-awful segments. Sample intelligently (all of the tail, fraction of the head).
  • Privacy negligence. Performance beacons carrying URLs, IPs and device details need consent handling and minimisation like any telemetry.
  • Dashboard without budget. RUM that nobody gates on rots into wallpaper. Tie budgets to alerts and release gates (see performance budget).

The practice: instrument real visits, percentile and segment relentlessly, optimise what the field shows, gate releases on budgets. Reality outranks simulation — measure where users are.