Blog

Why 100% utilization makes engineering teams slower

· 5 min read

A fully booked team looks efficient on a spreadsheet. Every engineer is assigned, every sprint is full, nobody is idle. It is also the reason a two-day request takes three weeks.

The math of waiting

When work arrives unpredictably and varies in size — which describes almost all engineering work — queues form. The simplest queueing model says waiting time is proportional to ρ ÷ (1 − ρ), where ρ is utilization.

  • At 50% utilization the factor is 1.
  • At 80% it is 4.
  • At 90% it is 9.
  • At 95% it is 19.

The curve is steep where it matters. Going from 80% to 90% busy sounds like a 10-point efficiency gain, and it more than doubles how long work waits. From 90% to 95% doubles it again.

Roads are the familiar example. A road at half capacity flows freely, and one near full capacity jams, even though the extra cars are few. Nothing is wrong with the drivers. There is no room left to absorb variation.

What it looks like in a team

  • Small requests from other teams wait behind large planned projects.
  • An incident or a sick day knocks the whole plan over, because nothing can give.
  • People multitask to cope, which adds context switching and slows everything further.
  • Other teams stop asking and build their own workarounds.

Slack is what you are buying

Unassigned capacity is easy to read as waste. In practice it buys responsiveness, the way a fire station staffed below full utilization can still answer a call. How much to keep depends on how unpredictable the work is. A team with a stable, well-specified backlog can run fuller. A platform, reliability or support team that exists to respond needs real headroom.

Plan with real capacity

Much apparent over-utilization is a planning error. An engineer does not have 40 project hours in a week. Meetings, code review, on-call, interviews, support and leave take a large share. Measure your team's focus factor — project hours divided by total hours — from the last few months, and plan with that. Six engineers at 40 hours and a focus factor of 0.6 have 144 project hours a week, not 240.

What to do on Monday

  • Plan sprints to about 80% of measured capacity and leave the rest unassigned.
  • Reserve an explicit share for unplanned and cross-team requests.
  • Limit work in progress, so that finishing takes priority over starting.
  • Track how long requests wait, not only how busy people are.

The objection is always that slack means paying people to do nothing. In practice the time fills with the work that never makes it onto a roadmap: reviews, documentation, small fixes, helping a colleague. The team ships sooner because nothing is stuck in a queue.

Try it — Utilization & queues

Queue wait factor

4.0×

Slack

20%

The wait factor is ρ ÷ (1 − ρ): how long work waits relative to how long it takes to do. From 80% to 90% utilization it more than doubles.

1.0×
50%
1.5×
60%
2×
70%
4×
80%
6×
85%
9×
90%
19×
95%

Wait factor at each utilization level. The curve is why a team planned at 100% is slow to respond to anything unplanned.

Practice this in a lesson

Each engMBA lesson pairs a short explanation with practice questions that grade themselves. Free, with no signup.

See the full curriculum

Keep reading