How long does a unit take from start to finish?
№ 205 · Cycle time distribution
Definition
MetricElapsed time per unit from start to finish, shown as a distribution rather than an average
Unithours or days, reported at p50 and p90
Histogram of days from opportunity creation to closed-won across deals. Right-skewed; the long tail signals stalled deals. Build from CRM date stamps per stage. Cut off the tail (force-close or kill stalled deals). Example: median 32 days; long tail at 180+ days (stuck deals).
- CRM
- Customer relationship management system. The system of record for accounts, contacts and pipeline.
Benchmarks
| Bottom 30% | Median | Top 30% |
|---|---|---|
| p90 above 5 times p50 | p90 around 2.5 to 3 times p50 | p90 under 1.5 times p50 |
Spread ratio synthesised across warehouse and software delivery cycle time data, 2024 to 2026. Absolute anchors: warehouse dock-to-stock best-in-class under 3 hours (WERC DC Measures 2024); software change lead time under one day for DORA's elite cluster against one to six months for low performers. · Absolute cycle times are not transferable across processes, which is why the bands above are expressed as a spread ratio. The ratio is a synthesis rather than a single published distribution, so treat it as a working rule and rely on the absolute category anchors where they apply. The main definitional trap is where the clock starts. Dock-to-stock runs from goods arriving to being recorded in inventory, customer order cycle time runs from order placement to delivery and includes weekends, and software lead time for changes runs from commit to production. Each excludes queue time that the others include. APQC publishes 25th, 50th and 75th percentile values for customer order cycle time in days through its Open Standards Benchmarking programme, but the values sit behind membership and could not be verified for this entry, so only the measure definition is used here.
| Category | Bottom 30% | Median | Top 30% |
|---|---|---|---|
| Warehouse dock-to-stock | over 24 hours | not separately published in the public summary | under 3 hours |
| Software change lead time (commit to production) | 1 to 6 months | 1 week to 1 month | under 1 day |
When it looks bad
The histogram is bimodal, a tight cluster of fast units plus a second hump far to the right, which means two different processes are running and only one of them is being managed.
p50 is 6 hours and p90 is 61 hours. The second hump is every order that triggers a manual credit check, and it never appears in the average.
What to do about it
- Publish p50 and p90 and drop the mean from the operating review. The tail carries the cost and the complaints, and averaging is the standard mechanism by which it disappears from view (Lorikeet, 2026).
- Find the second hump and name its cause before optimising anything. Bimodality almost always means an exception path with different handling, and making the main path faster does nothing for it.
- Attack queue time before touch time. Best-in-class dock-to-stock is under 3 hours and has not improved year over year across the WERC panel, and most of the distance to that number in a typical operation is waiting rather than working (WERC DC Measures, 2024).
- For software delivery, cut batch size rather than adding reviewers. DORA's elite cluster achieves change lead time under one day while low performers take one to six months, and the separating variable is deployment frequency and batch size, not individual speed (DORA 2024, via Taskade and Optimal).
Sources
- Warehousing Education and Research Council, reported by Hyster-Yale Best-in-class dock-to-stock cycle time is under 3 hours, unchanged year over year, indicating warehouses have maintained but not improved performance. Dock-to-stock ranked third among the most-tracked DC metrics in 2024. scg-lm.s3.amazonaws.com ↗
- DORA (Google Cloud), reported by Optimal Elite performers hold lead time for changes under one day and restore service in under one hour. High performers deploy once per day to once per week, medium once per week to once per month, low less than once per month. getoptimal.ai ↗
- Taskade, reporting DORA Elite teams deploy on demand with sub-one-day lead time and roughly 5% change failure rate, while low performers can take one to six months to ship a single change. That is a spread of two orders of magnitude on the same metric. taskade.com ↗
- Octopus Deploy, reporting DORA Elite performers show change lead times 127 times faster than low performers and recover from failed deployment 2,293 times faster. The high performance cluster shrank from 31% of respondents in 2023 to 22% in 2024. octopus.com ↗
- APQC Defines customer order cycle time as average days including weekends between order placement and delivery, equal to source cycle time plus make cycle time plus deliver cycle time, and publishes it at 25th, 50th and 75th percentile of the peer group. Percentile values require membership and were not accessible for verification. apqc.org ↗
- Lorikeet Recommends reporting median rather than mean because a small number of extremely delayed items inflates the average and masks the typical experience, and tracking the 90th percentile as the worst-case view for 10% of customers. lorikeetcx.ai ↗
- Fabrico Typical discrete manufacturing bands: below 40% poor, 40 to 60% typical, 60 to 85% good to high, 85%+ world class. Speed loss and micro-stops are a principal driver of the gap, and both show up as cycle time variance rather than as average slowdown. fabrico.io ↗
- Symestic Manually tracked performance is systematically 8 to 12 percentage points better than automatically measured performance, because micro-stops, short downtimes and optimistic cycle time assumptions are invisible in manual tracking. First accurate baselines always look worse. symestic.com ↗