A change can take 40 minutes of hands-on work yet nearly five hours to merge. Activity ratio makes that gap visible: divide time spent actively progressing the work by the elapsed time across the flow being measured. A higher ratio means more of that time is productive movement; a lower one points to waiting, reruns, queues, or context switching—not necessarily slow developers. section 5.4.2 section 9.1
A small change with a long wait
A routine code change takes 40 minutes of activity across implementation, tests, CI, review, and merge. From start to merge, 288 minutes pass: 40 ÷ 288 = 0.14, or 14%. The flaky CI rerun adds waiting; the review queue accounts for much more. Reducing the longest delay offers more leverage than asking for faster coding. section 5.4.2
Table 5.2 Activity ratio analysis of a simple workflow
|
Stage
|
Activity time
|
Waiting time
|
Cycle time
|
Activity ratio
|
|
Implement change
|
30
|
0
|
30
|
1.00
|
|
Run local tests
|
1
|
0
|
1
|
1.00
|
|
CI pipeline (1st run)
|
1
|
7
|
8
|
0.13
|
|
CI rerun (2nd run)
|
1
|
7
|
8
|
0.13
|
|
Code review
|
6
|
234
|
240
|
0.03
|
|
Merge
|
1
|
0
|
1
|
1.00
|
|
Totals
|
40
|
248
|
288
|
0.14
|
Compare activity and waiting time at each stage to see why the overall ratio is much lower than the coding-stage ratio. section 5.4.2
Name the measurement boundary before comparing ratios. For a change from first edit to merge, use activity time divided by cycle time over that interval. For a feature from request to delivery, include the earlier backlog wait in lead time. If estimating in working hours, keep the numerator and denominator on the same basis; one exercise uses active hours divided by eight working hours per lead-time day. Ratios calculated across different boundaries are not interchangeable. section 5.8 section 9.5
Use the ratio to find a delay worth changing:
- Trace one completed item with ticket, commit, pull-request, and deployment timestamps. Estimate active work separately from time spent waiting at each stage. section 6.1.2 section 6.6
- Look for a station where activity is a small fraction of elapsed time. Three hours of review spread over four days suggests a review queue, not three days of review work. section 9.2.3
- Change one constraint, then measure the next item. Review windows, for example, reduced review elapsed time from four days to 1.5 while active review remained three hours. section 9.2.3 section 6.6
The same question applies beyond tickets. In a checkout path, 65 ms of processing within 1,050 ms of total latency yields a 6.2% ratio; database contention and an external API account for much of the wait. Measure where time goes before optimizing the active step. section 11.3
