Measure automation savings without double counting

Define the work unit and baseline

Choose one workflow, beneficiary role and reporting period. Define one distinct eligible completion: a reconciled subscription, handled request or reviewed device record. Retries and monitoring polls count only if they replace the manual activity under comparison.

Observe comparable manual work where possible. Record active minutes, including the ordinary difficult cases. Waiting for a server or vendor is elapsed time; it belongs in active effort only when somebody must attend to it. If the old process cannot be observed, label the baseline estimate and record how it was formed.

Conduit's advanced workflows provide examples of reporting and triage activities.^[1] Their scope can help frame a work unit. Your own observations establish the baseline. Adam Hannemann's shiny-tool discussion similarly starts with understanding the work before choosing the product.^[2]

Use five inputs with non-overlapping effort

• N: distinct eligible units in the period.
• M: active manual minutes per unit before the change.
• H: routine human minutes per unit afterward, including review and approval.
• E: extra exception, fallback and rework minutes for the period.
• O: recurring monitoring and maintenance minutes allocated to this workflow.

Net recurring hours saved = (N × (M − H) − E − O) ÷ 60.

All effort inputs in that formula are minutes. Keep one-time setup and training separately in hours. Review already included in H stays out of O; fallback already charged in E stays out of the other categories. Use nonnegative effort inputs, while preserving a negative net saving if the change adds work.

Work a fictional month

There are 120 units, 8 manual minutes per unit, 1.5 routine human minutes afterward, 30 exception minutes and 90 maintenance minutes.

Manual effort is 16 hours. Remaining human effort is 5 hours. Net recurring savings are 11 hours. Eight setup hours leave 3 first-period hours.

For a team with 600 available and 610 workload hours, current aggregate headroom is −10. The recurring scenario becomes +1; the first-period scenario is −7. All these inputs are fictional estimates. The protected capacity sample uses the same arithmetic and includes unknown, existing-automation and added-work cases.

Decide whether the credit is incremental

If the automation already operates, current workload already reflects it. Give it zero incremental credit against that workload. Historical effort avoided can explain the baseline, but adding it again manufactures spare time.

For a proposed change, confirm the eligible work is still in the baseline and no other initiative claims it. Two scripts replacing the same categorization step cannot each take the full saving. Allocate it once or compare the scripts as alternatives.

Check who can use the time and when. Account-management time cannot fill an engineer's on-call shift. Small fragments can be useful without providing a block large enough for a project. Hours saved become cash savings only through a separately supported cost change.

Collect pilot evidence where the work lives

Use PSA task/time references, the automation's run evidence and a maintenance log. A successful RMM run is a useful reference to examine; the outcome test and remaining human effort need separate records. Autotask, HaloPSA, NinjaOne and Datto RMM are possible examples, not required systems or claims about a specific report. Verify the evidence available in your version.

Run a bounded, authorized pilot. Record ordinary runs, failures, fallback and maintenance. Have a reviewer check the outcome rather than relying only on the workflow's own completion flag. Retain the failed cases even when the average looks good.