Choose the unit of work.
Start with a request, document, onboarding or other unit that has a consistent beginning and end. Define which items enter the measurement and when they have had enough time to complete.
A faster-looking dashboard can hide a changed definition. If the old process counted all requests but the new process excludes difficult cases, the comparison does not establish improvement.
Build the baseline from real work.
Measure volume, manual effort, completion or turnaround, and exception handling. Use a representative period and record changes in staffing, workload and request mix.
For effort, observe a sample of actual tasks rather than multiplying a guessed time by every record. Time spent waiting and time spent actively working are different measures; both may matter.
Separate the benefit categories.
Different benefits support different business decisions. Avoid adding the same improvement twice under different labels.
| Benefit | How to assess it | Common mistake |
|---|---|---|
| Cash savings | Costs that actually stop being paid | Calling all released time a payroll saving |
| Released capacity | Manual hours no longer required at comparable volume | Assuming all capacity becomes billable work |
| Service improvement | Completion, response or turnaround under fixed rules | Changing the denominator or time window |
| Reduced complexity | Fewer handoffs, re-entry steps or training questions | Claiming a financial value without evidence |
Include the whole cost.
Count discovery, implementation, migration, rollout and training. Then include hosting, licenses, AI usage, monitoring, support and the time people spend reviewing exceptions.
A useful financial calculation is net benefit divided by total cost over the same period. If the benefit is mainly capacity, present hours and the assumptions separately rather than disguising them as guaranteed savings.
Use a range and revisit it after rollout.
Estimate conservative, expected and optimistic operating cases. Make volume, correction effort and recurring service costs explicit. A single precise number can hide large uncertainty.
After launch, repeat the baseline measurement. Explain what else changed and where the evidence is incomplete. One observed project result is useful proof of that project; it is not a guaranteed result for another business.
- A consistent completion definition and observation window
- Real manual effort, including exceptions
- Separate capacity and cash effects
- Implementation plus ongoing costs
- A named owner for reviewing the result
