Plan a baseline, controlled pilot and review of automation value. Separate capacity from cash savings and calculate ROI with explicit costs and assumptions.
Use this illustrative 90-day sequence to plan measurement before automation goes live. Agree a baseline, comparison method, costs and stop conditions with operations and finance. Extend the observation period when volumes or risk require it; the schedule does not guarantee a statistically conclusive result. A useful report separates measured facts, assumptions and forecasts. Record the original task, how much work was eligible for automation and the review or rework still required. A comparison group can help identify other causes of change, but its design must fit the workflow and protect customers from avoidable harm.
Why an ROI claim can be misleading
Common problems include a missing baseline, changing task mix, unrecorded human review and double-counting recovered hours as both capacity and cash savings. Record other process or staffing changes during the pilot. Show estimated inputs as estimates and test how the conclusion changes when they vary.
Days 1–14: Define the baseline
Define a unit of work and separate hands-on minutes from elapsed waiting time. Measure volume, review, rework, errors and service outcomes over a representative period. Record workload cost per completed unit using costs from the same period. Include source and collection method so the baseline can be checked later.
Days 15–28: Plan the comparison
Choose a comparison design suited to the work. Where safe and practical, assign comparable cases to assisted and existing processes; account-level effects may require grouping rather than individual randomisation. Agree success and stop criteria before looking at results. Record versions, case mix and the data needed for analysis. Get suitable analytical advice for high-stakes or complex comparisons.
Days 29–56: Observe the pilot
Run long enough to cover a representative range of tasks and gather useful evidence; four weeks is an example window, not a universal minimum. Capture failed runs, escalations and reviewer effort. Log changes to the model, prompts, tools or process. Fix safety issues promptly, then distinguish results before and after the change rather than treating the run as unchanged.
Days 57–70: Analyse value and cost
Compare task mix, completion, hands-on time, errors and service outcomes. Report sample sizes, exclusions and uncertainty; do not assume a calendar duration establishes statistical significance. For a defined period, ROI equals (attributable benefit minus total cost) divided by total cost, multiplied by 100. Include implementation, usage, review and support. Report realised cash savings, recovered capacity and forecasts separately. Use an appropriate analysis method rather than attaching confidence intervals to an unsupported extrapolation.
Days 71–84: Decide whether to expand
Expand only if the agreed quality, service and cost checks are met and an operating owner accepts the remaining limits. Increase use in stages suited to the workflow, watching case mix, failure rates and exception capacity. Keep a pause control and manual fallback. Extending or stopping the trial is a valid outcome when evidence is weak.
Days 85–90: Report the decision
Present the baseline, comparison design, results, limitations, complete costs and recommended next step. Show observed results separately from annual forecasts. For payback, divide one-off cash investment by positive recurring net cash benefit, stating when that benefit starts. Include rollout delay and additional setup costs; if net cash benefit is not positive, there is no simple payback period to report.
Owners and review cadence
Name an operations owner for workflow quality, a finance reviewer for benefits and costs, and a sponsor for the expansion decision. Agree how often they review evidence and what requires immediate escalation. Keep the assumptions and decision log accessible so a change in ownership does not erase the measurement method.
Mistakes to avoid
Do not cherry-pick easy cases, omit failed runs or treat every saved minute as a cash reduction. Do not count the same avoided work twice. Include recurring subscriptions and human review. Record product changes and other operational changes. Explain analytical limitations rather than claiming that a small pilot proves a result for all future work.
What a useful outcome looks like
A useful measurement exercise supports a clear choice: continue, change scope, gather more evidence or stop. Keep the baseline and method reusable, but recheck the economics and task mix for each new workflow. Neither a 90-day schedule nor a positive estimate guarantees savings, audit acceptance or future performance.