Open operational workbook

AI pilot business-case and acceptance workbook

For: Operational sponsors, finance reviewers and pilot owners

Decide whether one bounded workflow is worth testing, and what evidence would justify stopping or progressing. Time saved is not automatically cash saved.

All worked examples and target thresholds here are synthetic pilot-design assumptions, not customer incidents, measured Akima results or promised savings. Agree targets with your reviewers before testing.

Gather a defensible baseline

Choose one eligible case class and a fixed monthly period. Bring four weeks of case volumes, active handling times (not elapsed queue time), rework, loaded labour cost, quality outcomes and source-system ownership. Separate incident severity or order complexity. Do not multiply a simple-case result across the whole operation.

Finance must identify how released capacity could actually be used: avoided overtime, reduced contractor spend, or additional work with evidenced demand. Redeployment without an agreed use may have no realisable monetary value. Record the assumption owner, source, observation period and confidence for every input.

Editable monthly economics

Hypothetical defaults. Values remain in this browser session; nothing is submitted or saved. Print the page to retain your assumptions. All amounts are A$, excluding tax.

Eligible cases per monthBaseline handling minutes per caseAssisted handling minutes per case (excluding review)Additional review minutes per caseLoaded labour cost (A$/hour)Realisable capacity (%)Inference cost (A$/case)Monthly operating cost (A$)Monthly integration maintenance (A$)One-time setup and integration investment (A$)

Gross capacity value / month: $15,000.00
Realisable capacity value / month: $7,500.00
Additional review cost / month: $3,000.00
Total recurring cost / month: $6,000.00
Net capacity value / month: $1,500.00
First-year net value after setup: -$9,000.00
First-year value ROI (not cash ROI): -9.1%
Simple capacity-value payback: 18.0 months

How the model works

Gross hours/month = eligible cases × (baseline handling minutes − assisted handling minutes) ÷ 60. Assisted handling excludes additional reviewer effort.
Gross capacity value = gross hours × loaded hourly labour cost. Positive capacity is multiplied by the realisable fraction. If handling becomes slower, the extra labour cost is charged in full, even when realisable capacity is 0%.
Additional review cost = cases × review minutes ÷ 60 × loaded hourly cost. Recurring cost adds review, per-case inference, operations and integration maintenance. Review is deducted once, not also subtracted from assisted time.
Monthly net capacity value = realisable capacity value − recurring costs. First-year net = 12 × monthly net − setup investment. First-year value ROI = first-year net ÷ (setup + 12 × recurring costs). A zero denominator is unavailable, not infinite ROI.
Simple payback = setup ÷ positive monthly net; otherwise payback is not reached. Zero setup with positive net has zero-month payback. This model assumes steady state from month one, no discounting and no ramp-up; use a monthly cash-flow model before investment approval.

Review labour is an opportunity cost even if no additional salary is paid. Separate cash and non-cash values in the finance review. Include security review, connector development, evaluation and training in setup; include monitoring, support and quality audits in operations. Add error remediation to those inputs if material. Do not also book the same saved hours as headcount reduction and new-revenue benefit.

Three hypothetical cases, reproduced

All scenarios use 1,000 eligible cases/month, 30 baseline minutes, A$90/hour, A$0.50 inference/case, A$1,500/month operations, A$1,000/month integration maintenance and A$27,000 setup. These are planning assumptions, not quotations or measured outcomes. Change them in the calculator; the fixed table below remains a reference.

Synthetic scenario assumptions and outputs (A$)
Metric	Conservative	Base	Upside
Assisted / review minutes	24 / 3	20 / 2	16 / 1
Realisable capacity fraction	25%	50%	75%
Gross capacity value / month	9,000	15,000	21,000
Realisable capacity value / month	2,250	7,500	15,750
Additional review cost / month	4,500	3,000	1,500
Total recurring cost / month	7,500	6,000	4,500
Net capacity value / month	-5,250	1,500	11,250
First-year net after setup	-90,000	-9,000	108,000
First-year value ROI	-76.9%	-9.1%	133.3%
Simple payback	Not reached	18 months	2.4 months

The base case has positive monthly value but a negative first year after setup. Its 18-month payback is only a steady-state capacity-value estimate, not a cash recovery forecast. The conservative case should not pass an economic gate without changing scope or assumptions. Do not select the upside merely because it clears the hurdle.

Acceptance contract before the pilot starts

Fill in an owner and decision date for each gate
Gate / owner	Evidence needed	Proposed acceptance or stop rule
Scope / operations sponsor	One case class; eligible volume; excluded cases; manual fallback; no production write capability.	No scope expansion during evaluation. Unauthorised actions stop the pilot immediately.
Data / data owner	Permission-tested sources, retention decision, versioned baseline and held-out cohort.	Stop on data exposure, inaccessible citations or resolution-data leakage into historical inputs.
Quality / domain reviewer	At least 40 held-out cases; manual comparison; claim-level citation and correction log.	Zero material unsafe recommendations; at least 95% material claims supported. Any unsafe recommendation pauses testing.
Effort / operations lead	Median/p90 handling including review and failed cases; complexity mix and exclusions.	Target at least 20% lower median without worse p90 or quality. Report uncertainty before extrapolation.
Economics / finance	Observed eligible volume and time, agreed capacity-use plan, costs and cash/non-cash split.	Require positive base monthly net and sponsor-approved payback horizon (example: 12 months). This default base case fails that horizon.
Decision / sponsor and security owner	Signed review of gates, outstanding risks and rollback/retention plan.	Proceed only if all mandatory gates pass. Otherwise stop, narrow scope, or re-test with a newly held-out cohort.

Set a four-week evaluation timebox and a maximum approved spend before beginning. Do not extend a failing pilot indefinitely to find favourable examples. Record failures, zero-benefit cases and missing data. If the sample is too small or the comparison is biased, the outcome is “insufficient evidence”, not success.

Take the workbook to your reviewers

Download scenario inputs and acceptance fields (CSV). Edit locally in a spreadsheet, then transfer approved assumptions to the calculator and print the result. The CSV contains values and formula guidance, not an executable financial model. Do not enter confidential operational records into a public form.

Next step: name an operations and finance owner, baseline one workflow, and bring the completed assumptions to a baseline workshop discussion. For a worked operational boundary, use the read-only NOC pilot or order fallout workbook.

Sources and limits

NIST AI Risk Management Framework 1.0 informs the use of documented measurement, accountable review and risk decisions. It does not supply these economics, sample sizes or thresholds. The calculator is a planning aid, not financial advice or a forecast of Akima customer returns.