Use this kit to prepare a bounded pilot, record evidence, and make an expansion decision. Start with a copy for one team and one workflow. Blank fields are deliberate; they do not mean zero or approval.
The accompanying Neuronify pilot companion workbook has five editable sheets: Economics, Lifecycle (setup and continuing cost of sourcing options over a horizon), Inputs (the basis and source of each economics input), Evaluation (case records for all nine workflows), and Evidence. It contains no measured customer data and no preset ROI. Get the workbook (version 0.3, October 2, 2026). This download is available in the local development reader; these pages supply the same core prompts for readers working in another tool.
Define the pilot
| Field | Your entry |
|---|---|
| Workflow and team | To enter |
| Decision owner | To enter |
| Unit of work: the completed outcome, not an intermediate artifact | To enter |
| Business hypothesis and the mechanism that would realize value | To enter |
| Eligible population and exclusions | To enter |
| Start/end dates and follow-up observation window | To enter |
| Baseline process and comparator | To enter |
| Model, prompt, retrieval, and source versions | To enter |
| Evaluation design: cases, comparison and assignment method, reviewers | To enter |
| Operating limits: permissions, capacity, latency, recurring cost | To enter |
| What the system may propose | To enter |
| What requires human approval | To enter |
| Pause conditions and person who can disable it | To enter |
| Quality thresholds agreed before testing | To enter |
| Sample-size rationale and remaining uncertainty | To enter |
Use the evidence checklist before selecting test cases. Use the evaluation scorecard to record outcomes, including failed attempts and rejected drafts. Use the economics worksheet with actual inputs or explicitly labeled assumptions.
Use the evaluation case library to rehearse scoring before collecting an authorized sample of actual work. Its synthetic cases are not a representative traffic benchmark.
Fill the economics sheet
Each column is a different workflow, not a forecast scenario; rename the column headings to match the workflows you are assessing. Complete only the columns you need. Inputs and outputs use a month and USD. Convert proposal hours into minutes before entering them. Change inputs in place to examine a downside and save a dated copy if you need to preserve it.
- Enter volume, eligible share, adoption, baseline and assisted human minutes, hourly capacity value, and recurring capacity costs. Include review and comparable follow-up work in the time inputs.
- Record each input's observation window, basis (observed, estimated, or assumed), source or owner, and how it was measured on the Inputs sheet. Different task mixes should be analyzed separately.
- Enter avoidable cash expense and incremental cash cost independently. Do not copy capacity value into avoidable expense without an evidenced mechanism such as reduced paid overtime.
- Enter one-time cash setup spending if evaluating payback. Payback is unavailable when net cash benefit is zero or negative. A zero setup outlay can legitimately produce zero months.
Released hours = assisted units × (baseline minutes − assisted minutes) ÷ 60. Capacity value after recurring cost = released hours × hourly value − recurring capacity cost. Net cash benefit = avoidable cash expense − incremental cash cost. Simple payback = setup cash ÷ positive monthly net cash benefit.
Recurring capacity cost includes resources not already captured in handling effort. Recurring cash cost includes incremental spending. They may differ. Neither output is a revenue forecast or accounting classification. The workbook does not model ramp-up, tax, financing, or changes in demand.
Prepare the decision memo
| Question | Your evidence and decision |
|---|---|
| Which population and configuration were actually tested? | To enter |
| What quality gates passed, failed, or remain untested? | To enter |
| What changed in total human effort and repeat work? | To enter |
| What incidents and near misses occurred? | To enter |
| Which inputs are still assumptions? | To enter |
| What expense can actually be avoided, by whom and when? | To enter |
| What were the performance rates, each with its numerator and denominator, by cohort? | To enter |
| What scope is approved: expand, revise, simplify (adopt the simpler alternative), defer, or stop? | To enter |
| Who owns unresolved actions and the next review date? | To enter |
A completed sheet does not itself authorize deployment. Keep the decision and its supporting evidence together. Do not record an untested gate as passed because no incident has yet been observed.
Read the worked examples: support, proposals, and order intake.