Prepare the payment review note before making the decision
Bring payment signals into a review note. Compare the rule’s workload and missed fraud.
Recorded example · Read the desktop transcript · Read the mobile transcript
Luca’s payment is larger than his typical payment. The record also shows a new device and several recent payments. This example prepares a review note with four signals before a person records the decision.
Fictional records. Actions stay in the simulation.
Open this demo pageTry the example
Change the records, prepare the work, and review it before saving. This uses the same local simulation as the demo page.
Actions stay in this demo.
Open full demoWhat Does the Review Note Explain?
Select Prepare the review note. The first result uses the payment amount, configured signals, score, and review threshold. Examine the payment and account activity beside the editable note.
The score combines displayed rules. It is not a fraud probability. Each signal contributes to the review request, but those signals do not prove fraud.
Record an investigation or release decision after examining the note. The receipt keeps the decision and review note together. No payment is blocked or transferred.
Open the optional threshold comparison. Test thresholds 40 and 65 against the same 16 sample payments. Threshold 65 sends six payments to review instead of nine. It flags no legitimate payments instead of two, but misses one additional known fraud case.
Fourteen outcomes are known; two are unknown. Unknown outcomes count toward workload only. The comparison keeps the reduction in review work beside the missed fraud. Both results matter when a person chooses a rule.
What Potential ROI Could This Produce?
At 60% adoption, this example releases 18 hours per month. Assisted time includes human review.
Labor proxy: other financial clerks. BLS May 2025 US wages are $27.14/hour nationally.
The cost factor is 1.43, from US private-industry compensation, June 2026.
| Input | Sample assumption |
|---|---|
| Monthly tasks | 600 |
| Minutes, manual → assisted | 10 → 7 |
| Build budget | $5,000 |
| Monthly operations | $220 |
| Potential value | US example |
|---|---|
| Hourly wages + benefits | $38.81 |
| Monthly capacity value | $699 |
| Monthly net capacity value | $479 |
| First-year capacity ROI | 10% |
| Payback, months | 10.4 |
First-year cost: $7,640. ROI divides annual capacity value, less first-year cost, by first-year cost. Payback divides build cost by monthly net capacity value. Slower adoption delays payback.
Released capacity is not automatic cash savings. Workload and time reductions are illustrative, not BLS findings. Internal support is assumed; The AI Operations plan costs extra. The budget and operating allowance are illustrative, not a Ship AI Studio quote. Measure costs before deciding.
What Needs Human Review?
The reviewer examines the facts and records the decision. A real build needs verified outcome labels, permitted payment context, and an accountable rule owner. Test proposed rules on historical data before enforcement.
Measure decision time, legitimate payments sent to review, and confirmed fraud missed. Measure actual loss separately. This estimate covers reviewer capacity only and assigns no value to prevented fraud.
The production guide explains approval boundaries and exception ownership.