Client scenario
Northstar Industrial Supply is a fictional 42-person regional distributor. Customer-service staff receive emailed purchase orders, re-key line items into an ERP, resolve product-code mismatches, and send confirmations. The scenario assumes 310 orders per week and a blended administrative cost of $38 per hour.
Executive recommendation
Start with purchase-order intake and exception routing. It has the clearest volume, a bounded document set, an accountable operations owner, and a safe human-review point before an order enters the ERP. Do not begin with autonomous customer-email drafting; the exception language and commercial commitments are not yet standardized enough.
Observed workflow baseline
In a real Nano-Pilot, baseline figures are taken from interviews, workflow observation, exports, and client records. Each number is labelled by source and confidence.
Ranked opportunity shortlist
Order intake and exception routing
Extract line items, match product codes, and send uncertain rows to a reviewer before ERP entry.
Order-status preparation
Assemble shipment and back-order facts for staff to review before sending a customer update.
Demand forecasting
Potentially valuable, but historical SKU and stock-out data needs cleanup before modelling.
Visible value calculation
The illustrative baseline is 310 orders × 11 minutes = 56.8 staff hours per week. If assisted intake removes 55% of handling time after exceptions and review, the scenario returns 31.2 hours per week. At $38 per hour over 48 operating weeks, the directional annual capacity value is approximately $56,900.
This is arithmetic from fictional assumptions, not a savings promise. A real recommendation shows the source, confidence, implementation cost, adoption risk, and sensitivity range for every input.
Recommended pilot scope
- Input: emailed PDF and spreadsheet purchase orders from five representative customers.
- Processing: extract order fields, normalize known product codes, and assign confidence to each row.
- Human boundary: staff approve every exception and the complete order before ERP entry.
- Output: a review queue plus an ERP-ready import file; no autonomous submission in the pilot.
- Measure: handling time, extraction accuracy, exception rate, correction rate, and staff adoption.
Data and implementation boundaries
- Use sample orders with customer identifiers removed until the data flow is approved.
- Document where files are processed, retained, and deleted before production access.
- Do not automate pricing exceptions, substitutions, credit decisions, or customer commitments.
- Stop the pilot if extraction accuracy or reviewer confidence does not meet the agreed threshold.
Decision at the end of the Nano-Pilot
Proceed only if the baseline can be verified, representative documents are available, an operations owner accepts the review responsibility, and the conservative value range supports the written implementation estimate. Otherwise, clean the data or improve the current process before building.