Case study · Logistics & industrial
POD / invoice exceptions the Tuesday shift will run
Industrial distributor / logistics ops · North Houston / Port-adjacent freight mix
Primary metric: exception cycle time on invoice–POD mismatch queue · Results under NDA
- Industry
- Logistics & industrial
- Horizon
- 90-day pilot
- Systems
- WMS, ERP AP, email, scanned POD/BOL
- Primary metric
- Exception cycle time · NDA
Situation
The WMS was live. The dock was moving. AP and customer ops still lived in a spreadsheet-fed exception pile: missing PODs, partial deliveries, rate mismatches, and email chains nobody wanted to own.
Constraint
High volume of imperfect scans. Multiple customer formats. Small ops IT bench. Floor staff had zero patience for a tool that added clicks. Needed something usable on Monday—literally.
What we shipped
- Exception taxonomy and ownership map (who touches what, by when)
- POD/invoice document intelligence for match/flag
- Coordinator workqueue with suggested next action
- Thin ERP/WMS status checks where API reality allowed
- On-site training for the exception desk + written runbook
Controls / human-in-the-loop
Payment holds and customer-facing dispute steps remained human-approved. Low-confidence matches escalated automatically. Full action history retained for AP audit.
Results
Under NDA until client-approved. Primary metric: exception cycle time on the invoice–POD mismatch queue, baselined and tracked weekly. Secondary: manual touches per resolved exception. No fabricated public day counts.
What we refused to do
- Full “autonomous AP agent” with payment authority
- Custom vision science project for edge-case scans before happy-path volume was handled
- Multi-warehouse rollout before one desk proved the metric
Start with one metric—not a 90-slide roadmap
Tell us where cost, throughput, or labor is stuck. We’ll tell you if a 2–4 week assessment is the right next step—or if it isn’t.