What was done
Quebec industrial-automation firm Contrôles Laurentide, with 1,000+ customers and 16,000+ purchase orders since 2019, worked with AI consultancy Videns on two agentic AI tools: one sorts and routes the flood of uncategorized customer emails, and one reads unstructured customer POs and checks addresses, parts, quantities, prices and terms against ERP data.
Outcome
In testing, Videns reports the email classifier handled over 50% of incoming messages with 80% accuracy, and that about half of PO checks can now be automated, halving processing time and cutting related costs by 40%. These are test-phase projections reported by the consultancy, not yet full production results.
Evidence: Agency case study. Figures are as reported by the source and have not been independently audited by AgentsNorth.
Prompt Reconstructed by AgentsNorth
The source didn't publish its prompt. This is our reconstruction of a starting prompt for a similar workflow, not the company's actual prompt.
You are an order-desk assistant for an industrial automation distributor. Compare this customer purchase order with the matching quote and ERP record. Check ship-to and bill-to addresses, part numbers, quantities, unit prices, currency, payment terms and requested delivery dates. Output a table of each field with PO value, ERP value and match / mismatch, then a one-line verdict: ready to enter, or needs human review with the reasons. Never change prices or terms yourself.