AI Email Intelligence: From a Distributor’s Inbox to Tracked Obligations
Why Standard Cloud Stacks and Legacy Tools Failed
A Moroccan medical-supply distributor runs much of its business by email: customer orders, supplier price offers, payment reminders and regulatory notices arrive across 22 mailboxes, mixed in with internal threads. A commitment buried in a thread (a payment due, a promised delivery, a regulatory deadline) was only followed up if someone remembered it.
Custom Low-Latency Architecture & In-Country Execution
Vext built a pipeline that pulls every mailbox into one store and has an LLM read each email to record what it is about (31 intent types), who it involves, any amounts and deadlines, and how urgent it is. Emails that create a commitment become obligations with a due date and a status, so the team can see what is closed, what is coming up and what is overdue.
Key Engineering Deliverables
- Ingestion of 22 mailboxes with history back to February 2017
- LLM extraction of intent, counterparty, amounts, deadlines and urgency for every email
- Obligation tracker for payments due, deliveries and regulatory deadlines
- Overdue flags and an upcoming-deadlines view for the team
Measured Production Impact
- 6,441 obligations closed and 208 upcoming, all tracked in one place
- 763 overdue obligations flagged instead of staying buried in email threads
- 10,208 facts carry amounts and 7,112 carry deadlines, searchable by counterparty and intent
Architectural Context & Deployment Narrative
For a Moroccan medical-supply distributor, the inbox is where the business actually happens. Vext ingested 56,457 emails from 22 mailboxes, with history back to February 2017, and had an LLM convert 50,560 of them into structured facts. The most common intents are orders (8,715), payments (7,706), internal coordination (4,375), price offers (3,746), market watch (2,685) and regulatory matters (2,531). From those facts the system extracted 7,939 obligations: 6,441 are closed, 208 are upcoming and 763 are flagged overdue, so nothing promised by email depends on someone’s memory.
Have a similar engineering challenge?
Speak directly with the lead systems architect behind this deployment.
More Production Case Studies
Vext Voice: Morocco’s Sovereign Darija & French AI Voice Telephony Engine
Casablanca bare-metal telephony & WhatsApp VoIP gateway executing Moroccan Darija and French conversational agents with sub-400ms voice latency, direct IAM/Orange/Inwi SIP trunks, mid-call GPS & CMI multimodal actions, and 0.34 MAD/min runtime economics under 100% CNDP Law 09-08 compliance.
Finance Operations Hub: Sage Journals, Receivables and Cheques in One View
Sales and purchase journals imported from Sage (15,348 invoices across 375 active clients and 3,093 SKUs since January 2024), with 5,717 receivables and 1,943 cheques tracked from receipt through deposit to clearing.