Case study — Import & Distribution
A 7-agent AI ecosystem on Business Central for a $40M+ import & distribution company
Khaab AI built a seven-agent ecosystem on top of a $40M+ distributor’s Microsoft Dynamics 365 Business Central ERP. Sales reps request quotes in a Slack DM and the quotes agent drafts them; a person always approves. Agents share a governed memory layer, run inside the company’s own Azure environment and answer to an executive voice and ops assistant.
- purpose-built agents on one ERP
- 7
- operation running on the system
- $40M+
- of money-touching actions human-approved
- 100%
The problem
A $40M+ import & distribution business ran on a lean team stretched across purchasing, sales, compliance and accounting. Commodity prices were tracked by calling origin countries every month, certificates lived in inboxes, and every report interrupted management.
Quoting was the sharpest bottleneck: reps waited on management for prices, numbers were rebuilt by hand in spreadsheets, and a wrong figure could go out the door with nobody noticing. A previous vendor’s tools had made staff slower, not faster.
The stack was modern — ERP, voice, accounting, messaging — but nothing was connected. What was missing was the connective tissue.
What we built
Seven purpose-built agents on live Microsoft Dynamics 365 Business Central data: purchasing watches inventory, FX and commodity prices and recommends buy, hold or reorder; sales follows the playbook of the company’s top rep; quotes drafts customer pricing; data cleanup enforces clean records and tracks every certificate expiration; knowledge answers staff questions from live data; procurement and accounting run the back office.
The quotes agent lives where reps already work. A rep sends a Slack DM, the agent prices the request with a deterministic engine — the language model never invents a number — and returns a draft. The model can never save or approve a quote on its own: a person submits it, the price is re-checked, and approval happens only on a signed approval card.
A memory layer lets agents remember account and rep context — proposed by the agent, confirmed by the rep, never storing prices or contacts as “facts.” Everything runs inside the company’s own Azure tenant on Postgres synced with the ERP, with role-scoped access and an append-only audit trail. An executive voice and ops assistant sits in front of it all for briefings and questions.
Every action that touches money or leaves the building is gated behind human approval. Structured report in, one decision out.
Deliverables
The result
Reps get draft quotes in the chat tool they already use, management approves instead of rebuilding numbers, market coverage is systematic across every origin country, and the business gets agents that scale sales and back-office capacity without adding headcount — with a person in control of every decision.
