The problem
Ajusa’s small sales team was handling a long-tail B2B flow by phone: small retailers calling in to resolve doubts, find part references, and place low-ticket but critical orders. Operating cost was high, coverage was limited to business hours, and the reputational risk of slow response hung over every shift.
They needed to scale the channel without scaling the headcount — but the stakes ruled out any “demo-grade” chatbot. These are real orders worth thousands of euros, created against SAP.
What we built
A fully custom assistant deployed inside their own Azure subnet, connected securely to SAP and their catalog APIs. Multi-channel delivery through WhatsApp (and voice/phone where the flow calls for it) covers part lookups, doubt resolution, and order management.
Two things make it different from the usual chatbot:
- Built from scratch on Azure — no closed chatbot library lock-in. We leaned on Logic Apps for orchestration, Application Insights for end-to-end traceability, centralised logging, monitoring and operational alerting.
- It is not an FAQ bot — it creates and modifies real orders worth thousands of euros. So security and reliability were primary requirements, not an afterthought: role-based auth, validation against source systems (SAP / catalog), hallucination guardrails, explicit confirmation before any critical action, scope limits per intent, full auditing, and human escalation at any sign of ambiguity.
Why it matters
This is the shape of an AI project that actually matters in a traditional industrial business: deployed inside the client’s cloud boundary, talking to the source-of-truth ERP, treated with the same seriousness as any other system of record — not a toy widget on a marketing site.