BuyerCompanion: AI Procurement Agent

A global energy group operating in 120+ countries, with more than 30,000 employees on Microsoft 365 Copilot. Across that footprint, mid-value procurement (purchases under EUR 50,000) was the painful tier: too small to justify a dedicated procurement team, too important to skip compliance and supplier vetting. BuyerCompanion is the agent I co-built at Witivio to automate the entire RFP lifecycle for those purchases, delivered as an official partnership between the client, Witivio and Microsoft.

Buyers struggled to navigate internal purchasing rules, identify approved suppliers, and produce the right legal documents. The result was inconsistent data, real delays, and a steady risk of non-compliance with corporate governance.

The brief was unusual: deliver a full self-service experience to non-specialist employees, while enforcing strict policy, supplier vetting, and a complete audit trail. Make it work inside Microsoft Teams (where employees already are). Make it scale from a French pilot to 15,000+ employees globally without re-architecting. And ship it production-ready, not demo-grade.

I designed BuyerCompanion as a Microsoft Copilot Studio agent embedded in Teams, with Power Automate handling backend orchestration and AI Builder extracting structured data from unstructured supplier proposals. The agent guides the buyer through a conversational workflow: define the need, set the budget, scope the work, identify suppliers, generate the SOW and pricing grid, route for approval.

Two design choices made the difference. First, every RFP gets its own dedicated Teams channel that becomes the permanent project record. Documents, decisions, approvals, and supplier exchanges all land in one place, with no manual filing. Second, the SOW and pricing grid are generated as real Word and Excel documents (not chat output), which is what made senior approvers trust the agent for real EUR 50,000 decisions.

Key technical choices:

• Conversational front-end over forms - lower friction, higher completion rate for non-specialist buyers.
• Per-RFP Teams channel as the artifact hub - traceability without anyone having to file anything.
• SOW and pricing grid generated as proper Office documents, not chat output - critical for approver trust.
• Integrated approval flow in Teams - decision-makers approve in one click without leaving their daily surface.
• Connected to the corporate supplier database for vetted recommendations and automated compliance checks.
• AI Builder for structured data extraction from supplier proposals, accelerating the evaluation phase.

• 10% projected cost reduction on affected purchases, quoted directly by the client's Global IT Services Director.
• Deployed in France as the launch market, with the architecture designed to scale to 15,000+ employees globally.
• Audited and validated by Microsoft technical staff, confirming adherence to architectural and security best practices.
• Featured as an official Microsoft Customer Story, alongside the group's industrial maintenance agent, as proof points for its generative AI programme.

Microsoft published BuyerCompanion as part of the client's generative AI customer story, alongside the deployment of 30,000+ Microsoft 365 Copilot licences and an industrial maintenance agent. The client is anonymised on this page; the published story and the press coverage are on the media page.