SupRM Intelligence: AI for Supplier Decisions.
An enterprise AI assistant I led at Sanofi, SupRM Intelligence reached 80 unique production users and 572 conversations in its first six months. It supports decisions across 61 strategic suppliers and 34 Key Supplier Managers. I built its evaluation framework, reusable agent tools, and supplier performance reporting.
80
unique production users in the first six months
572
conversations in the first six months
97.9%
production evaluation pass rate
6+
AI products using shared capabilities
Skills used
chapter 01 —
An assistant for supplier decisions.
SupRM Intelligence supports supplier decision-making across 61 strategic suppliers and 34 Key Supplier Managers. I led its development and designed the supporting evaluation, reporting, and reusable agent capabilities. The work connects AI product delivery with measurable adoption and answer quality.
chapter 02 —
Make the answers useful and accountable.
Supplier performance and governance work spans different reporting cadences and information sources. I focused on making the assistant useful for day-to-day decisions while giving the team a repeatable way to assess answer quality and tool use.
chapter 03 —
Product decisions.
The decisions behind the work and the reasoning for each.
Evaluate business questions systematically.
I created an evaluation framework across 18 business scenarios and 25 deterministic checks. It achieved a 97.9% production evaluation pass rate and 100% tool compliance. Redesigning the responses also cut representative answer length by 65%.
Build capabilities other AI products can reuse.
I architected shared agent tools and orchestration capabilities used by 6+ AI products across budgeting, risk, compliance, procurement, and supplier management.
Build reliable reporting across cadences.
I developed supplier performance reporting across weekly, monthly, quarterly, and annual workflows, reconciled 1,822 historical metric records, and backfilled 211 missing reporting periods.
Keep proposed actions under human review.
I piloted a governed workflow that turns supplier files into action proposals for people to confirm, revise, or dismiss before an action is created.
chapter 04 —
What these results show.
the honest caveats —
- The adoption totals cover the first six months, not lifetime use.
- The 97.9% pass rate measures defined evaluation scenarios, not every possible supplier question.
- This public case study uses aggregate outcomes and does not show supplier records or internal screens.