A dedicated page for what ten scattered slides would not say well: what "augmented" means in indirect procurement, where you stand on the maturity scale, and exactly which guarantees you get when AI touches your data.
The thesis
I use this sentence as an analytic tool I apply on every AI use case presented to me. It legitimises AI when it shifts the transactional load and frees human judgment. It excludes AI when it just speeds up the wrong move.
"Augmented" means: tooled when it is needed, method when the tool is not enough, human judgment when the method reaches its limits.
Definitional thesis · The Augmented Indirect
AI maturity framework
Three levels of maturity. Most teams plateau at "Assisted". The real shift happens at "Autonomous", where AI runs the process end-to-end and procurement scales without adding headcount.
Maturity is not bought, it is earned. Skip a level and the next one collapses. Most teams stay stuck at the bottom for lack of data foundations, team literacy, or governance.
What most teams use today without yet calling it AI. Four use cases that already pay off, no operational risk.
Conditions for the climb
Maturity is not bought, it is earned on three foundations. Skip one and the whole stack stalls.
AI maturity rests first on the integrity of your data layer: a reliable supplier master, clean ERP extracts, a coherent category taxonomy. Without it, every model output inherits the entropy of your back-office, and AI does not accelerate, it scales the noise.
AI is only worth what your buyers do with it. Reading the outputs, challenging biases, probing use cases: that is the work. A team that sees AI as a threat will deploy nothing meaningful. The one that internalises it changes the cadence of the function.
Explicitly decide what is delegated to AI, log every decision, define the escalation paths. Without that frame, an autonomous deployment becomes a risk you cannot defend in front of the executive committee, and the experiment quietly dies before it scales.
The stack in place · what actually happens with your data
When I work with my stack, here is exactly the contract. Not one option among others: this is the standard guarantee, signed with each provider, documented, auditable. No engagement starts without it being laid out clearly to you, your CISO, and your DPO.
When I use my stack, your data goes via API (never through a consumer interface) on the frontier models listed below. This gives you access to the best models available, while keeping the protection level your internal contracts require.
Your data is never used to train the models. Contractual via API, not a marketing promise: verifiable in each provider's terms of use.
Inference on EU datacentres (typically Frankfurt or Paris depending on the provider). No transfer outside the EU. Compliant with French and German mid-market data-localisation requirements.
Data Processing Agreements signed with each provider. GDPR compliance documented, board-ready, verifiable by external audit at any time.
Full documentation provided at engagement start (DPAs, certifications, data flow diagram). Board validation included in the scope, no extra cost.
Request the documentation →Adaptations to your context
If your IT department has already deployed an enterprise AI environment (Microsoft Copilot, Gemini Enterprise, ChatGPT Enterprise, Mistral, internal deployment), I work with that one.
No duplication, no exfiltration to an external stack. Your internal policies stay the reference.
For sectors where data cannot leave the internal perimeter: switch to open source models running locally. No external call to a frontier LLM.
Typical stack: Mistral / Llama / Qwen on on-premise hosting or sovereign cloud (OVH, Scaleway, Outscale).
Experiments in progress · what passed the operational filter
Dedicated tool for managing the steel category: reference prices, index tracking, hedging scenarios, contract alerts.
See the tool →5 courses, 37 modules, 105+ lessons, 184 quizzes. Each module reviewed by a procurement operator. Never published raw.
formation.theprocurementor.com ↗Automated mapping, classification, anomaly detection. Two days instead of two months. Human judgment on interpretation.
Audit methodology →No pitch. No commitment. You walk away with: the target maturity level, the stack variant fitting your CISO, and the GDPR documentation to present to your DPO.