AI adoption states · indirect procurement
Assisted, Augmented, Autonomous: not a maturity ladder, but three distinct ways an indirect procurement team can use AI. Each has its own expected outcomes, its own operating model, and its own place where the buyer creates the most value. Further down, you will also see the four tiers that measure how widely those states are deployed across your function.
§ 1 · The three States
For each State, two questions: what does it produce (expected outcomes in € and time freed) and what does it change about the job (where the buyer spends time, where they create value no one else can create in their place).
State 1 / 3
Human decides · AI assists
Expected outcomes
Operating model
The buyer spends time on:
Deciding on better-prepared cases. AI clears the reading and drafting load. The buyer reclaims judgement time: picking a supplier, calling a clause, leading a negotiation. No change to procurement processes.
State 2 / 3 · 2026 target
Human validates · AI operates
Expected outcomes
Operating model
The buyer spends time on:
Calling decisions earlier and higher. Skills have already done the triage: categorisation, risk signal, comparison. The buyer no longer starts with a blank page, they start with a file. Reclaimed time goes into the strategic supplier relationship, the sensitive negotiation, the category work.
State 3 / 3 · horizon
AI operates · human supervises
Expected outcomes
Operating model
The buyer spends time on:
Governing a portfolio. The role shifts: designing rules of engagement, reviewing exceptions, piloting a portfolio of agents. Operational execution (recurring RFx, catalogue renewals) no longer consumes their time. Their value-add moves toward the strategic, the sensitive, and governance.
Fast read
What the three States share.
None of the three degrades the buyer's role, each redeploys it. Assisted frees judgement time, Augmented shifts the decision earlier in the flow, Autonomous opens a governance role. The point is not to replace the human, it is to move where the human creates value.
State 1 / 3 · the detail
Human decides · AI assists
A buyer opens ChatGPT, Claude or Copilot alongside their working tool. They use it like an assistant: draft a supplier email, summarise an 80-page contract, compare three offers in a table. The procurement process itself does not change, it accelerates.
Assisted is the first threshold for a team: it requires no integration, no data foundation, no enterprise licence. It is an individual licence plus a one-page usage policy. It is also the threshold where most teams stop, for lack of the foundations to go further.
Category · Facility Services
The buyer loads the supplier response (12 PDFs) into Claude, asks for a tariff / SLA / spec-gap comparison. The raw table becomes the base for manual analysis.
→ 4 h saved · lower risk of oversights
Category · IT / SaaS
The buyer dictates the 3 points to rephrase, ChatGPT proposes 3 tones (formal / direct / collaborative). They pick, tweak, send.
→ 20 min → 4 min
Category · Travel
The buyer submits the new travel partner's MSA to Claude, asks for the 5 risk clauses for indirect procurement. Legal validation stays on Legal's desk.
→ pre-read 80 % shorter
State 2 / 3 · the detail
Human validates · AI operates
AI is no longer alongside, it is in the flow. A Skill connected to your P2P reads spend, proposes recategorisation, surfaces three pricing levers, prepares the supplier QBR agenda. The buyer calls it, validates, signs. The job stays theirs, they operate better tooled.
Augmented requires a minimum data foundation: clean spend, a stable category taxonomy, P2P connected. From there, we build versioned and observed Skills, pieces of the sourcing / category management chain where AI does the first mile, and the buyer calls it earlier, with more angle. This is the tier where competitive edge widens.
Category · Indirect / spend analytics
The Skill reassigns the 12 % miscategorised spend, proposes a merger of two sub-categories, identifies 14 suppliers tagged « General services » actually billing « Building maintenance ».
→ −3 weeks of manual cleanup · base ready for sourcing
Category · Telecom / Connectivity
The agent aggregates real SLAs, invoices, contract deviations, proposes 6 quarterly review questions ranked by € stake. The buyer walks into the meeting with a signed file in 10 min.
→ denser QBR · recurring savings captured
Category · MRO
The Skill aggregates quality incidents, late deliveries, price gaps vs benchmark, proposes a composite supplier score refreshed weekly. The buyer triggers / de-lists a supplier in light of the score.
→ referencing decisions earlier · risks anticipated
State 3 / 3 · the detail
AI operates · human supervises by exception
The agent receives the purchase request, sources, negotiates on a contractually scoped perimeter, hands off for signature. The human no longer validates every decision, they review deviations, adjust the rules of engagement, step in when scope leaves the rails. A few perimeters at first (MRO catalogue, renewals < €50k). Not everything. Not all at once.
Autonomous is not « AI replaces the buyer ». It is the buyer supervises a portfolio of agents that operate on scoped perimeters. The role shifts: less execution, more governance. More rules-of-engagement design, more deviation reviews, more portfolio piloting.
Category · MRO < €5k
The agent renews the 240 catalogue lines, negotiates the price increases, flags 11 lines where the gap exceeds the 4 % allowed. The buyer reviews those 11 lines in 30 min, signs.
→ 240 h saved / year · renewal cycle −65 %
Category · General services
Request filed by an operational requester: the agent qualifies the need, sources 3 referenced suppliers, requests quotes, proposes the optimised choice. The buyer validates exceptions outside the envelope.
→ time-to-order −70 % · buyer freed from < €50k tenders
Category · Energy / Utilities
The agent monitors energy price curves, triggers automatic re-sourcing when the contract threshold is crossed, prepares the renegotiation file.
→ opportunities captured outside the traditional cycle
§ 5 · Where your team stands
A State (Assisted, Augmented, Autonomous) describes the nature of one AI-touched process. A tier (Early, Developing, Advanced, Leading) describes how many procurement processes touch AI across the whole function. A team can be Early with its sole buyer in Assisted, or Leading with 8 processes in Augmented and 2 in Autonomous. The grid below reads each of the 12 crossings.
0–2
A curious buyer trying ChatGPT on an RFP.
0
No data foundation to plug a Skill into.
0
Out of zone at this tier.
3–6
Several buyers use ChatGPT / Claude / Copilot on recurring tasks.
0–1
A first Skill (spend analysis) connected to the P2P.
0
No supervised-autonomy agent yet.
5+
AI-assisted is now the default reflex on drafting and analysis.
3–5
Skills / agents in the flow: spend, sourcing, supplier scoring.
0–1
First agent in supervised autonomy on a narrow perimeter.
7+
Standard. Every buyer prompts. Stabilised usage policy.
6+
Majority of decisions operated by AI, validated by the buyer.
2–4
Agents on catalogue, renewals, RFx < €50k.
Example · profile highlighted above
Indirect procurement team · mid-cap group · score 17 / 25. Advanced tier. Of the 9 procurement processes the team runs, 6 are in Assisted (RFx drafting, contract summaries), 4 are in Augmented (spend analysis, supplier scoring, QBR prep), and 1 is in supervised autonomy (MRO catalogue renewal < €5k). Diagnostic results read: « You are Advanced, mostly Augmented, with one Autonomous. »
≤ 10 / 25 · 0–2 AI processes
No enterprise licence, no usage policy. One or two curious buyers have tried ChatGPT on an email. The topic is not sponsored by procurement leadership.
Typical mix: 1–2 Assisted · 0 Augmented · 0 Autonomous
11–15 / 25 · 3–7 AI processes
The team has a usage policy, licences, several buyers prompting daily. A first Skill starts running (spend analysis, sometimes pilot).
Typical mix: 3–6 Assisted · 0–1 Augmented · 0 Autonomous
16–20 / 25 · 8–11 AI processes
Industrialised Skills (3 to 5), data foundation in place, first supervised autonomy starting on a narrow perimeter (catalogue or RFx < €50k).
Typical mix: 5+ Assisted · 3–5 Augmented · 0–1 Autonomous
> 20 / 25 · 12+ AI processes
Several agents in production on scoped perimeters. The buyer operates in governance, designing rules, reviewing deviations, piloting the portfolio. The role has shifted.
Typical mix: 7+ Assisted · 6+ Augmented · 2–4 Autonomous
§ 6 · Move to the next tier
Four recurring levers. Not in the « right » order, but what the teams we accompany actually cross in practice. A workshop tests one of them on one of your processes. The audit prices all four across your full scope.
Early → Developing
Before building the data foundation, move the whole team onto Assisted tools: usage policy, 1-day training, usage measurement. Most teams skip this step and regret it.
Developing → Advanced
Clean spend, validated category taxonomy, P2P plugged into an AI environment. This is what separates « a buyer prompts » from « a Skill operates in the flow ». The technical work is half data, half category-director alignment.
Advanced → Advanced+
Versioning, observability, data owner + procurement owner on each Skill, monthly review. This step separates teams that « have 2 Skills running » from those that « pilot 5+ Skills ». Not glamorous. Very differentiating.
Advanced → Leading
Define the exact perimeter where an agent can decide alone, the deviation thresholds that trigger a human, the review loops. It is a contractual exercise as much as an AI one. Start on a narrow perimeter (MRO < €5k). Widen later.
Three entry points, three efforts, the same thread: price where you are, name the next step, build with your team. The 5-min diagnostic places you on the grid. The three offers below take it from there.
Workshop · €950
We take a specific process from your daily run (IT sourcing, supplier negotiation, contract review). We see together how AI can improve it, Assisted or Augmented depending on your tier.
→ credited against the Audit
AI Procurement Audit · €9,999 · 3 to 4 weeks
Full diagnostic on 5 to 8 processes, priced opportunities, signed 12-month roadmap. Co-built with your sponsor (CPO or equivalent). We leave you autonomous.
→ credited against the Programme
Programme · 3 to 6 months · on quote
We build 3 to 5 Skills with your team, we frame the first Autonomous-agent rule. On exit: your team owns the steering. The programme includes the audit and the workshop.
−€10,949 if full cascade
§ 8 · FAQ
No. The tier is a global score on 25 (5 axes). An Advanced tier describes how many processes touch AI, not what each process does. Most Advanced teams have a majority in Augmented, plus a tail of Assisted, plus sometimes one Autonomous in pilot. That is exactly what the § 5 grid renders visible.
Technically: yes. Pragmatically: rarely. Assisted installs the AI reflex in the team, without which an Augmented Skill is underused even when it exists. Teams that skip Assisted end up coming back for training 6 months later. Recommended: 1 month of standardised Assisted before building the data foundation.
3 to 4 months for the first Skill in production, 12 months to have 3 to 5 industrialised Skills. The barrier is rarely tech, it is category-director alignment on the taxonomy and the cleanliness of the spend. That is what the audit prices in 3 to 4 weeks.
For Assisted: the CPO is enough. For Augmented: CPO + CIO (mandatory pair, because the data foundation runs through both). For Autonomous: CPO + CIO + Legal (framing the rules of engagement). We do not run the audit without a named procurement sponsor, that is the one non-negotiable.
Copilot alone places you in Assisted. Moving to Augmented requires connecting an AI environment (Claude, Anthropic API or equivalent) to your P2P and your clean spend. Copilot stays useful on the office layer, but does not do the work of a Procurement-domain Skill. Take the 5-min diagnostic to position your team precisely.
/ai-readiness · 5 min, no email
Take the 5-min diagnostic