AI Procurement Playbook/Part 3
Framework
A five-level framework for where procurement teams actually stand in their AI adoption — from individuals quietly experimenting on their own, to a true hybrid organization where agents carry named responsibilities alongside people.
Score your teamWhat we see across procurement teams
The overwhelming majority of the procurement teams we talk to sit at Level 2: they've issued Copilot or ChatGPT licenses, a few people ask it good questions, and that's where it stops. No workflow has actually changed.
The five levels
Score your own team honestly against each level — most industrial procurement organizations today land at Level 2. Click a level to explore it.
02 / 05
Most teams are here
Where most teams are today: tools are licensed, usage is shallow, and the workflow hasn't changed.
The organization has rolled out Copilot or ChatGPT Enterprise, maybe run a lunch-and-learn. People use it for one-off questions — a faster search engine or word processor — but nobody reuses a prompt, no process has been redesigned, and impact isn't measured.
Side by side
What changes in people, process, technology, governance and structure at each level.
| Dimension | |||||
|---|---|---|---|---|---|
| People & skills | No formal training; use is individual, quiet, and undocumented. | Individuals ask one-off questions; no shared prompting practice. | Role-based prompt libraries; teams reuse and refine what works. | Teams learn to manage and review agent output, not just write prompts. | Humans focus on judgment calls — negotiation, relationships, strategy. |
| Process | Workflows are unchanged — AI touches nothing official. | AI sits outside the workflow — opened separately, not embedded in it. | AI handles defined steps — RFQ drafts, scorecards, summaries — inside existing workflows. | Multi-step processes run with human checkpoints instead of human-in-every-step. | Workflows are redesigned around agents, not retrofitted onto old ones. |
| Technology | No licensed or sanctioned tool; personal accounts only. | General-purpose copilot or chat tool, no procurement-specific setup. | Prompting is structured and repeatable, often with saved templates. | Agents are integrated into procurement systems, not a separate chat window. | Orchestrated, continuously running agents across the function. |
| Governance & data | No policy exists on what can or can't be shared with AI. | Licenses exist, but usage and data exposure go untracked. | Basic usage tracking; some visibility into time saved. | Formal escalation rules and an audit trail for agent actions. | Agent performance is reviewed the way headcount performance is reviewed. |
| Org structure | Humans only — no role for AI has been defined. | AI is a personal tool, not yet a team capability. | Team-level AI champions; still entirely human-initiated. | The org chart begins naming agents as owners of specific processes. | A true hybrid org chart — humans and agents with named scope. |
Swipe sideways to compare every level.
Moving up
The jump between levels is rarely about buying more software. It's about what changes in how the team actually works.
1→2Unaware to Copilot
Get a sanctioned tool live, run baseline AI-literacy training, and set a clear policy on what can and can't be shared with it.
2→3Copilot to Augmentation
Build prompt libraries by role, name an AI champion on each team, and start tracking the hours it's actually saving.
3→4Augmentation to Agentic
Pick one narrow, high-volume process and pilot a single agent with a human checkpoint before scaling any further.
4→5Agentic to Autonomous
Formalize agent ownership and governance, then deliberately redesign the org chart instead of layering agents on top of the old one.
Self-check
Three questions that usually settle it.
01
Do you have a shared prompt library by role, or is everyone starting from a blank chat window each time?
02
Has any process actually been redesigned around AI, or has AI just been added on top of the old one?
03
If an agent ran a step of your sourcing process today, would anyone notice — or know what to do next?
A shared prompt library is the step out of Level 2. One agent, on one narrow process, is the step after that.