I was in Berlin, sitting with a mid-size company’s Chief Procurement Officer. He was fully behind his company's ambition to become AI-first and better versed in the Procurement AI market. So, I asked how he used AI personally. He talked about the tools his department had licensed, and about using ChatGPT for his emails and research. So, I dug deeper.

"How do you use AI to support your own decisions?"

There was no answer. Just a confused expression, and a change of subject. That stayed with me.

By now most of us have ChatGPT, Gemini or Claude on our phones and in our browsers. We use them to plan travel, tidy up presentations, draft emails, and help with the children's homework. You already know how quickly they produce something usable, and how a lazy prompt gets a lazy answer.

You as CPO are being asked to think about AI strategy. AI governance in procurement. AI agents that you should implement in procurement decisions. Yet very few CPOs have ever used AI for a decision that mattered to them personally. Not to draft something faster. But to decide something better.

That gap does not need a project. Twenty minutes and one honest problem will close it.

Start With the Decisions You Are Worried About

While writing faster emails is nice, it shouldn’t be your end goal with personal AI use. Instead, you should be building a thinking partner that knows enough about your priorities to challenge you before a decision, rather than after it has gone badly.

Think about what is important for you:

  • The negotiation you're worried about

  • The CFO constantly challenging procurement numbers

  • The stakeholder who disagrees with your sourcing call

  • The decision someone will challenge

None of that needs an AI agent or an integration.

What is the Difference Between a Personal AI and a Procurement AI Agent?

You don't need to build AI agents to make your decisions for you; instead you should be using a personal AI LLM to help sharpen your own decisions. The distinction is fairly simple:

Personal AI

Procurement AI Agent

Purpose

Helps you think

Performs a defined task/process

Decision-maker

You

May act within its scope

Data

What you give it

Organizational systems

Setup

Minutes

Integration, governance, and security

Scope

Personal

Organizational

Table 1: how personal AI usage differs from a procurement AI agent

Say, if you have supplier strategy discussion with your Business Unit head, your personal AI will remind you "Here are three risks to be considered before that call." rather than telling you, "I have already drafted the terms and conditions that business units should follow in their supplier negotiations". Then it becomes operational and more of a procurement AI agent job. And operational AI needs everything; an operational capability needs an oversight, an audit trail and a named owner.

Use personal AI when the task is about your own thinking: preparing, challenging an assumption, comparing options or even getting a second perspective.

Use enterprise AI when the task touches organizational data, needs to scale, perform an action, or must survive an audit.

Personal AI helps the CPO think. Enterprise AI helps the procurement organization act.

You will need both, but neither removes the need for the other.

How to Build Your Setup in 20 Minutes

Before you actually build something, always consider your company's data governance policies. Use LLMs (ChatGPT, Claude, Copilot, Gemini, etc.) you already have and are allowed to use within your company (“behind the firewall”).

Simple steps you can follow:

Minutes 0–5: Create a Dedicated Space

This would be your dedicated space to save your instructions, outputs, inputs, drafts, etc.

Depending on your tool, that might be a project, a saved assistant, custom instructions or simply a dedicated conversation.

Minutes 5–10: Give it Context

Write down who you are and what this will be used for. Feed information that is generally in your personal notes or even in thoughts. Keep it simple, no big documents or strategies to review but actions and thoughts that can be used further. Use a simple structure:

ROLE & REMIT

[your scope, categories, reporting line]

CPO at a global manufacturing company. Responsible for €3B procurement spend across Europe.

CURRENT PRIORITIES:

  • Deliver savings

  • Reduce supplier risk

  • Improve business adoption of procurement

DECISION PRINCIPLES:

Balance savings, risk and speed. Protect supply continuity.

STAKEHOLDER MAP:

CFO wants savings. COO wants continuity. Businesses want speed and flexibility.

LIVE ISSUES:

Strategic supplier pushing a price increase. CFO questioning the savings target. Two BUs bypassing procurement

Minutes 10–15: Tell Your AI How to Think

Vague instructions produce vague output. Be specific about the behavior you want:

“Challenge my assumptions. Tell me what I am missing. Give me the CFO's objection before I hear it in the room. Separate what I know from what I am assuming. Show me perspectives I will not like. Do not decide for me.”

Notice what you are asking for. Resistance.

Minutes 15–20: Give It One Real Problem

Take something live and work through five questions in order.

  1. What am I missing?

  2. What assumptions am I making?

  3. What is the strongest counterargument?

  4. What should I verify before the meeting?

  5. What five questions should I ask tomorrow?

No integration, no IT project, no agent. Just you, a tool you already have, and a real problem.

Craft/Borrow your own personal framework

For example:

Context → Challenge → Input → Output → Judgement

Each stage fixes a specific failure:

  • No context gets you generic answers;

  • No challenge gets you agreement;

  • No defined inputs makes quality variable;

  • No output spec gets you an essay when you need three questions.

  • No judgement leads to a lack of evaluation and improvement.

Make it a habit

You have just left a board meeting. Before you get absorbed in another critical discussion, take 5 minutes for yourself. Just simply write down your observations, thoughts, non-verbal clues you picked up, main discussion points, agreements and disagreements, multiple perspectives, your own comprehension and what to be done next. Pour everything in that folder. Let your LLM remember it instead of you.

Ask AI to structure it:

Facts. Assumptions. Decisions. Open questions. Action points. Follow-ups.

You make hundreds of consequential decisions a year, scattered across notebooks and email threads. After a few months, you can ask: what did the board agree in Q2 budget review, and what exactly was the CFO's objection to our numbers?

Let the tool hold the details. You hold the judgement. It works as a sparring partner too. Before a difficult update, ask it to act as the CFO, then the risk committee, then the COO and tell you what each would challenge.

What You Can and Cannot Put In?

Have one conversation with legal or IT first. As a rough guide: public supplier information, market research and your own reflections are generally safe. Internal strategy, performance data and meeting notes need checking. Live tenders, supplier pricing, confidential contracts, NDA-covered material and personal data can be a red line depending on your company position and/or local regulations.

Another thing to consider is that your setup will absorb your biases and reflect them back. That feels reassuring, right up until you act on it and it gives bad results... It will work confidently from wrong information. And it will invent a supplier name or a clause in the same tone it uses for things that are true.

Check the output every time. Give feedback to your personal AI LLM on a regular basis. Continuous improvements to shorten conversation lengths reduce AI hallucinations considerably.

Why This Makes You a Better Leader?

The real return takes a few weeks to appear…

Daily use builds an instinct for how these tools fail. It sounds most certain exactly when it is wrong like the invented clause reads as smoothly as the real one. It stays generic until you give them real context. It agrees with you too easily, and the useful answer comes from the follow-up question, not the first response.

That instinct changes what happens next. Someone on your team will bring you an LLM-supported tool; contract review, spend classification, negotiation bot, supplier risk. It will demo beautifully, because demos run on curated data.

The CPO who has never used AI personally may ask about the price, timeline, and references. The CPO who uses it every morning cross checks what it is grounded on, what it does when it does not know, whether it can show its sources, and what happens when the supplier master record is duplicated or the contract is missing.

Then: “show me how it fails, not just the happy path.”

These questions come easily to anyone who has watched a model to be fluent and wrong about their own work.

Governance works the same way. Regulators across the world have settled on meaningful human in the loop for important decisions, not a checkbox in a workflow, but someone who knows what the system can and cannot be trusted to do. You cannot delegate that if you have never used it.

There is another more subtle effect, too. When your team knows you use these tools daily, people stop overselling and start telling you the facts.

The CPO in Berlin was not behind in AI adoption. He knew the market better than most people I spoke to. He had simply never used it to support his own decision making... Which made him vulnerable.

So, before you build AI for procurement, build it for yourself. Twenty minutes, one real problem, and a tool you already have.

Using AI personally doesn't make you an AI expert… But it does make you a more informed buyer, sponsor, and governor of AI.