How to run calibration sessions for procurement officers

Pick ten real purchase requests the agent held in the last fortnight, have each procurement officer decide them alone with a written reason, then compare. Spend the session on the cases where officers split, agree the right answer and why, record it as a gold standard, and send the result to whoever owns the agent's brief.

What is a procurement calibration session, and who owns it?

A procurement agent typically reads purchase requests, checks them against the catalogue and policy, compares quotes, drafts purchase orders for an officer to release, and raises routine orders inside agreed limits. Whatever it cannot settle, it holds for a named officer. Calibration is the habit of getting those officers to settle the same held case the same way.

The head of procurement owns the practice. A respected officer facilitates. The agent's owner attends often enough to hear how officers read the drafts, because their agreed reasons become changes to the brief. Without calibration, two officers apply two versions of the threshold policy, suppliers get different answers, and the agent receives contradictory corrections.

Procurement has a particular problem: much of the policy is judgement dressed as rules. "Value for money", "justified single source" and "genuine urgency" all leave room. Calibration is where the room gets narrowed.

How do you run the session, step by step?

1. Agent, depth 1: pulls the candidate cases. It lists held requests from the period with the verdict, the bound that decided it and the officer's final action.
2. Facilitator: chooses eight to twelve cases. Include splits, near-threshold orders, one new supplier and one renewal with a price rise. Remove officer names.
3. Each officer: decides alone before the session. Approve, change or decline, with a two-line reason and the source they relied on.
4. Agent, depth 1: tabulates the spread. How many approved, changed or declined each case.
5. Group: discusses the widest splits first. Reasons, not outcomes. Which document did you trust? Which limit did you think applied?
6. Head of procurement: confirms the gold standard for each disputed case, or sends it to the policy owner if the policy itself is unclear.
7. Agent owner, with a builder: turns standards into changes to the brief or bounds, adds the cases to certification and re-certifies before the change goes live.

Which procurement cases belong in the pack?

The best cases are the ones officers argue about in the corridor. Five kinds earn a place in almost every session.

• Split orders. Three requests from the same cost centre, each just under the approval threshold, raised on consecutive days. Is it a pattern or coincidence? The agent can spot the pattern; officers must agree when it becomes a decline.
• Single-source justifications. A requester says only one supplier can do the work. Officers need a shared view of what evidence makes that credible.
• New supplier with changed bank details. The supplier's onboarding form arrived by email. Under provenance rules, a detail that only arrived from outside cannot authorise a payment, so the agent holds it. Officers must agree what verification clears it.
• Renewals with uplift. A contract renews with a price rise above inflation-linked terms. Approve, renegotiate or escalate?
• Urgency claims. "Needed by Friday or the project stops." Officers should agree how much urgency changes the checks, which is usually not at all.

Keep a running gold-standard library. Each entry holds the case, the agreed decision, the reason, the sources that counted, and the date the head of procurement confirmed it. New officers learn from it; the builder tests against it.

Which controls keep calibration honest, and what evidence does each leave?

Independent decisions first. Officers submit before the session. Evidence: timestamped submissions that precede the discussion.

Anonymity during discussion. The spread is shown without names. Evidence: the session pack carries case numbers only.

A named owner for each gold standard. Evidence: the confirmation entry with the head of procurement's name.

Policy gaps go to the policy owner. Calibration does not quietly rewrite policy. Evidence: a logged referral and the policy owner's answer.

Agent changes are re-certified. A gold standard that changes how the agent decides is tested on replayed cases before it applies. Evidence: the certification run that followed the change.

What should you measure after each session?

• Agreement on shared cases, session by session. It should rise and then settle.
• Overturned decisions in sampling. A second officer re-checks a sample of released orders each week. Fewer reversals on calibrated issues means the standard has landed.
• Corrections to the agent by cause. If officers keep correcting the same thing after the brief changed, the change did not work.
• Time held cases wait. Calibration should make officers quicker, because they spend less time second-guessing.

Do not use calibration results in individual performance reviews. Officers must feel free to show the decision they would really have made.

Where do procurement calibration sessions usually go wrong?

The loudest buyer wins. Senior category managers can dominate. The facilitator asks the quietest officer to explain their reasoning first.

Cases are too easy. A pack of clean catalogue orders produces full agreement and teaches nothing. Choose the ones that caused friction.

Standards never reach the agent. The group agrees, writes a note, and the brief stays the same. The agent then keeps holding or drafting the old way, and officers keep correcting it. Make the agent owner a standing attendee.

Sessions stop once things settle. Calibrate again after any change to thresholds, the agent's depth or its sources, and whenever sampling shows drift. Teams in other functions face the same pressure; compare how to run calibration sessions for compliance officers and how to run calibration sessions for loan officers, where the case mix differs but the discipline is identical. The general method is in how to run calibration sessions for AI agent reviewers, and the wider programme sits in our guide to training teams for AI agents.

Where the Academy fits

Pryme Intelligence Academy's Operator track teaches calibration inside its human review and running the rhythm modules, with module tests set in the learner's own industry and role, so a procurement officer practises on purchasing cases. The practicum then checks two weeks of real review decisions against a high accuracy bar with no missed red line, which is calibration measured on the live queue.