How to run calibration sessions for underwriters

Underwriters calibrate by writing terms on the same submissions independently, then comparing. Take eight to twelve risks the agent referred or drafted terms for, have each underwriter decide accept, load, exclude or decline with a reason, then discuss where terms diverge. Record a gold standard for each and change the agent's brief so its drafts follow it.

What are underwriters calibrating?

Underwriting guidelines leave deliberate room. Two underwriters can accept the same risk, one with a modest loading and one with an exclusion, and both sit inside authority. Over a book, that room becomes inconsistency: similar risks priced differently depending on who picked up the referral. When an agent drafts terms, the inconsistency multiplies, because the agent is corrected in two directions at once.

An underwriting agent typically reads the submission and supporting documents, extracts risk data, checks it against guidelines and appetite, drafts terms for an underwriter to issue, and refers anything outside its bounds. Its drafts sit at depth 2: a person issues the terms. Referral rules, authority limits and the agent's bounds decide which risks reach which underwriter.

The chief underwriting officer or the head of the line owns calibration. A senior underwriter facilitates. The agent's owner and, where pricing models are involved, the pricing actuary attend.

How do you run the session?

1. Agent, depth 1: assembles referred and drafted submissions from the period, with the draft terms, the guideline clauses applied and the underwriter's final terms.
2. Facilitator: picks the pack. Include one risk with incomplete information, one with adverse history, one near the edge of appetite and one where the broker pressed for speed. Remove underwriter names.
3. Each underwriter: writes terms alone, with the reason and the evidence relied on.
4. Agent, depth 1: lays the terms side by side.
5. Group: discusses the widest divergence first. Ask which guideline clause each person applied and how they read the evidence.
6. Chief underwriting officer: confirms the gold standard. Questions about appetite go to the underwriting committee; questions about price adequacy to the actuary.
7. Agent owner and builder: update the brief or referral bounds, add the cases to certification, and re-certify before the change applies.

Which submissions teach the most?

Declared versus verified history. The broker's submission states three years claims-free. The insurer's own claims system, a sealed source, shows a small claim two years ago. Underwriters must agree which governs, how the discrepancy affects terms, and what the agent should do when the two disagree. It should hold.

Missing information. A property risk arrives without a survey. Some underwriters quote subject to survey; others decline to quote. A gold standard settles it.

Edge of appetite. A trade the guidelines list as "refer". The agent refers it, correctly. What happens next is pure judgement, and the place calibration earns its keep.

Loading or exclusion. Same risk, two remedies. Agree which the line prefers and when.

Broker pressure. "Need terms in an hour or we lose it." Urgency does not change the evidence required. Agree that, and agree the reply.

Keep each case realistic. Include the full submission, the broker's covering note and any loss runs, not a tidy summary. Underwriters diverge most when they must decide which document to believe, and a summarised case hides exactly that judgement.

For personal lines, especially life and health, remember that risk assessment and pricing of individuals is treated as high-risk under the EU AI Act. Calibration records are part of how you show that people, not the agent, set terms for those risks.

Which controls make calibration credible?

Independent terms first. Evidence: submissions timestamped before discussion.

Clause references. Each underwriter names the guideline clause behind their terms. Evidence: the clause on each submission.

Authority respected. Calibration does not extend anyone's authority. Evidence: gold standards checked against the authority schedule.

Re-certification after change. Evidence: the certification run following a brief or bound change. The agent's certificate lapses after 90 days in any case, so its renewal replays the gold standards.

Library versioning. Evidence: each gold standard records the guideline version it rests on.

What should you measure?

• Spread of terms on shared cases, session by session.
• Peer review findings. Most underwriting teams already peer review a sample of files. Track how often reviewers disagree on calibrated issues.
• Agent drafts issued unchanged, by risk type.
• Referral turnaround, which should improve as underwriters stop seeking second opinions on settled points.

Where does it go wrong?

Experience becomes veto. The longest-serving underwriter's view becomes the standard because nobody challenges it. Have them speak last.

Price discussions swallow the session. Calibration is about how guidelines are applied. Rate adequacy belongs to pricing.

Only declines get discussed. Accepted risks with generous terms matter as much to the book.

The agent never hears the outcome. Without the agent owner, the brief keeps drafting yesterday's standard.

For the theory, read what calibration is. The general method is in how to run calibration sessions for AI agent reviewers, and calibration sessions for operations managers shows the same discipline on queue decisions. The full programme sits in training teams for AI agents. Pryme Intelligence Academy covers calibration in its Operator track, with module tests set in the learner's own industry and role.