Underwriting authority is the backbone of an AI agent competency matrix for underwriting teams. Agents gather risk information, pre-fill submissions and draft referral notes; underwriters must test the agent's summary against the source documents, recognise risks outside appetite, and keep every acceptance, decline and price outside guidelines with a person who holds the authority to give it.
How does underwriting authority shape the matrix?
Underwriters already work within written authority: classes they may write, limits per risk, and rules for referral. An agent fits inside that structure, not beside it. It may read submissions at depth 1, draft risk summaries and referral notes at depth 2, and in some lines apply standard rating to simple risks at depth 3, always below the most junior underwriter's authority.
The skills therefore track authority levels. A junior underwriter must check what the agent extracted from a broker's submission: sums insured, locations, claims history. A senior must judge the agent's referral note and decide whether its reasoning holds. The underwriting manager must decide whether the agent's evidence justifies any change in its depth. Some lines raise the stakes: under the EU AI Act, risk assessment and pricing for life and health insurance are listed as high-risk in Annex III, which brings duties of human oversight and staff AI literacy. The meaning of underwriting itself does not change; the evidence an underwriter relies on does.
How do you fill it in?
- Copy the template and add your authority bands as columns, lowest to highest.
- For each skill, enter 1 (understands), 2 (applies unaided) or 3 (decides exceptions and coaches).
- Where a skill relates to an agent capability, note its depth, so nobody oversees a depth 3 capability with level 1 skill.
- Re-check whenever underwriting guidelines change, since the agent's sources change with them.
The template
| Skill | Agent capability (depth) | Assistant underwriter | Underwriter | Senior underwriter | Underwriting manager | Evidence |
|---|---|---|---|---|---|---|
| Checks extracted data against the submission | Extraction (1) | |||||
| Weighs provenance of broker and third-party data | Extraction (1) | |||||
| Judges the agent's risk summary | Summary (2) | |||||
| Recognises a risk outside appetite the agent missed | Summary (2) | |||||
| Checks standard rating applied by the agent | Rating (3) | |||||
| Decides referrals the agent drafts | Referral note (2) | |||||
| Writes reasons for declines | Decline draft (2) | |||||
| Reads a certification result for the rating capability | All | |||||
| Reconstructs a quote from the record | All |
A worked example
A hypothetical commercial property insurer with four assistants, six underwriters, two seniors and a manager. Its agent extracts and summarises submissions and applies standard rating to small, simple risks at depth 3.
| Skill | Assistant | Underwriter | Senior | Manager | Evidence |
|---|---|---|---|---|---|
| Checks extracted data | 2 | 2 | 2 | 1 | 20 files a week re-checked |
| Weighs provenance | 1 | 2 | 3 | 2 | Scenario test |
| Judges risk summary | 1 | 2 | 3 | 2 | Peer review |
| Spots out-of-appetite risk | 1 | 2 | 3 | 3 | Referral audit |
| Checks standard rating | 1 | 2 | 3 | 2 | Rating sample |
| Decides referrals | 0 | 1 | 3 | 3 | Authority letter |
| Writes decline reasons | 1 | 2 | 3 | 3 | Reason audit |
| Reads certification result | 0 | 1 | 2 | 3 | Owner review |
| Reconstructs a quote | 1 | 2 | 2 | 3 | Walkthrough |
Assistants reach level 2 on checking extracted data at once, because it is the first place an error enters. A broker's schedule lists a warehouse's sum insured; the agent reads it into the summary; if the figure is wrong, every later step is wrong too.
Provenance matters in a particular way here. A broker's presentation is signed: a named party answers for it. A survey report commissioned by the insurer is stronger. A statement on the client's own website is external. Seniors at level 3 can tell when the agent's summary leans on the weakest source for its most important claim.
Standard rating at depth 3 needs a level 2 check from underwriters, through weekly sampling. If the rating capability's certificate lapses, every small risk becomes a held item for underwriters until it is recertified. Plan for it in the reviewer rota for the underwriting team, and set referral limits in the approver rule for the underwriting team. The connections and sources behind extraction are covered in the IT team's competency matrix.
Underwriting teams can evidence the review skills with Pryme Intelligence Academy's Operator certificate, whose module tests are set in insurance for insurance staff. The broader view is in AI agent skills by role.
Questions people ask
Can an agent accept a risk?
Only within a narrow, pre-approved class, below the lowest underwriting authority, and only once it has earned that depth on the insurer's own past cases. Everything else is a draft for an underwriter.
Who should own the underwriting agent?
The underwriting manager for the line it serves. They answer for its bounds and decide whether to seek a higher depth.
What should a new underwriter learn first?
To check the agent's extraction against the source document. It is quick, it catches most errors early, and it builds the habit of never relying on a summary alone.