EXPOSED
Actuarial work is entirely screen-based, and the mechanical layer — cleaning claims data, running reserve triangles, building pricing models in Excel/R/Python, drafting exhibits and rate filing memos — is exactly what AI does well and cheaply now. What holds is the regulatory structure: statutory reserve opinions and rate filings require a credentialed actuary (FSA/FCAS, appointed actuary status) who is personally accountable for assumption choices under NAIC and state rules. The modal actuary today spends more time on model plumbing than on defensible judgment, so the occupation compresses toward fewer, more senior signers.
Mixed — a routine tier and a judgment tier. Chain-ladder development, loss-ratio trending, mortality table blending and experience studies are pattern work an LLM plus a scripting environment reproduces in minutes, but setting the assumption basis for a Statement of Actuarial Opinion — deciding whether last year's loss trend still holds after a reinsurance program change or a court ruling on bad-faith claims — still requires someone who knows the book, which is why this sits at 10 rather than in the 4-6 range of pure reporting roles.
Fully desk- and screen-based. Nothing in the job leaves the terminal: data warehouse queries, AXIS/Prophet/GGY runs, Excel and R, and the occasional trip to a rate hearing or board meeting that is itself a conference room, so the 1 reflects only that you physically show up somewhere.
Licensed human required and personally liable. An appointed actuary signs the reserve opinion under state insurance law and is subject to the ASOPs and the Actuarial Board for Counseling and Discipline — real professional exposure — but the credential is conferred by the SOA/CAS rather than a state licensing board with practice-prohibition power, and most pricing, valuation and reinsurance analysts work for years under someone else's signature, so this lands at 12 rather than in the high teens where physicians and PEs sit.
Some relationship component. Regulators, reinsurers and CFOs care about the credential and the opinion, not about you specifically; consulting actuaries at Milliman or WTW who hold pension and self-insured clients through triennial valuations do have real relationship stickiness, which is what pulls this up from the 3-4 of an in-house valuation analyst to 8.
Exists to be accountable for ambiguous calls. Picking a discount rate, a lapse assumption, or an IBNR selection at the low end of a reasonable range is a call with no verifiable right answer that moves millions of dollars of reported surplus and can be second-guessed by a regulator, an auditor or a plaintiff's expert years later — the 14 is earned by that unfalsifiable-yet-defendable quality, not by decision volume.
Has AI actually changed your work?