// Roles
The unstaffed seat isn't the AI engineer.
It is the person who sits between the model and the underwriter. Carriers are hiring model builders rather than the people who redesign the work. These five pages answer one question, desk by desk: what changes on Monday.
// The situation
The hiring is happening one layer away from the work.
Shift Technology reviewed 100 current US insurance job postings. 72% of technology and data-science roles explicitly mention AI, against 6% of operating roles in claims, underwriting and SIU. Only 10% of operating roles mention exception handling, escalation or QA — the exact skills a deployment creates demand for.
Shift Technology · 100 US insurance job postings · 2026 · vendor research.
So we scope that seat with the build.
Who reviews the output, who owns the correction loop, and who is named on the override go into the first statement of work, with a written runbook and the training behind it. The review role is a deliverable, not an assumption about your org chart.
// The five desks
What changes depends on which desk.
Underwriter
The file arrives assembled, contradictions surfaced rather than resolved, and appetite screened before the queue. We are not touching the part that is actually underwriting.
Underwriter →ClaimsClaims professional
The most negative occupation in the analysis, and the data says the opinion was earned. Every extracted fact cites its page, or the build does not ship.
Claims professional →OperationsOperations leader
Capacity without a hiring cycle, in sequence: instrumentation, then intake, then exception routing a person owns, then the QA layer, staffed.
Operations leader →PortfolioCUO & portfolio leader
Accountable for loss ratio, appetite, aggregation and rate adequacy, on data that arrives late by construction. Ingest first, dashboard second.
CUO & portfolio leader →TechnologyIT & data leader
Sanctioned integration surfaces only, the semantic layer as a deliverable you own, and the exit plan written into the first scope.
IT & data leader →These pages are written from the desk rather than the buyer persona, because the cost of a broken workflow is paid by the person working it.
// What's hard about this
Two things the role axis gets wrong.
The titles don’t map across carriers.
A VP of operations owns claims throughput at one carrier and underwriting support at the next. The same is true of every title on this page, which is why a build scoped from a job description misses the person who actually holds the workflow. So we scope by workflow and by accountability, and we read your org chart in week one instead of inferring it from a title.
The seat AI creates has no job family yet.
Exception handling, escalation and QA barely appear in operating-role postings, which means there is no pay band, no career path and no requisition template to hire against. So the role definition, the runbook and the training material ship as deliverables, and the reviewer is trained before go-live rather than named after it.
Shift Technology · 100 US insurance job postings · 2026 · vendor research.
What changes on your desk, and what doesn’t.
Five desks, five pages, and the same discipline on each: the work as it actually runs, the constraint named before you find it, and the review seat scoped with the build.