Carlson BCM
Insight · October 5, 2026

Where Can an AI Agent Act?

Most insurance work can be automated. Very little requires an agent. What the machine may decide is a separate question.

Where can an AI agent act? Of 2,427 functional capabilities, 67% can be automated and 4% retain human authority. Of 406 business capabilities, 24% need intelligence and 8% qualify as agents. An example: the business capability Ability to deny a claim.
Of 2,427 functional capabilities

Can be automated

67%

1,620 could be autonomized. Most insurance work can be automated.

Of 406 business capabilities

Needs intelligence

24%

97 have high fit for language models.

Of 406 business capabilities

Is an agent

8%

33 qualify for agentic design. 81 of the 97 that need intelligence are not agents. A model that can read and reason is not a worker that can hold a goal and act on it.

Of 2,427 functional capabilities

Retains human authority

4%

94 retain the human decision. 30 of them could otherwise be fully autonomized. The machine can do the work. It cannot own the decision.

An example · The business capability “Ability to deny a claim”

Four of its five functional capabilities can be autonomized, and carriers already automate much of the denial work. Yet the business capability does not qualify for agentic design: the work can be automated, but a person still owns the decision to deny. A denial shifts the loss to the insured, and the carrier answers for it.

Where that ownership has shrunk to a signature on denials reviewed in seconds, it has drawn class actions and new state law requiring a qualified professional to make the decision.

Derived from the Carlson BCM insurance corpus · Group and Individual · Commercial Property and Casualty

Download the plate
How it was derived

One population, two levels, each finding measured against one.

The population is 3,122 business operations capabilities across 17 value chains, 6 in Commercial Property and Casualty and 11 in Group and Individual: 116 enterprise capabilities, 470 business capabilities and 2,519 functional capabilities. IT platform and technical capabilities are excluded.

Business capabilities describe what the business does. Functional capabilities describe how, functionally, it does it. Automation and authority are properties of the work, so they are measured on the 2,427 functional capabilities scored for both. Intelligence and agency are properties of what the business does, so they are measured on the 406 business capabilities scored for both.

Every finding states its own denominator, and no comparison mixes the two levels.

The question is not what an agent can do. It is what the enterprise has authorized it to do.

All insights