Carlson BCM01
Automate the processing completely
Where no decision is retained, the zero call center logic applies in claims too.
Insights · October 19, 2026
AI Can Replace the Work. Who Replaces the Expertise?
The claims operation reveals what a carrier must still know when AI does the work.
The Issue
Carriers are asking the same AI question across the value chain.
In billing, policy administration, finance, HR, IT, sales and marketing, carriers are asking how much of the work AI can eliminate. Claims shows why the answer cannot be the same everywhere. In the claims operation, carriers are asking whether AI can eliminate the call center and whether it can eliminate the adjuster, usually without distinguishing between them.
Travelers
Has cut its claim call center population by a third and is consolidating four claim call centers into two.
Lemonade
Reports that its AI takes 96 percent of first notices of loss without human intervention, and that claims it is not authorized to settle go to human claims experts.
Allianz
Uses AI agents to process small claims, and states that payout decisions are never automated.
The labor market
Entry-level adjuster postings are down about half since early 2024. Senior postings are about 80 percent above 2017 levels.
The entry-level work is disappearing while demand for senior judgment rises. Yet both questions still treat claims as a labor pool waiting to be automated. The call center and the adjuster are not the same kind of work. One is processing. The other is made of decisions.
The Distinction
Zero call center is a strategy. Zero adjuster is a liability.
The Evidence
The call center holds no retained decisions. Claims holds 31.
I took every business capability in my Carlson BCM insurance model owned by service roles and by claims roles, and placed each in one of three categories: AI can run it end to end; AI can do part of the work while a person stays in it; or a person must make the decision. Each capability falls in exactly one, under the role accountable for it.
Business capabilities
Call center76 assessed
Claims92 assessed
AI can run it end to end
4423AI does part of the work; a person stays in it
3238A person must make the decision
031Claims findings come from the Group and Individual model; commercial claims is only lightly represented.
The difference is not how much work AI can perform. It is what the carrier must remain capable of deciding.
The call center can shed its mass-processing structure because no retained decision holds it in place. Its obstacles are execution, not authority. Claims cannot be judged on the same basis. A person must make the decision in 31 of its capabilities today, and every one is rated high for risk and compliance value: determining a coverage position, determining causation and liability, settling a claim, denying a claim, terminating a benefit.
What Does Not Move
Those decisions may move to AI over time. The carrier’s accountability for them does not.
The Insight
The work AI takes is the work that builds the judgment.
How far AI can go · the weakest limit decides
A(c, t) = min [ I(c, t), D(c, t), G(c, t) ]
I
Intelligence
What AI can do with the work.
Moves fastD
Delegation
What the carrier and its regulators allow AI to decide.
Moves slowlyG
Governance
The capacity to make the decision, or govern the AI that does.
Moves as fast as people developToday the carrier still has its experienced people, so intelligence and delegation set the limit and cutting the workforce looks free. One of the 31 decisions shows where governance comes from.
Business capability
Ability to determine a coverage position
Accountable role: Senior Claims Examiner
Authority
Retained by the carrierHigh
How far AI can take it
AbsorbHigh
AugmentHigh
AccelerateHigh
AutonomizeLow
Intelligence it requires
RulesHigh
AnalyticalHigh
GenerativeHigh
Reading the policy against the facts
01Ability to identify an applicable benefit provision
02Ability to identify an applicable exclusion
03Ability to identify an applicable limitation
04Ability to identify an applicable rider
Taking and recording the position
05Ability to determine a coverage position
06Ability to register a coverage position
The four functional capabilities in gold are not simply tasks AI can perform. They are how an examiner develops the judgment to make this decision, or to govern the AI that makes it. Automate them without replacing the experience they provide, and governance becomes the limit: the carrier keeps its senior experts for now while losing the means of producing their successors.
The Paradox
The machine becomes more capable. The institution becomes less capable of governing it.
The Implications
Delegating the decision does not delegate the answer for it.
If the AI is defective, its maker answers for the product. If it is applied badly, the carrier answers for the deployment. Either way, the carrier is still the one examined, sued and held to account, and someone inside it must still understand the decision.
Insurers have made this trade before. I worked on the transition at a specialty carrier that had handed the maintenance of its policy, claims, billing and sales systems to an offshore provider, and when it changed providers, no one inside could explain the systems its business ran on. It still owned them and still answered for them, but it no longer understood them.
02
Design the apprenticeship on purpose
The 38 capabilities where a person stays in the work are the population to examine. Practitioners identify which experiences actually build judgment, and those are protected.
03
Measure the bench
For every retained decision, the people qualified to make or govern it, those developing toward it, and the time to proficiency.
The Board Test
Before approving an automation business case, require an assessment of the expertise needed to make or govern every decision it touches, and a funded plan to sustain that expertise as the workforce changes.
Automation can remove the person who performs a capability. It cannot remove the carrier’s need to understand it, govern it and answer for it.