The safest first win for claims AI is not deciding whether a claim should be paid. It is getting the file ready for that decision.
That distinction matters as insurers look for practical places to use artificial intelligence without handing high-stakes claims judgment to a black box. Claims departments are full of work AI can improve quickly: reading documents, summarizing notes, comparing photos, pulling policy language, checking prior losses, surfacing public records and consolidating information that may sit across multiple systems. The risk rises when that support layer becomes the decision-maker.
The National Association of Insurance Commissioners says AI is already being used in claims processing to estimate repair costs or assess damage using photos and historical data, while warning that AI-generated information should be reviewed carefully when used for important decisions. A recent fraud release from Aviva shows why that balance is becoming more urgent: the insurer reported a growing number of claims supported by AI-generated images and manipulated documents, particularly in motor insurance, and said advanced analytics and AI-enabled tools are being used with human oversight to identify suspicious claims earlier.
For Rob Galbraith, CEO of Forestview Insights and a former insurance innovation leader, that is the right starting point: AI as an evidence workbench, not an autonomous claims authority.
Galbraith sees meaningful interest in claims because so much of the adjuster’s work begins with information gathering. “So much of the time today might be simply data information gathering from eight, 12 different sources, whatever, and consolidate into one,” he said. The next step is where human judgment belongs: the adjuster uses claims training to decide “how do I want to handle this account moving forward, what are the next steps that are required.”
That is a different proposition from letting AI deny a claim, set a settlement value or trigger an investigation without meaningful review. The lower-risk opportunity is assembling the evidentiary record faster and more consistently: policy terms, loss history, photos, repair estimates, weather information, communications and prior claim notes brought into one workflow before the adjuster decides what they mean.
It also gives claims leaders a clearer business case: cycle-time reduction, fewer missed documents, better fraud triage and a stronger audit trail when a decision is challenged.
Galbraith’s caution is straightforward. AI may be able to perform many tasks, but insurers do not always understand how a model reaches its output. That gap becomes dangerous when the output affects a claimant.
“We’re kind of having the AI assist the user, but we’re still having that human in the loop to kind of make decisions before we, for instance, deny a claim,” he said.
The distinction is especially important because claims decisions are visible to regulators, policyholders and courts in a way many internal efficiency tools are not. A bad summary can be corrected; a wrongly denied claim creates a compliance, reputation and customer-trust problem. Deloitte has similarly warned that overemphasis on AI-driven automation during claims processing can reduce the human touch needed in customer servicing, particularly when policyholders are already dealing with loss.
That does not mean claims teams should avoid AI. It means the first deployment should be designed around traceability. Claims leaders should know what source material the system used, summarized and flagged before an adjuster acted. Evidence gathering is useful because it can make the human decision better. It becomes risky when the system’s output is treated as the decision itself.
For carriers, the practical investment is less glamorous than autonomous claims: document ingestion, image analysis, workflow integration, source tagging, escalation rules, quality control and explainable summaries. The goal is to reduce the adjuster’s search burden without reducing the adjuster’s authority.
That makes claims AI a strong Transformation use case because it sits at the intersection of business function and technology capability. It gives executives a focused place to invest: not “AI for claims” broadly, but AI that improves evidence collection, file completeness and investigation prioritization. Those categories are also easier to measure than sweeping promises of automated adjudication.
The C-suite question is not whether AI can touch claims. It already does. The question is which parts of the claims process should become more automated first. Galbraith’s answer points to the work before the decision: gather the facts, organize the record, surface what may need attention, and leave the judgment where accountability still sits — with the insurer’s people.
