Insurers keep AI inside old workflows

Insurance carriers are adopting artificial intelligence inside familiar underwriting and claims routines rather than replacing those routines outright. Recent industry commentary, trade reporting and regulator guidance all point to the same pattern: AI is being used to summarize files, score submissions, support triage and accelerate review, while human referrals, approvals and legacy handoffs remain in place. That approach is showing up at a moment when insurers face tighter governance expectations from state regulators and continued pressure to document how automated tools affect decisions. ### Where is AI actually showing up inside insurance operations? InsuranceNewsNet described the current pattern as carriers applying AI to existing underwriting and claims processes rather than redesigning the workflow itself. That framing matches a broader industry view that adoption is landing first in narrow operational steps where speed gains are easiest to measure. (insurancenewsnet.com) IBM said insurers are primarily using AI to boost operational efficiency, with 40% of AI spend allocated to operational effectiveness and cost reduction. In a separate insurance report, IBM also said many early generative AI uses center on tasks such as question-answering and support inside existing business processes. ### Why are carriers choosing narrow use cases instead of full redesigns? (insurancenewsnet.com) New York’s Department of Financial Services said on July 11, 2024 that insurers using AI and external consumer data should establish governance and risk-management frameworks to mitigate potential harm to consumers. The circular letter covers underwriting and pricing, and it ties AI use to existing insurance-law obligations around unfair or unlawful discrimination. (ibm.com) The NAIC adopted its Model Bulletin on the Use of Artificial Intelligence Systems by Insurers on Dec. 4, 2023, and its adoption map shows multiple states have since implemented it in bulletins or related guidance. That regulatory backdrop favors use cases carriers can explain, test and document inside existing controls rather than wholesale operating-model changes. That last point is an inference from the guidance’s emphasis on governance, transparency and compliance. (dfs.ny.gov) ### What does “inside the old workflow” look like in practice? InsuranceNewsNet’s reporting on insurance AI adoption said large language models are being used to process large amounts of data and speed claims processing, underwriting and other tasks. Those are acceleration functions, not necessarily changes to who makes the final call. IBM said insurers are “walking a tightrope” between building new generative AI capabilities and managing risk and compliance, while customers remain concerned about privacy and inaccurate AI-generated information. (content.naic.org) That combination helps explain why many deployments are aimed at helping adjusters, underwriters and service teams work faster inside existing review paths. (insurancenewsnet.com) ### Does this mean insurers are skeptical of AI? Equisoft executive Brian Carey told InsuranceNewsNet that misconceptions are a barrier to broader AI adoption in insurance. That suggests some of the restraint is about confidence and implementation readiness, not only technology limits. IBM said insurance CEOs were nearly split on whether generative AI is more of a risk or an opportunity, at 49% and 51%, respectively. (ibm.com) In the same body of research, IBM said investments were expected to rise as organizations moved from one or two pilots to more functions, but operational trust remained central. ### What should readers watch next? The NAIC said in 2025 and 2026 its Big Data and Artificial Intelligence Working Group has been developing an AI Systems Evaluation Tool for regulators to use in exams and reviews. (insurancenewsnet.com) That tool, together with state adoption of the model bulletin, is likely to give the next concrete evidence of how deeply insurers have embedded AI and where they still keep humans and legacy controls in the loop. (content.naic.org) (ibm.com)