A New Variable in Insurers' AI Race: 'Token Costs'?
Computing Costs: The "Token Dilemma"
Complex Insurance Reviews... Cutting Input Raises Error Concerns
Case of Sharp Rise in Token Use at an Overseas Insurer
As insurers expand their use of artificial intelligence (AI), managing computing costs is emerging as a new challenge. The more information they input to process complex insurance policies and claims documents, the more tokens they use. AI agents may also consume large numbers of tokens in a short time when they repeatedly search and analyze information. Costs that seemed manageable during pilot operations could become a strain on profitability as the technology is deployed for large-scale operations.
According to the insurance industry on November 18, Deloitte, a global consulting firm, said in its recently released 2027 insurance industry outlook that operating costs associated with token usage and model selection will become important areas for insurers to manage as they expand their use of AI agents. In its report on insurers' AI investments, Deloitte said, "Traditional IT budgeting alone may no longer be sufficient," adding that insurers "need to understand AI costs and the value they generate to prevent costs from rising excessively beyond expectations while continuing to invest in innovation."
A token is the basic unit that a large language model (LLM) uses to process information. Generative AI divides input text into tokens for analysis and also generates responses in token units. When insurers use external AI models through an application programming interface (API), they generally incur costs based on the number of tokens used for input and output. Some reasoning AI models may also use additional tokens internally while generating responses, which can increase costs.
This cost structure matters in insurance because insurance work is information-intensive. For example, determining whether to pay an insurance claim requires checking the policy in effect when the customer enrolled, coverage terms, exclusions, the circumstances of the incident, medical records, and past claims history. Unlike general customer inquiries, this work involves checking conditions scattered across multiple documents and applying them to each individual case. If an insurer provides the AI with the entire policy and all claims documents, the number of input tokens increases. Repeated rounds of processing may also cause costs to accumulate as the same information is processed again. However, if some information is omitted to reduce token usage, the AI could miss exceptions or contract terms needed to determine whether a claim should be paid.
A financial industry official said, "Claims documents may contain duplicate medical records or omit test results that are needed. If these documents are input into AI as they are, unnecessary processing may result." The official added, "Rather than having one model handle every task, insurers may be able to work more efficiently by using models suited to each step, such as document classification or information extraction."
Hot Picks Today
"Two Spend 500,000 Won in a Flash, Yet Wait Two Hours?" ... 'Dreamland' Shunned by Kids and Foreign Tourists
- The Era of Getting 200 Million Won After 30 Years Is Over... A Japanese Bank's "Perpetual Department Head" Who Just Clocked In Cries Out
- Spain Hit by Its Worst Tragedy... Lee Kangin, Moreno and Real Madrid Join Soccer World in Mourning
- They Said, “Make It Your Home for Life” - Then Suddenly Kicked 200 Passengers Off... What Happened to the Around-the-World Cruise?
- "130 Rice Bowls Sold Out in a Day"... 'Unfiltered Face' Adds 8 Million Won to Night Market Sales: China's Bungeunwoo
Analysts say token costs could become more conspicuous as insurers expand their enterprise-wide use of AI from pilot operations to actual large-scale processing. In an analysis published last September, Deloitte cited a case in which one insurer's monthly AI token usage rose from 3 million to more than 150 million in less than a year. As AI use expanded, token consumption surged in a short period. Deloitte noted that "costs that seemed manageable at the proof-of-concept stage can become a significant burden as operations scale up."
An industry official said, "When the results of an insurance review are uncertain, an agent may repeatedly search and analyze information, using more tokens than expected even for a single task." The official added, "External AI models have different input and output token rates, as well as different fees depending on the provider. Even when a less expensive model is used, errors can lead to more reviews by staff and drive up total processing costs, making it difficult in some cases to demonstrate the benefits of AI investment."
© The Asia Business Daily. All rights reserved. Unauthorized AI training and use prohibited.