How Crawford & Company’s Broadspire aligns technology and AI to client outcomes

AI insurance claims technology cuts Broadspire processing time 40%

Sarah Mills | GlobalBeat

Broadspire, the third-party claims administrator owned by Crawford & Company, said it reduced average claims processing time by 40 percent after embedding machine learning models across its liability, workers’ compensation and auto lines.

The company told reporters its new AI engine now handles 2.3 million documents annually and flags potentially fraudulent claims within 5 minutes of submission.

Insurers have pushed vendors to show measurable cost savings since loss ratios on U.S. commercial lines rose to 73.4 percent in 2023 from 68.9 percent in 2021, according to AM Best data.

Broadspire’s technology team trained the models on 12 years of anonymized claims records covering $18 billion in paid losses, chief technology officer Mohit Sharma said in an interview. The platform scores each incoming claim from 0 to 100 for complexity, litigation risk and reserve variance, allowing adjusters to prioritize high-impact files.

“The system is not replacing adjusters,” Sharma said. “It tells them where to look first.”

Early adopters included a regional workers’ comp carrier in California that cut average claim duration to 18 days from 31 days during a pilot, Sharma said. The carrier, which he declined to name, reported a 12 percent drop in legal expenses because the AI surfaced claims likely to result in attorney involvement.

The rollout began in January 2023 and reached all U.S. clients by December, Broadspire confirmed. Executives said the company spent $14 million on the project, funded through Crawford’s annual technology budget of roughly $90 million.

Competitors such as Sedgwick and Gallagher Bassett have introduced similar analytics dashboards, but Broadspire’s system updates risk scores every 15 minutes, compared with nightly batches used elsewhere, Sharma said. The difference matters when attorneys file liens or medical providers add new billing codes, he added.

Privacy regulators reviewed the platform in March after client insurers raised questions about feeding historical data into proprietary algorithms. Broadspire said the review found no violations because personal identifiers were stripped before records left client servers.

Consumer advocates warned that algorithmic scoring can embed past biases. “If older claims reflect discriminatory patterns, the machine will learn them,” said Birny Birnbaum, executive director of the Center for Economic Justice. He urged regulators to audit model outputs annually.

The National Association of Insurance Commissioners adopted a model bulletin in December that asks insurers to document how outside vendors use AI. Broadspire said it provided regulators with a 47-page disclosure covering training data, model performance and override protocols.

Background

Third-party administrators process about 42 percent of all U.S. commercial claims, according to a 2023 Conning study. Carriers outsource the work to cut overhead and gain access to specialized medical and legal networks.

The sector has lagged behind personal auto and homeowners insurers in adopting AI because commercial claims involve bespoke policies and higher dollar amounts, said Donna Popow, legal counsel at the Self-Insurance Institute of America.

What’s Next

Crawford will extend the AI engine to its property operation in Britain during the third quarter, chief executive Rohit Verma told analysts on April 30. European clients will have opt-in rights to exclude claims data from training sets to comply with the EU AI Act, he added.