BCG Tells Insurance CEOs to Build AI Once and Scale It: 10% Algorithms, 20% Tech, 70% People
Insurers do not have an AI shortage; they have dozens of pilots and no way to scale them. BCG’s Sebastian Schmoeger, Christopher Freese, Yohei Takabe, Stefan Bleyhl and Christoph Fritsch make that argument in The Insurance CEO’s Blueprint for Scaling AI. Their answer is to build common capabilities once and adapt them across businesses instead of funding each unit’s pilot separately.
The numbers attached to the six priorities are ranges, not guarantees. A shared conversational layer can reach 50 to 70 percent call containment and a 20 to 30 percent service-to-sales improvement. Process redesign can yield 30 to 60 percent efficiency gains in underwriting and claims, and agent-assist tools have produced 20 to 50 percent gains and up to 90 percent fewer call transfers. Technology delivery gains run 10 to 20 percent at AI-assisted level, 25 to 40 percent at AI-driven and three to five times at fully agentic. The authors also expect insurers to become discoverable to AI tools first, then conversational, then transactional, with agent-to-agent transactions likely to appear first in B2B.
The CEO checklist is the useful part. Concentrate on a few owned initiatives, redirect an estimated 30 to 40 percent of planned project resources, build shared data and infrastructure, and limit partners to avoid vendor sprawl. Then split effort roughly 10 percent algorithms, 20 percent technology and data, and 70 percent people and processes.
That split matches BCG’s earlier work on health insurers, which recommended the same 10-20-70 rule and put gains from reshaping workflows around agents at 30-80% versus 10-20% for off-the-shelf tools. The measurement gap is the other half. A BCG study of Latin American technology companies found 85 percent run AI agents in production while only 22 percent validate the financial impact.
The lesson travels beyond insurance: pick fewer initiatives, own them, and prove the outcome before starting the next pilot. The figures are BCG’s own ranges, so treat them as targets to test, not benchmarks.