Examples of companies turning AI from a cost center into operating cost savings, customer service leverage, logistics optimization, and developer productivity
Which buyers of AI are getting real value? How?
The sellers of AI have been obvious winners: chips, power, data centers, neoclouds.
But buyers have been less obvious. Who's turning AI from a cost center into real, hard ROI?
UnitedHealth Group ($UNH) plans roughly $1.5B of AI spend in 2026 and expects nearly $1B of AI-enabled operating cost reductions. One target is outpatient medical coding, the back-office job of turning a doctor visit into the right billing and claims codes, where UNH reported a 73% productivity gain. With AI, coders clear far more visits per hour, which lowers admin labor per claim and lets them process more volume without adding people in proportion.
In a similar vein, Astrana Health ($ASTH) is spending about $15M this year on AI and automation and expects at least $10M in annual operating efficiencies by early 2026. The target work administrative grind: insurance pre-approvals, claims, and clinical paperwork. Fewer manual hours per task = more operating leverage as patient volume grows.
Booking Holdings ($BKNG) lowered customer service costs in absolute dollars even as volume grew about 10%. Bookings has a lot of agents helping travelers with issues during and after booking. More self-service resolution = fewer human touches per booking, so support cost falls even as transactions rise. Booking also has a broader savings program targeting $500M to $550M of annual run-rate savings, but that is wider than AI.
XPO ($XPO) is a freight carrier. Linehaul is the long-haul movement of trucks between terminals, and it is about a $1.6B cost category. AI routing cut empty miles, trucks driving with no paid freight, by 12%, and cut inefficient reroutes by more than 80%. Fewer wasted miles, less fuel, fewer driver hours, and a stated dollar bridge.
Walmart ($WMT) says AI tools saved about 4M engineering hours last year, with more than 40% of new code AI-generated or AI-assisted.
Citi ($C) reports roughly 100,000 hours of weekly developer capacity from AI code review across a very large bank.
But what's missing is a concrete figure for cost savings or revenue contribution. It's hard to put a dollar figure to faster engineering cycles! Ironic, since AI has made by far the biggest impact in software development.
Codex, using CapRelay MCP, gave us a list of 100 companies that reported some gains from applying AI internally. All we had to do was type in a single prompt. We chose a handful to write this article. You can try CapRelay for free.