AI in regulated industries: BofA and S&P on the huge gains to be had, and the risks of rushing in

Bank of America plans to double its AI expense budget next year after investing about $400 million across roughly 140 AI use cases that generate an estimated $800 million in benefits. Its Erica assistant has handled 3.6 billion transactions and reportedly avoids the need for 11,000 additional call-center staff, while the bank emphasizes human oversight, privacy, bias and model-risk controls. S&P Global is expanding its Kensho-centered AI strategy, citing an unnamed tier-one bank deployment that reached production six times faster and improved accuracy to 98% from about 60%.

Analysis

The investable read-through is stronger for SPGI than BAC: regulated, source-linked data can be monetized as a higher-value workflow layer rather than merely used to remove internal labor. If client deployments move from pilots into embedded credit, research and compliance workflows, SPGI can capture incremental API, platform and seat revenue while lowering servicing cost; this supports both organic-growth durability and a premium multiple versus data vendors with less proprietary or auditable content. The key second-order risk is that frontier-model providers commoditize generic research synthesis, widening the valuation gap between owners of licensed primary data (SPGI, MCO, LSEG) and workflow vendors reliant on public or replicable datasets.

For BAC, the relevant question is not whether AI spending rises but whether it displaces enough controlled operations expense to offset implementation, model-governance and cyber costs. Near-term EPS sensitivity is likely modest because the highest-return deployments are typically narrow, assistive workflows with human review, while the largest labor pools require longer control-validation cycles. A successful rollout could nevertheless reinforce digital-service advantages over regional banks, whose fixed compliance and technology costs are spread over materially smaller revenue bases; this is a 6-18 month competitive issue, not a near-term earnings catalyst.

Consensus may overvalue autonomous-agent narratives and undervalue data provenance, permissions and audit trails. That favors SPGI's commercial positioning, but the cited productivity and accuracy outcomes are vendor-reported rather than independently audited; investors should require evidence in net new bookings, retention and revenue-per-client before underwriting material estimate revisions. For BAC, a doubling of a budget line without disclosed gross savings could initially be an expense-ratio headwind, especially if broader rate-driven NII pressure persists.

Catalysts over the next 1-3 months are SPGI's disclosures on AI-linked bookings, platform attach rates and margin trajectory, plus BAC's 2027 expense guidance and efficiency targets. Falsification for SPGI is decelerating subscription growth or margin dilution despite AI investment; falsification for BAC's efficiency thesis is rising noninterest expense without improved digital-service penetration, headcount leverage or operating-jaw expansion.

Key Decisions for Investors

• Accumulate SPGI on market-driven weakness rather than chase the news; target a 6-12 month long with 12-15% upside if AI-enabled data/platform products sustain organic-growth and margin expectations, versus roughly 7-8% downside if bookings evidence fails to emerge. Reassess after the next earnings call if management does not quantify platform or AI-related commercial traction.
• Express the data-provenance thesis as long SPGI / short a basket of lower-moat information-workflow exposure rather than a broad AI long; a practical watch pair is SPGI versus FDS, sized beta-neutral, over 6-12 months. Exit if FactSet demonstrates comparable AI-driven net-sales acceleration or if SPGI's Market Intelligence margin fails to improve.
• Do not add BAC solely on this development. Maintain BAC exposure only where supported by the separate rates/credit thesis; use the next two reporting periods as a monitor for positive operating jaws and quantified automation savings before treating AI as an EPS catalyst.
• Monitor MCO, LSEG and ICE for corroborating pricing and usage signals. Broad evidence that licensed financial datasets are receiving incremental AI monetization would validate a sector allocation to financial-information owners; evidence of customers shifting to cheaper model-native alternatives would argue against multiple expansion.