AI adoption is stalling. Finance needs to price that in
Investors bet on a rapid surge in enterprise AI use. The latest federal data says otherwise.
As of late November, the employment-weighted share of Americans using AI at work sits around 11%, down roughly one percentage point. The pullback is sharpest at large firms with 250+ employees-exactly where budgets and scale were supposed to fuel lift.
Three years into the generative-AI wave, demand inside businesses looks softer than the narrative. That has real consequences for forecasts, budgets, and valuations.
What the survey actually measures
The U.S. Census Bureau asks firms if they used AI "in producing goods and services" in the prior two weeks. That high-frequency pulse now points to flat or falling usage, especially among big enterprises.
It's a narrow lens-short window, self-reported-but it tracks real activity on the shop floor and in back-office workflows. Directionally, it's a useful brake on overenthusiastic models. Source: Business Trends and Outlook Survey.
Why big companies are hesitating
- Compliance and model risk: auditability, explainability, PII handling, and retention rules raise the bar.
- Quality variance: hallucinations and edge cases make "fully autonomous" promises hard to trust in production.
- Integration tax: connecting models to systems of record, permissions, and data contracts is slower than demos suggest.
- Fuzzy ROI: pilots save minutes, not hours; scaled benefits require workflow redesign, not just a chatbot.
- Cost control: inference bills can spike unpredictably; unit economics still wobble outside narrow tasks.
- Data readiness: fragmented, unclean, or gated data caps accuracy and adoption.
- Change friction: incentives, training, and process ownership lag the tech.
Implications for your models and portfolio
- Push the productivity uplift to the right. Treat broad margin expansion as a 2026-2027 story, not this quarter's.
- Expect a slower revenue ramp for "AI-native" apps; pipeline ≠ production. Pricing power likely compresses.
- Infrastructure spend stays resilient, but application layer winners will be those that own a workflow, not a feature.
- Budget mix tilts to small pilots, guardrails, and data work-less to big bang rollouts.
- More M&A and vendor consolidation as buyers standardize on fewer platforms.
What to watch next (leading indicators)
- Share of core workflows with AI in production vs. sandbox.
- Gross margin lift tied to AI line items in MD&A or earnings calls.
- Revenue per employee and ticket resolution times where AI is deployed.
- Unit cost per task (inference + orchestration) vs. human baseline.
- Procurement cycle length for AI deals and security approvals.
- Vendor indemnity terms and IP clauses-tighter terms signal maturing risk views.
Action plan for finance leaders
- Run a 90-day audit: where AI is actually used, who owns it, spend by vendor, and measured outcomes.
- Pick 3 workflows with measurable payback (FP&A variance analysis, invoice processing, KYC refresh). Define pre/post metrics.
- Cap TCO per task. Set a gate: no scale-up unless ROI clears a threshold (e.g., 30% time saved or higher quality at lower cost).
- Mandate model-risk controls: data lineage, prompt/response logging, red-team tests, human-in-the-loop for high-impact actions.
- Fund the plumbing: data quality, permissions, and integration work. That's where most of the return hides.
- Align incentives: managers get credit for adoption and measured outcomes, not for piloting the newest tool.
- Negotiate usage floors and price protections; avoid lock-ins until volumes stabilize.
- Refresh capitalization policy for AI spend (software vs. services) and disclosure of AI-related savings.
If you need a fast scan of practical tools to test in finance workflows, see this curated list: AI tools for Finance. For role-based upskilling across teams, browse courses by job.
Bottom line
The market priced an S-curve. Operations are giving us a plateau. Treat AI as targeted process improvement for now, not a blanket productivity boom.
Price patience, fund the boring plumbing, and demand proof of value before scaling. The winners will be the ones who measure.
Further reading: the Stanford AI Index tracks adoption, spend, and performance trends across sectors.
Your membership also unlocks:
