AI Spend, Thin Returns: What Finance Leaders Need to Do Next
Many finance chiefs are staring at the same spreadsheet: rising AI spend, limited payback. A recent PwC survey shows more than half of CEOs say their companies aren't yet seeing a financial return from AI. Only 30% reported increased revenue over the last 12 months, and 56% said AI neither lifted revenue nor cut costs. Just 12% saw both.
That disconnect is fueling investor worries about an AI bubble while productivity gains stay out of reach. Meanwhile, executives fear falling behind if they slow down. As PwC's leadership put it, a small group is turning AI into measurable results while many are stuck in pilots - and that gap is widening.
Why ROI Is Stalling
- Weak foundations: no clear roadmap, fuzzy ownership, and immature model risk management.
- Use cases chosen for novelty, not unit economics or data readiness.
- Pilots never leave the sandbox due to change resistance and missing process redesign.
- Quality issues: hallucinations, brittle workflows, and poor retrieval erode trust and value.
- Security and compliance friction slows deployment and adds overhead.
- Costs outpace benefits: compute, vendor minimums, integration, and shadow projects pile up.
The CFO Playbook: Turn Hype Into Cash Flow
- Start with a financial question, not a model: what P&L line or balance sheet risk will move, by how much, and by when?
- Quantify unit economics: cost to serve, cycle time, error rates, and rework. Set a target delta upfront.
- Prioritize three use cases with clean data and short paths to production (90 days). Kill the rest for now.
- Run controlled pilots with a real baseline and A/B measurement. No vanity metrics.
- Design for human-in-the-loop and exception handling from day one; don't bolt it on later.
- Impose cost discipline: track GPU hours, inference cost per transaction, and vendor overage fees weekly.
- Negotiate contracts around outcomes: service credits tied to uptime, latency, and quality thresholds.
- Account correctly: capitalize what qualifies, expense the rest. Don't hide Opex in "innovation."
- Stand up model risk, data governance, and audit trails before scaling. Saves pain later.
- Publish a simple ROI scorecard monthly: business KPI lift, quality metrics, and fully loaded cost.
Where AI Is Actually Paying Off (Now)
- Collections and dunning automation: smarter outreach, better prioritization, tighter cash conversion.
- Fraud/AML triage: higher signal-to-noise and faster investigator throughput.
- Customer service deflection: policy-grounded assistants that reduce handle time and refunds.
- FP&A scenario generation: faster variance analysis and narrative drafts with strict source citations.
- Procurement spend classification: cleaner categories that unlock renegotiations and compliance.
These work because they live on structured data, have clear controls, and map straight to cost or revenue. Keep the scope small, wire into existing systems, and measure the cash impact weekly.
Questions to Put on Your Next Steering Committee Agenda
- What precise KPI will this use case move, by how much, and how will we measure it?
- Is our data complete, permissioned, and governed for this purpose?
- What is the full cost per transaction (infra, licensing, integration, people)?
- What is our fallback process when the model is uncertain or wrong?
- Who owns run-state performance (not just the pilot) and what's the escalation path?
- What controls satisfy audit, privacy, and model risk requirements?
- What will we stop doing to fund this? Where does the headcount/time actually come from?
- What's the kill switch if ROI misses plan for two consecutive quarters?
The Market Context
The debate isn't about potential - it's about timing and discipline. Some firms are getting measurable results, but many are burning cash on experiments with no line of sight to scale. Until the basics are in place, more spend won't fix the returns problem.
Next Steps
- Pick one use case that hits cash in 90 days. Ship it, measure it, then scale.
- Stand up a weekly "AI P&L" review: benefits realized, costs incurred, risks surfaced, actions taken.
If you're mapping the first wave of practical tools, this curated set is a useful filter for finance teams: AI tools for Finance. Keep it boring, measurable, and tied to a KPI - that's where AI starts paying its way.
Your membership also unlocks:
