AI is getting cheaper every year.
Yet companies are spending more on it than ever.

This is the paradox many finance leaders are now facing.

Research from Epoch AI shows LLM inference costs have declined roughly tenfold annually since 2021 [1], while total enterprise spending continues to rise. With global AI investment expected to exceed $3.3 trillion by 2027 (Gartner) [2], Finance leaders running cloud platforms, SaaS products, or AI-powered services now face a distinct form of planning complexity. Cost behaviour, revenue patterns, and margin dynamics are all shifting in ways that traditional budgeting frameworks were not designed to accommodate. This article examines how forward-looking finance organisations are adapting their approach and what practical steps they can take.

The Cost Behaviour Paradox

Traditional cost forecasting assumed stable relationships between costs, inputs, and outputs. Cloud and AI have inverted this logic. Unit costs decline continuously as AI infrastructure matures. Yet total spending rises as consumption scales faster than efficiency gains. This creates two distinct planning challenges. Transaction-level economics improve quarterly. Portfolio-level consumption compounds, producing net cost growth unrelated to traditional inflation.

Supplier negotiation alone cannot resolve this dynamic. Finance must understand how operational decisions drive infrastructure consumption: from the architecture chosen, to how features are designed, to how customers are onboarded. Each of these decisions carries a cost consequence that compounds over time. Gaining that visibility requires finance to engage with engineering and product teams earlier, before plans are finalised and spending trajectories are set. The teams that manage this well do not wait for budget variances to surface. They are present when consumption decisions are being made. 

Revenue in Layered Pricing Structures  

Pricing has changed with costs. Pure subscriptions provided predictability but did not capture value from heavy users. At the same time, pure usage models aligned revenue with consumption but created cost volatility. Most organisations now use hybrid structures that combine base subscriptions with usage allowances and extra fees.  

Subscription-based pricing simplifies baseline forecasting but obscures the effects of elasticity. Usage components offer demand signals but increase timing mismatches between costs and revenue. Hybrid structures require finance to track multiple variables simultaneously, from base subscription uptake to overage rates, without losing sight of overall margin.

Revenue depends on customer behaviour patterns that change faster than contract cycles. Finance must model not just what customers buy, but also how they use it, how usage leads to charges, and how both aspects respond to product changes and competition.  

The strongest pricing structures synchronise three variables: consumption patterns, infrastructure costs, and the value customers actually realise. When these are reviewed regularly, pricing adjusts to shifts in consumption rather than trailing by quarters. 

Margin as Governance Discipline  

Margin outcomes largely come from three variables that change at different rates. Operational usage patterns change constantly. Infrastructure costs adjust every quarter. Pricing structures are reviewed once a year or even less frequently. This timing gap creates exposure windows before traditional planning cycles catch up.  

Leading finance organisations view margin as a governance discipline rather than just a forecasting tool. Instead of predicting a single outcome, they keep scenario ranges that reflect different usage paths and efficiency assumptions. These are readiness tools, not theoretical models. The leadership team already understands the implications and has evaluated response options in advance. 

This approach requires connecting usage signals, cost-to-serve data, and pricing performance into a regular review cycle. Finance teams managing significant cloud and AI expenses hold regular reviews, often monthly, where they check usage trends, cost behaviours, and pricing effectiveness with teams from finance, engineering, product, and commercial areas.  

From Usage Signals to Predictive Intelligence  

A growing number of finance teams now treat usage metrics as financial inputs rather than just operational indicators. Machine Learning (ML) models can translate operational usage signals directly into financial forecasts. By combining usage trends with time-series analysis, finance teams gain earlier visibility into potential margin pressure.

This enables a critical transition. Finance shifts from a reporting function to an active orchestrator of forward-looking decisions. Instead of reactive variance analysis, teams can initiate discussions about resource optimisation, pricing adjustments, and capacity planning while options remain available.

A similar approach, documented in an Association for Financial Professionals case study [3], showed how a fintech identified margin pressure weeks before it was reflected in financial statements. Heavy users consuming more resources than pricing structures anticipated created silent margin erosion. Early detection through usage monitoring enabled coordinated response across product, engineering, and commercial teams before the issue became a crisis. This approach is now standard practice at leading SaaS and cloud infrastructure companies. 

AI Augmentation in FP&A

ML models turn operational data into financial forecasts and scenarios that previously required extensive manual work. Finance teams can now evaluate a wider range of trajectories more frequently and with less analytical overhead.

Agentic AI is moving from theory into the daily reality of finance teams. When it operates inside clear governance boundaries, it can quietly take on a large share of routine spend monitoring and intervention work. These systems track how resources are actually consumed, flag patterns that don’t look right, and surface pragmatic next best actions. In more mature deployments, they also execute tightly defined, pre-approved decisions so humans are not forced to sign off on every small adjustment.

The most advanced use cases today are in cloud infrastructure, where the link between usage and cost is both direct and measurable. Here, finance is shifting from retrospective cost policing to real-time stewardship of unit economics. Yet the line of accountability has not moved: finance still sets the boundaries within which autonomous actions can occur, and any decision that could materially move the numbers is escalated back to humans.

For FP&A leaders, the real shift is in how the function operates day to day. Teams with these capabilities don’t just look at results after the fact; they review usage data alongside financials, sit down with engineering and product to discuss cost trade-offs, and bring AI-generated scenarios into their regular planning rhythm rather than treating them as occasional side projects.

Three Priority Actions for Finance Leaders

Foundation: Build Usage-Cost-Revenue Data Architecture

Build a unified data architecture that connects operational usage telemetry, cost attribution, and revenue recognition in a single source of truth. Without this foundation, finance operates without visibility into the variables that drive margin outcomes in cloud and AI businesses. 

Many finance teams still see infrastructure costs only at an aggregate level in monthly reports. By the time unusual consumption patterns become visible, the underlying operational decisions have already been made.

This is not optional infrastructure. It is the prerequisite for every planning discipline that follows.

Intelligence: Implement Forecasting and Scenario Frameworks

Machine learning models that translate usage patterns into financial outcomes, combined with pre-evaluated scenario ranges, preserve decision readiness as conditions shift. The goal is not prediction precision but response speed. Maintain scenarios reflecting different trajectories. When actuals track toward one scenario, you already understand the implications and have evaluated the options. Over time, the FP&A process itself improves as new usage data continuously updates planning assumptions.

Governance: Align Pricing with Cost-to-Serve Economics

Establish regular reviews that bring together three things: how intensely customers are using the product, how infrastructure costs are moving, and how pricing is structured. Treat pricing as a financial control, not just a commercial lever you adjust once a year. When prices trail cost behaviour by several quarters, margin risk builds quietly in the background. A simple, recurring governance cadence helps catch this early and protect margins as the business scales.

Looking Forward

Cloud and AI are forcing FP&A teams to rethink how planning really works. Forecast accuracy on its own is no longer enough. What matters more is the ability to respond quickly and confidently when usage, costs, or customer behaviour shift.

Finance leaders who have already made this transition often say the biggest change was not in their models, but in how they see the role of FP&A. Instead of only predicting outcomes, finance now helps the organisation spot signals earlier and shape decisions before issues become visible in the P&L.

This is not a theory. Variants of these approaches are already in use at SaaS and cloud infrastructure companies, where usage-driven economics have made traditional planning cycles too slow. The organisations moving fastest are the ones that have stopped optimising for perfect forecasts and started focusing on decision readiness — being prepared to act when conditions change.

Useful Links & Acknowledgements

1. Epoch reports LLM Inference Price Trends 

2. Gartner - Worldwide AI Spending

3. AFP - Case Study Fintech

4. FP&A Trends - Five Visionary Trends Redefining FP&A

5. Managing Volatility in the Age of Autonomy