AI in Finance: Stop Chasing Cost Cuts and Follow the Revenue Signal

Boards commonly see finance AI business cases framed around cost. Automate reconciliations, shorten the close, reduce manual journal entries, and the case largely writes itself. The savings are real, but for many enterprises they are smaller than the value available in the decisions finance governs. For AI in finance function portfolios, those decisions include what price is accepted, which customer gets credit and on what terms, how much cash is tied up in inventory and receivables, and where margin leaks between quote and invoice. A finance organization that automates its own processing while leaving those decisions untouched may report efficiency gains and still miss the earnings effect.

The efficiency case is real, which is why it crowds out everything else

There is credible evidence that AI helps accounting work. Research from Stanford Graduate School of Business found that accountants who use AI finalize monthly statements 7.5 days faster than those who use traditional methods and spend 8.5% less time on routine back-office processing. The Stanford AI Index 2026 points to a broader pattern: productivity gains from AI are largest in structured, measurable work where outputs are easy to monitor.

The close, reconciliations, payables matching and intercompany eliminations are exactly that kind of work. Outputs are discrete, errors are visible and the baseline already sits in the ledger. That makes these use cases easy to approve, measure and defend in a steering committee. It also makes them a budget magnet.

The trap lies in what the measurement captures. A faster close pays off mainly when someone acts on the earlier numbers, whether by repricing a product line, tightening terms for a deteriorating account or releasing excess safety stock. If decision cadence does not change, a week saved in the close often becomes capacity absorbed elsewhere. It may not show up as lower cost and is unlikely to raise earnings on its own. The lean principle discussed in WGA’s analysis of operational excellence in the agent era applies here: automating a process that feeds no better decision only makes the waste arrive faster.

What the evidence says about revenue and adoption

Revenue effects are common but modest per function

The Stanford AI Index 2025 draws on a McKinsey survey and reports that 71% of respondents using AI in marketing and sales report revenue gains, as do 63% in supply chain management and 57% in service operations, yet the most common level of revenue increase is less than 5%.

Two readings follow. First, revenue effects appear most often in the functions closest to customers and supply, domains where finance often sets or shapes the rules on pricing authority, credit terms and inventory policy. Second, the gains are incremental rather than transformational. That argues for discipline, not dismissal. On a large revenue base, a low single-digit improvement in price realization, or a modest release of working capital, can outweigh the labor savings from automating accounting tasks. The CFO is well placed to make that comparison honestly, because finance sees both sides of the ledger.

Adoption is early, so allocation habits are forming now

Broad business adoption remains limited. US Census Bureau survey data show overall AI usage hovered between 17% and 20% from December 2025 to May 2026. The Federal Reserve Board, citing Census data, reported that about 18 percent of firms had adopted AI as of year-end 2025. More advanced deployment is rarer still: the AI Index 2026 found that AI agent deployment was in the single digits across nearly all business functions.

For finance leaders, the issue is timing. Portfolio habits set in the next few budget cycles are likely to shape where finance AI spending concentrates for years. If the first wave goes entirely to back-office automation, the governance, data and talent that decision use cases need may not get built. For a grounded view of where agents fit, see what an AI agent workforce actually is.

Four decision domains where finance holds the lever

The finance AI use cases that move earnings share a profile. Each involves a recurring, high-volume decision with a measurable economic outcome, and finance either owns the decision or sets its rules.

Pricing and discount governance

Where discounts flow through layered approval authorities designed for a world of spreadsheets, AI can score each deal against comparable transactions, flag discounts outside a defensible range and estimate the margin cost of each approval. The decision stays with commercial leaders, while finance owns the guardrails and the price realization metric.

Credit and collections

In many organizations, credit limits and payment terms are set at onboarding and revisited infrequently. Continuously updated risk signals let finance tighten exposure earlier and extend terms selectively where that wins volume at acceptable risk. Prioritizing collections by expected recovery rather than by ageing bucket changes where scarce collector time goes.

Working capital

When inventory policy, payables timing and receivables terms sit with different owners who have different incentives, AI-assisted forecasting and scenario analysis can show the cash cost of those tradeoffs in one view. At that point, cash discipline becomes an operating instrument rather than a quarterly review item.

Margin leakage

Leakage can hide in several places:

  • rebates that are unclaimed or overpaid
  • freight charges that are not recovered
  • contract terms that are never invoiced
  • price changes that do not reach the order system

These are pattern-detection problems across large transaction sets, and each finding carries a cash value that can be verified.

Governance changes once AI touches decisions

Moving AI from processing to decisions changes the risk profile. A misclassified invoice is an error. A mispriced contract or a misjudged credit limit is an exposure. Model risk discipline therefore matters more, even outside banking.

Banking supervision offers a useful reference point. Under the Federal Reserve’s SR 26-2, non-compliance with the revised model risk management guidance will not result in supervisory criticism against a banking organization. The agencies also note that the revised guidance discusses considerations specific to vendor and other third-party products, including validation of these products. Corporate finance teams often encounter AI embedded in planning, treasury and order-to-cash software. For them, the takeaway is practical: a vendor model is still a model, and someone must own its validation.

Credit use cases add a regulatory dimension. According to the EU AI Act text published on a non-official mirror maintained by the Future of Life Institute, AI systems intended to evaluate the creditworthiness of natural persons or establish their credit score are classified as high-risk, with an exception for systems used to detect financial fraud. Enterprises extending credit to individuals in the EU should confirm the official text and have legal counsel settle where business-to-business credit models fall.

Strategic and operational implications for enterprise leaders

  • Operating model. Decision use cases need a named business owner for the outcome, not only a systems owner for the tool. Pricing AI without commercial ownership risks becoming a report; credit AI without a credit committee mandate risks becoming advisory noise.
  • Workforce. Finance roles shift from processing transactions to reviewing exceptions and challenging model recommendations. That calls for judgment training and clear escalation rights, not just tool training.
  • Cost. Integration with ERP, CRM and billing data, plus monitoring and validation, can outweigh model costs (see WGA’s analysis of the economics of an AI agent workforce).
  • Data. Decision use cases depend on transaction-level data that runs from quote to cash and crosses functional boundaries. Ownership and quality rules for that data, defined per use, are prerequisites.
  • Measurement. Revenue and margin effects need counterfactuals, such as holdout groups, controlled comparisons or reconciled cash findings. Every figure should be labeled as measured or projected.
  • Board oversight. Directors can ask three questions. What share of the finance AI portfolio targets decisions rather than processing? Who owns each outcome? How are gains verified?

A practical way to redirect the finance AI budget

  1. Inventory the portfolio. Classify every funded finance AI initiative as either processing efficiency or decision economics.
  2. Baseline the decisions. Use the ledger to establish current performance: realized versus list price, bad debt and days sales outstanding, inventory and payables days, and recovered versus unrecovered leakage.
  3. Size value at stake conservatively. Use modest ranges consistent with the evidence, and include full integration and governance cost.
  4. Re-weight funding. Let efficiency use cases self-fund where their savings can be verified. Direct incremental budget to one or two decision lighthouses, each with a business owner.
  5. Stage-gate on production results. Release further funding only when measured results clear agreed thresholds. Where a program has lost sponsor confidence, a structured AI strategy turnaround can separate what to keep from what to retire.

Conclusion

The CFO’s AI agenda does not need to abandon efficiency. It needs to stop treating efficiency as the destination. Faster closes and leaner processing create capacity and credibility. The earnings effect comes when that capacity is pointed at the decisions finance governs: price, credit, cash and margin. With adoption still early, the allocation choices made now will shape whether finance becomes a cheaper ledger or a sharper instrument of decision economics.

Turning these implications into action

Redirecting finance AI from processing to decision economics is mainly an economics and operating-model problem. Through AI value realization and economics, WGA Consulting builds value-at-stake models from the workflow up, with finance involved, which can be applied to pricing, credit, working capital and leakage. The work includes workflow baselining, full-cost business cases, target setting and value tracking, with every figure labeled as measured or projected. Value audits can test whether existing finance AI initiatives are delivering what their business cases assumed.

Where the question is portfolio shape and ownership, WGA’s enterprise AI strategy and operating model service covers portfolio design and lighthouse workflows, operating model and AI governance design, and a sequenced roadmap with stage-gated funding. For CFOs linking AI choices to corporate priorities, strategy consulting adds option framing, economic modeling and AI governance for board review.

Start a conversation about AI value realization and economics

Relevant WGA client experience

A $30 billion international holding company headquartered in Australia faced a problem familiar to diversified portfolios: each business accumulates costs and processes in its own way, and budgets built year on year can roll prior spending forward without asking whether it still earns its place. Operating across several competitive sectors, the client needed to release efficiency, raise shareholder value and fund sustainable growth.

WGA applied a blank-sheet, zero-based approach using WGA360, its proprietary zero-based transformation solution. The work combines an AI-powered diagnostic of hidden costs across functions and geographies, scenario modeling of operational and financial outcomes, and transformation spanning zero-based budgeting, workforce optimization, process automation and digital reinvention, supported by change enablement throughout. The same test of whether spending still creates value applies to finance AI budgets. Documented outcomes: “Deliver substantial cost savings and greater agility” and “Set a new benchmark for zero-based transformation in the Asia-Pacific region.” Read the case study.

Sources

  1. accountants who use AI finalize monthly statements 7.5 days faster than those who use traditional methods and spend 8.5% less time on routine back-office processing (gsb.stanford.edu)
  2. Stanford AI Index 2025 (hai.stanford.edu)

About the author

Director, WGA Consulting

With over 23 years of cross-industry expertise, Liv Turner is a leader in business transformation, strategy, and operational excellence. She is a trusted advisor known for her ability to deliver sustainable performance improvements and drive transformational change.