Finance Needs a New ROI Playbook for AI

Instead, AI should be managed as a portfolio of different bets, not as one uniform ROI challenge. That is because different AI initiatives create value in different ways—and on different timelines. A tool that automates invoice coding or account reconciliation may deliver relatively clear efficiency gains within months. A forecasting or decision-support model may create value by improving speed, accuracy, and business responsiveness before the financial benefit is fully visible. And a more ambitious initiative, such as AI-driven scenario generation or autonomous decisioning, may require longer time horizons and staged funding because the upside is real but uncertain.

A better starting point is to classify the use case before trying to assign a value to its benefits. Gartner experts advise breaking AI investments into three broad categories:

• defend, where AI improves existing operations;
• extend, where it enhances performance and supports measurable business outcomes; and
• up-end, where it aims to enable new ways of working or new business models.

Sorting AI spending into these three buckets forces finance to consider: Is this initiative mainly about efficiency? Is it about better business performance? Or is it a strategic bet with venture-style risk?

For treasury and finance teams, this approach means moving past the simplistic question of whether AI has "paid for itself" yet. The more useful question is whether a use case is performing as expected for its type and stage. In practice, that means assessing value on four dimensions.

1. Look at the nature of the outcome, not just the output. Too many AI programs are measured by activity—number of pilots launched, dashboards built, or models deployed. Those metrics may show momentum, but they do not show value. Better measures ask what business problem has changed: Has cycle time fallen? Has forecast accuracy improved? Has the close become faster? Has fraud risk or leakage declined? Has treasury gained earlier visibility into cash or working capital pressure?

2. Evaluate feasibility as rigorously as benefit. Many AI ideas look attractive in theory but disappoint in practice because of poor-quality data, fragmented systems, heavy customization requirements, or low trust from users. Gartner experts recommend scoring use cases not only for strategic relevance, but also for technical feasibility, internal readiness, and external readiness. In treasury and finance, that is especially important because high-value processes often involve sensitive data, control requirements, and stakeholders who may not readily trust black-box outputs.

3. Develop a clearer view of the full cost curve. One of the biggest mistakes in AI investment cases is underestimating ongoing cost. Up-front implementation may get the most attention, but the longer-term economics often depend on data preparation, integration, model maintenance, vendor contracts, licensing, token- or usage-based pricing, training, and adoption. There are three main recurring drivers of sustained AI spend: data management, vendor dependency, and pace of consumption. The question is not just what the tool costs to launch; it is what it will cost to run, scale, and govern over time.

4. Look beyond immediate financial return. Revenue growth, cost takeout, and cash flow improvement are the most common measures, for good reason, but many AI initiatives show their value first through nonfinancial gains: better decisions, stronger agility, faster adaptation, wider analytical reach, improved control, or a more strategic finance role. These benefits are not a substitute for financial value, but they are often the leading indicators of it. A treasury team that improves visibility into cash flow risk earlier in the cycle, or a finance team that cuts the time needed to generate planning scenarios, may not see the entire gain immediately in the P&L—but the enterprise is still benefiting.

Where in Treasury and Finance Is AI Proving Useful Today?

Across the wide swath of companies that have deployed AI within treasury and finance, the clearest near-term impact is still in high-volume, rules-based processes. Gartner identifies travel and expense management, SOX testing, account reconciliations, intelligent document processing, and automated invoice processing as among the more feasible use cases already delivering practical value. These applications reduce manual work, improve consistency and compliance, and free up finance staff for more analytical tasks.

The next wave of impact is emerging in decision support and predictive work. Cash flow forecasting stands out for treasury because it sits directly at the intersection of liquidity, working capital, and risk. AI can help analyze historical patterns, customer behavior, and external drivers to produce more dynamic forecasts of inflows and outflows. Credit risk analysis is another area with strong potential, using payment behavior, public data, and client signals to sharpen risk decisions.

The finance organizations that extract the most value from AI will not be those chasing a single breakthrough or insisting that every initiative meet the same short‑term ROI threshold. They will be the ones that evaluate AI through a portfolio lens, assessing each investment by its intended role, the effort required to operationalize it, and where value is expected to materialize first.