In the latest MIT study, they show that the majority of AI pilots do not generate economic impact; the clearest return comes from productivity increases in repetitive tasks and back-office. The main datum of the study is emphatic: approx. 95% of organizations do not register a measurable return on their investments in generative AI projects, which underlines the gap between spending and real value creation.
By Strathens Media | 27 August 2025 | Reading time: 3 min
Recent studies —academic syntheses linked to MIT and the State of AI in Business 2025 (MLQ.ai) report— converge on a practical and urgent idea: AI delivers its greatest value when it improves productivity in financial and operational processes that can be measured. Deploying models is not enough; it will be necessary to reduce man-hours in closings, automate reconciliations, speed up validations and, above all, be able to quantify those savings through KPIs defined from the start.
Why do so many projects remain demos? The report points to three recurrent bottlenecks: quality and availability of data; technical integration with transactional systems and workflows; and organizational governance that allows measurement, iteration and scaling. In practice, the documented successes do not come from the most sophisticated model, but from its ability to integrate into operations with clean data and standardized processes.
Sector evidence provides context: while one recent study calculates that 95% of pilots show no measurable return, other analyses (for example, prior consultancy studies) point to modest returns when projects are applied at scale without adequate infrastructure; for example, reports from recent years place average enterprise ROI in contained figures (orders of single-digit percentages in some samplings). This reinforces the idea that investing is not enough: you must invest with a focus on measurable cases and with a prepared data base.
A key component that enhances AI performance in finance and operations is a FP&A platform capable of delivering ready-to-use data in real time, in a shared environment where finance, controlling and operations interact transparently and aligned with the company’s short- and long-term objectives. By consolidating sources, normalizing histories, controlling versions and enabling automatic data pipelines, these platforms facilitate financial planning, accelerate analysis and automate reporting —all while reducing manual work and increasing the likelihood that an AI pilot moves into production with measurable impact.
In short, MIT and MLQ.ai refine AI’s promise: the most reliable return today is measurable operational productivity. Capturing that return requires combining models with solid data infrastructures and FP&A platforms that provide real-time data, shared processes and traceability. Those who achieve that combination will not only optimize costs, but will gain speed and quality in decision-making —the key competitive advantage in the era of applied intelligence.