Why Clean Data Beats Better AI as the CFO’s Top Priority in 2026

MIT Sloan just published takeaways from its CFO Summit, and one line from Arm Holdings CFO Jason Child deserves more attention than it will get: “I find that getting that clean data layer is really hard.”

He said it plainly, without drama. And then he said the thing that matters: once you have that clean layer, “you can start to do the automation layer, you can do the machine learning; you can do a lot of different things.”

That sequence is correct; and almost universally ignored.

The article reports that 60% to 80% of time on a data analytics project goes to acquiring and cleaning data. That number has been roughly the same for a decade. It has not improved because the industry keeps optimizing the wrong end of the stack. Better models. Better agents. Better interfaces. Meanwhile, the foundation those tools depend on is still built by hand, project by project, team by team.

Child also made a point about LLMs that finance leaders need to internalize: “LLMs are probabilistic. Finance is deterministic.” An LLM pointed at unresolved data does not produce a wrong answer because it is a bad model. It produces a wrong answer because the data gives it nothing trustworthy to reason about. The model is not the constraint. The material it reasons with is the constraint.

This is the part the AI conversation keeps skipping. Model capability is commoditizing. Every company has access to the same reasoning power. The differentiator is not intelligence; it is what you point the intelligence at. The company that gives AI access to clean, harmonized, contextualized operational data gets answers no competitor can replicate; because no competitor has that data.

The CFOs at this summit are learning the same lesson from different angles: start with the data. Not because it is a prerequisite you check off before the real work begins. Because it is the real work. The resolved data layer is not infrastructure supporting AI. It is the primary artifact. Everything else is a consumer of it.

The bottleneck was never intelligence. It was producing material worth reasoning with.

Source: MIT Sloan Ideas Made to Matter, “4 takeaways for finance teams as they implement AI” (Feb 2026)