Any analysis now takes minutes. A metric, a chart, a full report on demand: work that used to occupy an analyst for a week is a prompt away. For most of the history of management reporting, production was the bottleneck. Leaders waited for numbers, and the wait shaped how companies worked. That constraint is gone, and the scarce asset has flipped. Production used to be the bottleneck. Verification is now.
The question that became the whole job
When any number can be produced instantly, one question carries the weight: where did this number come from, and who agreed on its definition?
Margin can be calculated before or after freight, rebates and returns. On-time delivery can be measured against the date the customer requested or the date the company promised. An active customer can be one who bought in the last 3 months or the last 12. Each version is defensible, and each produces a different chart. A tool that writes fluent analysis will pick one without telling you and present the result with the same confidence either way.
That is the new risk. A confident chart built on an unagreed definition is now the fastest way to make a bad decision look rigorous.
Slow reporting was an accidental control
Slow reporting used to protect companies that skipped definitions. Building a report took weeks, and during those weeks someone caught the gaps. The analyst asked finance which revenue figure to use. The controller noticed that two entities booked freight differently. A plant manager objected to how scrap was counted.
Nobody designed that as a control. It was friction, and friction did the job. Remove the friction and the protection goes with it, unless the definition work is done on purpose.
Example
What a polished report can hide
A public case arrived in 2025. Australia’s Department of Employment and Workplace Relations commissioned a 237-page review from Deloitte’s Australian firm, at a reported cost of about A$440,000. After publication, Chris Rudge, a University of Sydney researcher, noticed that the report cited academic papers that do not exist and quoted a federal court judgment in words the judge never wrote. The revised version disclosed that a generative AI tool had been used in preparing it, and Deloitte agreed to refund the final installment of its fee. The firm said the corrections did not affect the report’s findings and recommendations.
The lesson is not about one firm. The document looked finished. Its length, format and tone signaled rigor, and it passed through a professional delivery process and a government client before one reader checked the references against their sources. Production had never been easier. Verification was left to chance, and it happened outside the process.
Do the definition work before the machine starts
The answer is not less AI. It is doing the work AI cannot do for you, before it starts producing. In practice that means three things.
- Agree the definitions first. For each number leadership decides with, write down the calculation, the source and the owner, and have finance and operations sign the same page.
- Make every number traceable. A figure in a review should lead back to its source and its definition in one step. If it cannot, it is an opinion with a chart.
- Give verification an owner. Someone is accountable for confirming that a new analysis uses the agreed definitions before it reaches a decision meeting.
None of this slows the machine down. It gives the machine something agreed to produce. Companies that do this work get the full benefit of cheap analysis: more questions answered, faster, on numbers that hold up when challenged. Companies that skip it get more reports and less certainty than they had before.
Sources: Fortune, October 7, 2025.