Most financial reporting fails not because the numbers are wrong but because nobody reads
them. A twelve-page PDF emailed on the 15th of the month, dense with tables, lands in inboxes
and dies there. This week: how to build reporting that actually changes behavior—and how AI is
dissolving the wall between “the numbers” and the people who need them.

Start with the purpose. Financial reporting has three distinct audiences with three distinct needs.
Owners and executives need the strategic picture: are we on plan, where are we winning and
bleeding, and what needs a decision? Managers need operational drivers they can influence
this week: pipeline, utilization, job margins, inventory turns. Outsiders—banks, investors, the
IRS—need standardized statements. The classic mistake is producing only the third kind and
hoping it serves the first two.

The modern answer is a layered approach. At the base, clean books maintained continuously
(see last week). In the middle, a live dashboard with a small number of metrics—revenue vs.
plan, gross margin, cash runway, DSO, and two or three drivers specific to your business—
visible to the people who own them. On top, a short monthly narrative: what happened, why,
and what we’re doing about it.

AI is transforming each layer. The most visible change is natural-language analytics: instead
of waiting for an analyst to build a report, a manager can now ask their finance platform, in plain
English, “what drove the increase in cost of goods this quarter?” or “show me margin by
customer for the last six months” and get an answer with charts in seconds. Underneath,
automated variance narratives—generative AI drafting the “why” behind the numbers by
tracing a variance to its largest components—turn the monthly commentary from a half-day
writing task into an editing task.

Just as valuable is anomaly surfacing: rather than requiring someone to notice that software
spend crept up 30%, the system notices and tells you. Reporting shifts from pull to push—from
“here’s everything, find the story” to “here’s the story, and the data if you want it.”
Two disciplines keep this honest. First, define every metric once, in writing; the fastest way to
destroy trust in a dashboard is two versions of “gross margin.” Second, always pair AI-

generated narrative with human review—models occasionally explain a variance with confident
fluency and complete wrongness. The machine drafts; a human signs.

Good reporting is a decision-support product, and your colleagues are the customers. Build it
like one.

Next week: scenario planning—rehearsing the future before it happen