For CFOs

Why does our AI give a different revenue number than finance?

Because the AI and finance are answering two different questions that share one word. Revenue before or after refunds. Booked or recognised. This month in which time zone. Nobody wrote the choice down once, so every tool made its own.

The usual fix is called a semantic layer: one shared dictionary of what each metric means, which every report and every AI assistant reads from. It is a good idea. It is also one more thing to build and run, and what decides whether it works is not the software. It is who is allowed to say what revenue means.

I keep seeing companies buy the dictionary before they agree on the words. Who signs off the definition of revenue where you work?

See the workshop on what to stop paying for

A working session on the data spend you cannot explain. It can be credited toward the Audit.

Have three answers for one number? Price the checking.

The hours spent reconciling figures are the first cost. About five minutes, no sign-up.

What to look at first

01

Put the answers side by side

Pick the five numbers your board sees: revenue, margin, churn, customer count, cost to win a customer. Ask finance, marketing and the AI assistant for each. Write the answers in one table. The gaps are the work.

02

One owner per number

One person who can say what it means and say no to a second version. Often in finance, sometimes not. Without a name, every fix lasts until the next dashboard gets built.

03

Price the dictionary before you buy it

A semantic layer needs people to build it and keep it current, every year. Ask what it costs to run, which reports it replaces, and which of the three answers it ends. If it replaces nothing, it adds cost.

The longer version: One number, one meaning