Departments count revenue differently. Whose number will the AI use?

#Sales reports 12%, finance reports 8%

At the executive meeting, the sales division reports that revenue grew 12%, and the finance team says 8%. Neither is wrong. Sales counted orders booked, finance counted tax invoices issued net of returns, and the revenue in the executive report is yet another number, with intercompany transactions taken out as well. Active customers, new accounts, and utilization work the same way. The word is the same; the calculation differs a little from one department to the next.

In a company like this, before you ask an AI about revenue, you have to settle what "our revenue" means. Otherwise the AI comes back with a third number that matches no department. In OwlAI, that definition goes into the organization glossary, down to the formula and where the data lives, and the AI counts by that formula no matter who asks.

#Without the definition, the AI produces a plausible third revenue figure

Ask "What was revenue this month?" and an AI that doesn't know the definition finds a database column that looks like revenue and adds it up. It has no way to know whether returns should come out or intercompany transactions should be excluded. The number it produces matches no department's figure.

The trouble is that this number doesn't look wrong. The table is clean and the query conditions look reasonable. Only someone who knows the departments define revenue differently senses that something is off.

#The glossary ties company terms to the data

Each entry in the organization glossary holds the term, its definition, synonyms and abbreviations, a formula, and the database, table, and columns the term refers to. The formula can be a sentence, such as "sum of tax invoice amounts issued, net of returns," or, for a data owner, a query fragment.

When a question comes in, the AI first looks up its terms in the glossary. Ask with "revenue," "net sales," or "NS," and you hit the same entry. The AI reads the definition and formula before it writes the query, and when it chooses which tables to look at, it starts with the tables linked to the term. Instead of guessing at a column that looks like revenue and adding it up, it calculates the way the company has defined it.

The glossary isn't limited to data questions. It is the same glossary that corrects misheard names in meeting transcripts, and the AI follows the same definitions when it writes reports or meeting minutes. So "revenue" said in chat and "revenue" written in the minutes mean the same thing.

#Define it once, and every question is counted the same way

With the definition in the glossary, nobody has to explain the basis each time they ask. Whether a sales rep or a finance analyst asks, the AI reads the same entry and calculates the same way. If departments really do need different bases, split the term into "booked revenue" and "invoiced revenue" and define each. When a question could mean either, the AI says which basis it is counting on and then answers.

Next to the glossary sit table notes. When a fact turns up while someone is working with data in a conversation, such as "amounts in this table are in thousands of won before 2023," the AI saves it as a note on that table. The two records change in different ways. A note is a hint gathered in the field, so the AI rewrites it when it turns out to be wrong. A glossary entry is a definition the organization has approved, so it changes only with human approval.

#People set the definitions, and the AI finds the missing terms

Administrators manage the organization glossary. We recommend starting with the core metrics used in executive reporting, such as revenue and customer count. The people who know the data best set each definition, and an administrator registers it.

The rest fills in with use. When someone says in a conversation, "Our revenue is invoiced, net of returns," or an in-house term that isn't in the glossary keeps coming up, the AI proposes the term or definition as a candidate for that person's personal glossary. Because the candidate comes from their own conversation, only they can see it; it goes into the glossary only when they approve it, and it is not used in answers before then. Candidates for the organization glossary need an administrator's approval. The authority to define terms stays with people; the AI takes on the work of finding the ones that are missing.

Before you ask the AI about revenue, settle three things inside your company:

• The list of core metrics used in executive reporting (revenue, customer count)
• The data owner who sets the definition of each metric
• The terms whose basis differs by department, and the names you'll use to split them (booked revenue, invoiced revenue)