AI Readiness
You can't put an AI on numbers your own departments don't agree on.
Before anyone connects Copilot, Claude, or an AI agent to your operations data, the data has to agree with itself. Here is what ready looks like, and a ten-minute way to find out how far off you are.
The Situation
The board asked. Now what?
AI assistants can now query your data directly. Copilot sits inside Power BI. Assistants like Claude connect to a semantic model through a connector (an MCP server, for your IT team) and answer questions in plain English. That is real, it is inexpensive, and it is already on someone's desk at your company.
What decides whether it helps or hurts is not the model you pick. It is the data underneath. An assistant inherits exactly what your data means today: every undocumented field, every metric two departments define differently, every spreadsheet that quietly patches two systems together.
The work that makes data AI-ready is the same work that makes Finance and Operations agree on a number. We were doing it before it had this name.
Ungoverned Data
What an AI does with numbers nobody agreed on
It picks a definition and does not tell you
Finance books shrink as a P&L variance. Operations counts waste in units. Ask an AI for "loss and waste" and it chooses one, answers confidently, and never mentions the other exists.
It reads the field name, not the meaning
A column called Adj with no description is a guess. A measure with no logic documented is a guess. The assistant fills the gap with something plausible, which is the worst kind of wrong.
It answers fast, and leadership believes it
The number arrives in seconds, formatted, with a sentence of explanation. Nobody checks it against the spreadsheet, because the point of the AI was to stop checking spreadsheets.
What Ready Looks Like
Four things an AI depends on. None of them are AI.
This is the foundation we build in every engagement. The assistant is the last thing to plug in, and the easiest.
01
Agreed definitions
One written definition per metric that Finance, Operations, and IT signed. If two departments would give two numbers, the AI inherits the argument.
02
One source per metric
Each number traces to one system, one table, one calculation. Not a chain of exports. The assistant queries the source, not a copy of a copy.
03
Documented fields
Every column and measure carries a plain-English description: what it is, where it comes from, how it is calculated, how to use it. People read it as a tooltip. AI assistants read it before they answer. This is the piece almost every model is missing, and the cheapest to fix.
04
Governed access
Data classified into tiers, access granted through groups. An assistant answering a plant supervisor should not be able to see payroll, and with governance in place it cannot.
Documented Fields
The standard we apply to every field. Yours to use.
Every column and measure in a model we build carries one description in one format. It reads as a tooltip for the person hovering over a chart, and it is what an AI assistant reads before it decides what a field means. Five sections, pipe separated, so people and machines parse it the same way.
[Plain-English summary]. | Source: [origin system]. | Logic: [how it is calculated]. | Report alias: [display name, if different]. | Note: [how to use it correctly].
Of the recipients who opened the message, the share who clicked something. | Logic: DIVIDE([Total Clicks], [Total Opens]). Returns blank when opens are zero. | Note: CTOR = Click-To-Open Rate. Compare within channel and audience type.
The rules matter as much as the format: the note never carries a number that will age, every abbreviation is decoded where it appears, and a field that displays under a different name in reports says so. The full standard, with worked examples, is free. There is nothing secret in it. It is just work most teams have not done.
Find Out Where You Stand
Ten questions. Ten minutes.
Yes-or-no questions a CFO or COO can answer without calling IT. Count the yes answers and the scoring guide tells you whether to move, prepare, or start with the foundation. Three of the ten:
- Can you pull your top ten SKUs by margin in under five minutes without calling IT?
- Do Finance and Operations use the same number when they talk about production performance?
- If you opened your reporting model, would every field carry a description of what it is and how it is calculated?
Scored Low?
The AI Readiness Assessment
Two weeks and $5,000, fixed. We score your data on the four pillars above, trace where each core metric actually comes from, and hand you a written fix list in priority order. Credited in full against a Foundation Build signed within 90 days.
If the honest answer is that you are further off than an AI project can absorb, we say so, and you keep the report.
You talk to Mitch.
Get the data ready before the AI arrives.
A 30-minute discovery call. Bring the question the board asked you; we will tell you what has to be true about your data before an AI can answer it, and how far off you are.