You have the score. You have the gap list. You have a defensible plan for the next two quarters and a specific ask attached to it.
You also have ten minutes on a calendar, and the last person who used those ten minutes to explain column-level lineage lost the room by minute two.
The problem is rarely the argument. It's that the vocabulary which earns credibility with engineers spends credibility with executives, and most technical champions discover this while it's happening.
Translating the vocabulary
A CDO is accountable for a small number of things: whether the numbers the business runs on can be defended, whether AI initiatives ship or stall, whether an audit becomes a project, and whether the budget she defended last year produced something.
Everything you want to say maps onto one of those. It just has to be said in those terms.
What you would tell an engineerWhat she needs to hearWe only have column-level lineage on 61% of the estateFor 39% of our reporting, we can name the system a number came from, and no more than thatOur lineage is inferred from run logsOur provenance answers are best guesses that hold until someone checks themDefinitions have no version historyWe cannot show what a metric meant last year, which is what an auditor asks firstDefinitions are re-implemented in each BI toolThe dashboard and the AI assistant answer the same question differently, by designWe have no impact analysis before schema changesWe find out what a change broke by waiting for someone to complainCoverage went from 61% to 84%The share of our reporting we can defend on request went up by a third
Read the right column aloud before the meeting. If a sentence contains a technology, it goes back to the left column.
Slide one: what we can answer today
One slide, three rows, no architecture.
Pick the three questions her stakeholders will actually ask this year. Something close to: where did this figure in the board deck come from, can we ground an assistant on our own reporting data, and can we show an auditor how a regulated number was produced.
For each, give today's honest answer and the time it takes.
Where did this board figure come from? Answerable in about a week, by two engineers, for one metric at a time.
Can we ground an assistant on our reporting data? Partially. It will answer confidently across 61% of the estate and guess across the rest, with no way to tell which.
Can we show an auditor how a regulated number was produced? By reconstruction. Six weeks, and the result cannot be reproduced next quarter by anyone else.
That slide establishes the exposure without naming a single tool. It also has to be honest, because she will test one of these three claims with someone else within a week.
Slide two: the number, the gap, and the ask
Four elements. Resist adding a fifth.
The score. Metadata completeness across our three most-reported metrics is 9 out of 20 today. State the denominator every time.
The two gaps that move it. Name specific systems. "The billing extract and the legacy claims database have no harvested metadata" beats "we have coverage gaps in legacy systems."
The target and the timeframe. 14 out of 20 by end of Q2, with those two systems covered and definitions under version history.
One ask. Headcount, a budget line for a source connection, or a decision from governance about who owns definitional authority. One. A slide with three asks gets none of them.
What to leave out
The five stages. The five dimensions. The word MetadataOps, unless she uses it first.
That last one is counterintuitive and it matters. Introducing a term she has never heard costs you a minute of explanation and invites the question of whether this is a real discipline or something you read about. Lead with her exposure and your number. If she asks what to call this, then name it, and by then the term is arriving as an answer to her question.
Also leave out any comparison to what the organization already bought. If the catalog was her decision, criticising it makes the conversation about that decision.
The three questions coming back
"Isn't this what our catalog is for?" The catalog tells us what exists. This is about whether what exists is described deeply enough to answer questions with. Agree that the catalog was the right first step and position this as finishing it.
"What does it cost?" Price the first gap on its own. A number attached to a defined outcome gets approved, and a program budget gets deferred to planning.
"Is anyone else doing this?" Yes, and it's worth having one external reference ready. Practitioners have started naming the role, the analyst community has been pointing at metadata maturity as an AI prerequisite since the Magic Quadrant returned in late 2025, and the argument is public. Send her something written by someone other than you: an analyst note, a job posting from a company larger than yours, or a practitioner thread.
Before you book the meeting
Do the scoring first. All of the above depends on having a real number with a real denominator, and there is no version of this conversation that works from an estimate.
Then write the right column of that table out longhand and read it back. If any sentence would confuse a CFO, it will confuse her too, and you have ten minutes.
