You have the mandate now. Somebody senior agreed that the metadata estate needs an owner, and the owner is you.
The mandate is also enormous and vague, which is the actual difficulty. Every system is in scope, every definition is arguably yours, and four teams have a list of things they would like fixed.
Most new metadata functions respond by starting a tool evaluation or an estate-wide glossary project. Ninety days later they have a vendor shortlist or a spreadsheet of terms, and no evidence that anything improved. The mandate gets quietly reabsorbed the following quarter.
Here is a way to spend the first quarter that ends with a number.
Treat the estate as a product with three users
The framing that makes the scope tractable: your metadata estate is a product, and it has exactly three kinds of user.
Reports need definitions that agree across tools. Agents need definitions that exist in retrievable, governed form. Auditors need definitions with a history and a provenance chain.
All three consume the same underlying supply. One supply, one product. That framing is the argument that keeps your scope whole when three teams each want their own version of it.
A product has a measurable state and a roadmap. Ninety days is enough to establish the first and draft the second.
Days 1 to 30: measure, and fix nothing
The strongest thing you can do in month one is resist repairing anything.
Take the three business metrics your manager assigned, or pick them yourself using one rule: each appears in at least two consuming systems and someone senior has asked about it this quarter. Trace each one backward from the report to the source, by hand, one hop at a time.
At every hop, record five things: whether harvested metadata exists, whether it goes to column level, whether the lineage came from parsed code or observed execution, how current the record is, and whether a named human owns the definition.
Budget an hour per metric and expect it to take three. The tracing is where the findings come from, and doing it by hand is what makes them yours to defend.
You end month one with two artifacts: a score, and a list of the exact hops where the trace went dark. The gap list is the more valuable of the two. It's specific, it names systems, and it's usually surprising to the people who assumed the estate was in better shape.
Circulate both. This is also your baseline, and you will need it in ninety days.
Days 31 to 60: close one gap completely
Now pick one gap and run the whole lifecycle against it, for your three metrics only.
Harvest the source that was dark. Model the definitions properly, with an owner named and the concept bound to specific physical assets. Put those definitions under version control with a change history. Compile them into at least one consuming system so a downstream tool inherits the definition. Build one verification check that tells you when it drifts.
Three metrics, one gap, all five stages. Depth before breadth.
The reason is that a partial implementation across the whole estate teaches you nothing and demonstrates nothing. A complete implementation across three metrics gives you a working reference: a thing you can point at when someone asks what this function actually does, and a template for metric four.
Expect the Model stage to take longer than the four engineering stages combined. Getting two teams to agree on what active customer means is a calendar problem.
Days 61 to 90: make it repeatable, then report
Month three converts a one-time result into a practice.
Stand up the promotion path: a place to make a definition change, review it against production, and move it deliberately. Then run one real change through it end to end, including the rollback, so you know the path works before you need it.
Write down the intake process. Who requests a definition change, what they have to supply, who approves it, how long it takes. A page is enough. Without it you become the queue, and the queue is how this role gets converted back into a help desk.
Then rescore the three metrics against your day-30 baseline and report four things: the score delta with the denominator, the gap list that remains, the one process now running, and one specific ask for the next quarter.
Make the ask concrete. Headcount, budget for a source system connection, or a decision you need from governance. A function that reports a measurable delta and a specific ask is a function that gets funded again.
What to refuse, and how
Four requests will arrive in the first month. Each one is reasonable and each one will consume the quarter.
The estate-wide glossary. Someone will want every term documented. Agree to the goal and scope it to your three metrics for now, on the stated basis that a glossary nothing enforces is a document, and you would rather prove the enforcement path on three terms than document four hundred.
The tool migration. Evaluating platforms feels like progress and produces no evidence. If a purchase is genuinely needed, your day-90 gap list is the requirements document, so the evaluation gets better by waiting.
Data quality requests. Bad values, null rates, and failed loads belong to data engineering and data quality tooling. Say so early and say it consistently. Accepting the first one makes you the owner of all of them, and it's the fastest way to lose the scope you just won.
Reconciliation help desk. You are the person who gets pinged when numbers disagree, which is probably how you got this job. Keep answering, and start logging each one against which of the five dimensions caused it. Six weeks of that log is the most persuasive document you will produce all quarter, because it converts a nuisance everyone tolerates into a pattern with a cost.
The one number
At the end of ninety days you should be able to say a sentence like this: metadata completeness across our three most-reported metrics went from 9 out of 20 to 14, the remaining gaps are these four systems, and definition changes now promote through an environment with a rollback.
That sentence is the job. Everything above is a way of being able to say it.
