I've set up EA tools for organizations ranging from a single architect getting started to teams running one as the backbone of a group-wide EA practice. Most platforms are flexible enough to support almost any approach — which is exactly why it's easy to get the first few months wrong.

1. Don't start with the metamodel workshop

It's tempting to spend the first weeks designing the perfect metamodel before entering any real data. Don't. Start with one real question — "what applications support this capability?" — and build only the components and references that question needs. Extend the metamodel as new questions come up, not in advance of them.

2. Assign ownership before you assign fields

Data quality problems are almost always ownership problems. If nobody's name is attached to keeping an application's data current, it will drift within a quarter. Decide who owns what before you decide what fields to track.

3. Use workspaces to protect trust, not just to organize

A single workspace with everything in it looks efficient early on, but it means one team's half-finished draft data sits next to another team's production-quality model. Separate draft from trusted, even if it means more workspaces than feels tidy.

4. Build one dashboard before you build ten

Most EA tools make it easy to build views. Resist building a dashboard for every stakeholder in month one. Build the single view that answers the question your sponsor actually cares about, get it in front of them, and let the next view be a response to a real request rather than a guess.

5. Train the second wave of users differently than the first

The first people you train are usually architects who think in models already. The second wave — business stakeholders, product owners — need a completely different onboarding: less about the tool's mechanics, more about "here's the one view relevant to you, here's how to read it."

None of this is really about the tool. Whatever platform you use will hold whatever structure you give it — the lessons above are really about sequencing the org and data work correctly, so what you build in month three is still what you're using in year two.