Comment by bytefish

To me it's a business problem, a communication problem and a technical problem. You would have to solve all for a unified graph to succeed. And that's super hard.

For the communication problem, you'll need to break the "sweet isolation within autonomous teams". You'll need someone to hop between the teams and analyze their data models. It's usually not enough to share a data schema. You'll need to actually sit down with humans and ask questions.

The technical side is by far the easiest one. Share a fat schema, like Microsoft does with Microsoft Graph, and dictate all of your teams to use it. Problem solved. Yes, this requires a lot of empathy and frustration tolerance, because in my experience all software developers (me included) are drama queens.

The technical side can only be solved by dictating and enforcing it. You need management buy-in and authority for this, if you don't have the authority and all teams are still free to roam... you can come up with the best solutions, it won't work.

The hardest part is the business-side. If you have highly optimized processes people have been trained on for 20+ years, it's usually not possible to change processes and associated terminology.

So... You need 100% buy-in from decision makers, that it's useful and the cost of this monumental work is going to pay off in a lifetime.