Data team structure: embedded or centralised?

Embedded data teams are closer to business problems but ownership of what to do when data goes wrong is complicated