Ownership and operating model in data programs
Unclear ownership, limited leadership direction, and weak upstream requirements rank above technical debt as constraints on data programs. Architecture cannot resolve an ownership gap.
The requirement this answers
“Define the target operating model for data and AI across the enterprise”
Ownership, decision rights, and decommission accountability are first-phase deliverables.
The questions I ask first
Who is accountable by name for the quality of this domain's data? Where the answer is a committee, the effect is the same as no owner, and the cost of that gap shows up in everything downstream.
The second question is what gets switched off when this is delivered. A program without a decommission commitment tends to produce two platforms, two cost lines, and two versions of the same number.
Structure follows consumption
Whether you land on mesh, fabric, a central platform team, or a federated model matters less than whether the boundaries follow how the business consumes data. Domain ownership works where domains are real and have leaders with budget authority.
Run the platform as a product: named owners, published SLAs, unit economics, and a governance cadence for cost, quality, and evals comparable to the rigor of financial close.
Roles change, and that needs to be said
Data engineers move from pipeline authors to platform and context owners. Stewards move from clearing queues to adjudicating ambiguity. That is an elevation, and it needs to be framed and resourced as one. Presented as an efficiency measure it tends to be resisted, and around a third of workers already report quietly resisting AI tools.
Adoption is a people problem. A working prototype that nobody uses returns nothing.
How I apply it
- Name accountable owners per domain before designing anything
- Publish decision rights across business, platform, and governance
- Commit decommission dates with an accountable executive
- Fund role redefinition and skills uplift as a workstream
What good looks like
- Every critical domain has a named owner who would be contacted first
- The platform publishes SLAs and its own unit economics
- Legacy systems are retired on the dates committed
- Adoption is measured, and low adoption triggers a response