Data Ownership, SLAs and Accountabilities

• Assign owner, steward and producer roles to any dataset using a RACI-like model
• Translate business SLAs into per-table targets an owner can be held to
• Design an ownership review so datasets don't become orphans when people leave

• Quality initiatives fail without a person accountable — shared ownership is usually no ownership.
• Declared consumers let producers know who they break when changing a contract.
• SLAs give incidents a clear escalation path: the owner is paged, not 'the data team'.
• Onboarding and offboarding become possible: you know who to hand a dataset to.

Every floor, machine and janitorial route has a person accountable for it. When the elevator fails, you page the facilities manager, not 'building-logy' — data needs the same named accountability.

• Enumerate datasets and declare a catalog entry per dataset (grain, purpose, owner, SLA).
• Assign roles: Accountable (owner), Responsible (producer team), Consulted (stewards), Informed (consumers).
• Derive SLA targets from business expectations and write them into the contract (freshness, completeness, uptime).
• Publish the ownership map in the catalog; make every page/issue route to the owner.
• Run periodic ownership reviews so datasets follow people who left and new datasets get owners at birth.
• On SLA breach, the owner executes the escalation/repair playbook and reports back.

• A bank registers every risk table with an accountable VP; the quarterly SLA review is in the risk committee minutes.
• A product analytics team names the consumer squad per metric definition so reorganizations trigger contract re-negotiation, never silent breakage.
• When an engineer leaves, the catalog's ownership report lists their datasets days before the handoff, not months after.

• One accountable owner per dataset — a named person, not a generic mailing list that none answer to.
• Write SLAs you can verify mechanically (freshness < X, completeness > Y%), not vibes.
• Require onboarding: any new dataset is created with an owner, an SLA draft and a consumer survey.
• Remind owners quarterly: review their dataset's quality scorecard and deprecate dead datasets.

• Platform teams owning everything — they can't answer business questions, so SLAs stay vague.
• Two teams both 'own' a dataset until it breaks, then neither does.
• SLA targets that can't be verified (e.g. 'data shall be reasonable') — unmeasurable promises are unenforceable.
• No offboarding process — datasets survive their owners as orphans sustaining nobody's priority.

• When an incident has no assignee, search the catalog for the dataset's accountable owner rather than escalating up the chain.
• If the SLA is breached but nobody noticed, check whether the monitoring query uses the catalog's role fields — alerts should target the owner's contact.
• For orphaned datasets, run an ownership report sorted by 'owner left date' to spot risk.
• When two teams argue over a column, the steward (business definition) is the tie-breaker, not the pipeline team.