A data quality SLA is a customer-facing (or stakeholder-facing) promise about quality attributes of a dataset: usually freshness, availability, and sometimes accuracy/completeness targets.
Example promise: "fct_orders is available by 09:00 IST on weekdays, no more than 1 hour behind source, with primary-key uniqueness and not-null on order_id / amount."
Quality SLA bundle: freshness: age <= 1h by 09:00 weekdays availability: table queryable after successful publish completeness: required columns 100% non-null uniqueness: 0 duplicate order_id
How you operationalize it
- Define measurable thresholds (not vibes)
- Monitor continuously
- Alert owners on warn vs breach
- Publish status in catalog / status page
- Have a remediation runbook (rerun, backfill, communicate)
SLA vs hope
"We'll try to keep it fresh" is not an SLA. An SLA is explicit, measurable, and owned.
Interview tip: Frame a quality SLA as a measurable promise (often freshness + key tests), with owner, threshold, and escalation.