Topic starter
25/06/2026 10:29 pm
TL;DR: Gartner expects organizations to abandon 60% of AI projects through 2026 when they lack AI-ready data, while poor data quality already costs the average organization $12.9 million annually, underscoring how fragmentation turns detection into delay and undermines trust, according to Collibra. Data governance has become an operational control, not a reporting layer.
Editorial analysis by NHI Mgmt Group, based on content published by Collibra: “Bad data doesn't slow AI down. It scales the wrong answer.”.
By the numbers:
- Poor data quality already costs the average organisation $12.9 million every year.
- Gartner predicts that organisations will abandon 60% of AI projects that are not supported by AI-ready data through 2026.
Key questions
Q: What breaks when data quality alerts lack lineage and ownership context?
A: Alerts become hard to act on because teams know something drifted but not which model, policy or business process is affected.
Q: Why does poor data governance create risk in AI pipelines?
A: Poor governance lets the wrong data enter training and inference workflows, which can distort outputs, leak sensitive information, and violate privacy obligations.
Q: How do teams measure whether unified observability is actually working?
A: Look for end-to-end dashboards that connect API performance with AI behavior, token consumption, and security signals in one view.
Practitioner guidance
- Map every data quality signal to lineage and ownership Require each anomaly, rule violation or observability alert to resolve to a consuming model, data product and accountable owner before it enters the queue.
- Standardise rule templates across domains Use reusable standards so local fixes do not turn into isolated exceptions that fragment governance across teams and tools.
- Treat AI-suggested rules as governed changes Route machine-generated rule suggestions through steward approval, version control and audit logging before deployment.
Bottom line: Bad data quality does not stay local when AI systems consume it. It scales wrong answers into reporting, forecasting and operational decision-making.
Explore further
View Full Forum → | NHI Foundation Course → | Our Services → | Read the full analysis →
25/06/2026 11:02 pm
Data quality has become an operational governance control, not a reporting afterthought: When AI systems consume fragmented data, they scale the same error pattern faster than humans can spot it. The control failure is not absence of monitoring alone, but absence of a governed route from anomaly to accountable action. Practitioners should treat quality as part of operational identity and trust architecture, not a downstream analytics concern.
A question worth separating out:
Q: Should organisations standardise data quality controls before expanding AI programmes?
A: Yes, because AI expansion magnifies the cost of inconsistent rules, missing lineage and unclear ownership. Standardisation makes quality checks reusable across domains and gives governance teams a stable basis for audit, remediation and trust decisions. If the control model cannot scale, the AI programme scales uncertainty instead of reliability.
👉 Read our full editorial: Bad data scales wrong answers in AI-ready data governance