- Companies do not trust the accuracy of their AI/ML models but it is due to bad data subordinates, reports claims
- Only one in three has implemented or optimized data observability programs
- Obsability must be default across the entire date life cycle
New research from Ataccama has claimed that a significant part of companies still do not trust the output of AI models – but it may simply be because their data is not in order yet.
The study found that two out of five (42%) organizations do not trust their AI/ML model outputs, yet only three out of five (58%) have implemented or optimized data observability programs.
Ataccama says this can be a problem because traditional observability tools are not designed to monitor unstructured data, such as PDFs and images.
Don’t trust AI? Lack of appropriate data may be the problem
The report also revealed the ad-hoc approach that companies often take, with observability often implemented reactively, resulting in fragmented governance and silos throughout the organization.
Ataccama defined an effective program as proactively, automated and embedded across the data life cycle. More advanced observability could also include automated data quality control and remedy of workflows, which can eventually prevent further problems upstream.
“They have invested in tools, but they do not have operationalized confidence. It means embedding observability in the full-day cycle, from ingestion and pipeline execution to AI-drive consumption, so problems can be surface and solved before they reach production,” explained CPO Jay Limburn.
However, the ongoing lack of skills and limited budgets is still challenges along the way. Ataccama also noted that unstructured inputs continue to grow as a result of increased generative AI and RAG adoption, yet feeding fewer than one in three organizations unstructured data in their models.
The report continues to explain: “The most mature programs close this hole by integrating observability directly into their data technical and management framework.”
With proper observability in place, companies can expect improved data reliability, faster decision making and reduced operational risk.
