Data quality is back at the top of the list

What the numbers describe

Salesforce surveyed 3,800 analytics and IT decision makers and 3,852 line-of-business leaders in 18 countries between June and August 2025. Besides the 26% estimate, it found that 89% of data and analytics leaders with AI in production had seen inaccurate or misleading AI outputs, and that only 43% had formal data governance frameworks and policies in place.³

IBM and Oxford Economics asked 1,700 senior data leaders across 27 geographies between July and September 2025. Most had a data strategy tied to their technology roadmap, at 81%, yet only 26% were confident their data could support new AI-enabled revenue streams.⁶ Gartner’s survey of 432 organisations, published in June 2025, found data availability and quality among the top barriers to AI, named by 34% of leaders at low-maturity organisations and 29% at high-maturity ones.⁷

The practitioner view comes from dbt Labs. In its 2025 report poor data quality was the challenge data teams cited most often, at 56%.⁸ In 2026, trust in data and data teams as an organisational priority rose from 66% to 83%, and 71% of respondents worried about incorrect data reaching stakeholders.⁹

Why it outlasts every tool cycle

Two findings point at ownership more than technology. dbt Labs found ambiguous data ownership to be a persistent obstacle, cited by 41% of respondents in 2026.⁹ Precisely and Drexel University’s LeBow College of Business found that 71% of organisations with a data strategy and a governance program reported high trust in their data, against 50% of those without.¹⁰ That second figure is a correlation, and organisations with governance programs probably differ in other ways too. It is still consistent with what the rest of the evidence suggests: quality improves when somebody is accountable for it.

Gartner’s 2025 forecast puts a price on doing nothing. It expects organisations to abandon 60% of AI projects through 2026 where the data is not AI-ready, and it found 63% of organisations lacking, or unsure they have, the right data management practices.¹¹

How much weight to give each survey

Most of these studies come from companies that sell data tools, and every one of them has a reason to find a data problem. None of that makes the figures wrong, though it matters for how you quote them. Note the sample, the country mix and the fieldwork dates. The dbt Labs 2025 report, for example, was published in 2025 and based on fieldwork from late 2024. BARC’s Trend Monitor rates importance, which is a measure of attention and says nothing about how good anyone’s data actually is.

What gives the picture its weight is agreement across sources with different methods and different commercial interests. On data quality, the agreement is unusually strong.