Enterprise AI Analytics Agents: Plenty of Insights, Less Than Half Are Actionable

Generative AI has been integrated into enterprise analytics for only a few years, and it has significantly lowered the access threshold for analytics use. AI agents are taking this progress a step further, with the goal of autonomously running entire end-to-end workflows and taking targeted actions based on final analysis results. The core problem emerges right here: the quality of insights generated by current generation agents is still far from satisfactory.

A joint report from MIT Technology Review Insights and Snowflake reveals several key metrics. Among the surveyed enterprises, an average of only 48% of insights generated by analytics agents can trigger actions that improve business performance. While this figure is up from 37% a year prior, more than half of all generated insights still deliver no tangible business value. The report warns that low-quality insights that trigger erroneous actions could lead to reputational damage or financial losses.

The report categorizes generated insights into two types: descriptive and prescriptive. Most respondents recognize the value of agents' descriptive insights, which explain what exactly has happened in business operations. However, ratings diverge sharply when it comes to prescriptive insights — recommendations for specific actions to take. This dynamic is analogous to equipping every business team with a highly talkative junior analyst: they churn out outputs at great speed, but their judgment capabilities are still very much in a probationary period.

A small subset of enterprises is classified as "insight leaders" in the report. For these organizations, over 70% of agent-generated insights lead to tangible business improvements, and they also give far higher ratings to the quality of prescriptive insights. The biggest performance gap between regular firms and insight leaders lies exactly in the quality of prescriptive insights.

The top barrier to quality improvement is the lack of business context, cited by 55% of surveyed executives. 45% of respondents pointed to fragmented or inaccessible data. Another very practical pain point stands out: 52% of surveyed enterprises say analytics agents generate far too many insights for their teams to evaluate one by one, making it impossible to implement all of them naturally. Only 24% of enterprises are able to evaluate every single insight from their agents; most firms process insights on a case-by-case basis, with workflows still heavily dependent on human labor.

The core improvement directions center on governance and semantic layers. 60% of surveyed enterprises list strengthening AI and data governance as a top priority, and 45% have added a semantic layer to their data architecture. Experts interviewed for the report argue that semantic layers are critical for high-performing analytics agents.

In other words, the bottleneck for agentic analytics no longer lies exclusively in the insight generation phase. Context availability, semantic maturity, governance frameworks, and evaluation throughput determine whether generated insights can be translated into actual actionable steps. The full report download link is available here.

Source: Custom content from MIT Technology Review Insights, sponsored by Snowflake. The report was researched and written by human authors, with AI tools only used under human supervision for production workflows.