Raising the quality of insights in agentic analytics

Generative AI has transformed enterprise analytics by making tools accessible to a broader range of users, while agentic AI aims to automate entire workflows and execute actions based on agent-generated insights. However, a new report highlights that the quality of these insights currently lags behind expectations. Fewer than half of the insights generated by agents at surveyed organizations result in business improvement. While most respondents rate descriptive insights positively, assessments of prescriptive insights, which drive specific actions, are mixed.

The report indicates that agentic analytics is only beginning to deliver significant value. On average, 48% of insights generated by analytics agents lead to actions that improve the business, a notable increase from the 37% average reported by respondents one year ago. Despite this progress, more than half of current agent-generated insights fail to assist businesses. The report warns that a failure to produce effective insights deprives organizations of their desired return on investment, while flawed insights that trigger incorrect actions can cause reputational or financial damage.

A distinct group of organizations termed “insight leaders” demonstrates significantly higher performance than the average. In this cohort, more than 70% of agent insights lead to business-improving actions. These leaders also rate the quality of their agents’ insights higher than the overall respondent pool. The gap in performance is particularly pronounced regarding prescriptive insights, where the distinction between high-performing and average organizations is most evident.

According to the survey, insufficient context is the primary reason for substandard insight quality, a factor cited by 55% of executives. Fragmented or inaccessible data is the second major contributor, mentioned by 45% of respondents. Insight leaders report less struggle with accessing context compared to the broader group of surveyed companies. Additionally, insight overload inhibits effectiveness at 52% of organizations, where agents generate too many insights for proper evaluation. Fewer than 25% of organizations are able to evaluate all agent insights, with most relying on a human-intensive, case-by-case process.

To address these challenges, surveyed companies are prioritizing initiatives to strengthen AI and data governance, a strategy cited by 60% of respondents. Another 45% are adding a semantic layer to their data architecture. Experts interviewed for the report agree that this semantic layer is crucial for ensuring high-performing agents. The content was produced by Insights, MIT Technology Review’s custom content arm, and was researched and written by humans with AI tools limited to production processes under human oversight.