Nearly half of the analyses produced by the data analytics agents of the surveyed companies lead to actions that improve business performance. According to a study published by Insights, the custom content branch of MIT Technology Review, separate from its editorial staff, this rate reaches an average of 48%, compared to 37% a year earlier. Therefore, more than half of the results remain, according to the study, without useful effect for the companies.
The authors identify a small group of organizations, which they call “insight leaders.” In this group, over 70% of the agents’ analyses lead to actions beneficial for the business, and these organizations rate the quality of these analyses higher than the overall survey respondents. The gap is particularly marked for prescriptive analyses, meaning those that recommend an action to be taken.
A Lack of Context and Dispersed Data
When analyses are deemed of insufficient quality, the main cause cited by the surveyed executives is the lack of business context, mentioned by 55% of them. Fragmented or inaccessible data comes second, at 45%. Insight leaders are less likely to say their agents struggle to access the necessary context.
Volume is also an issue. In 52% of the surveyed organizations, the agents generate a number of analyses too high to be properly evaluated, and therefore to be acted upon. Less than a quarter (24%) manage to evaluate all analyses. Most proceed on a case-by-case basis, a process that, according to the report, currently heavily engages human teams.
Governance and Semantic Layer
To improve the quality of analyses, the surveyed companies prioritize strengthening AI and data governance, cited by 60% of them. 45% also add a semantic layer to their data architecture, meaning a level that gives data a common and shared meaning within the company.
The experts interviewed by the study’s authors agree on the role of this semantic layer, which they consider crucial for agents to be effective.
