Generative AI has transformed enterprise analytics, but agentic AI—automating workflows and acting on insights—faces a critical flaw: poor insight quality. Fewer than half of agent-generated insights lead to business improvement. While descriptive insights are praised, prescriptive ones—those driving actions—are unreliable.
Flawed insights risk financial and reputational damage. A new MIT Technology Review report explores how companies are improving agent performance. On average, 48% of agent insights now lead to beneficial actions, up from 37% last year.
Still, over half fail to deliver value. A small group of 'insight leaders' sees more than 70% of agent insights drive improvement, citing better context access and governance. The main barriers are lack of business context (55%) and fragmented data (45%).
Additionally, 52% of organizations face insight overload, with only 24% able to evaluate all agent outputs. To improve quality, 60% are strengthening AI governance, and 45% are adding semantic layers to data architecture—deemed essential by experts for high-performing agents.
Source: MIT Technology Review AI · Summarized by HeadlinesBriefing