The Promise Runs Ahead of the Results
Generative AI has already changed how companies look at their own data. Questions that once required an analyst and a week of waiting can now be answered in seconds by almost anyone in the building. The next step, agentic analytics, goes further. Instead of simply describing what happened, software agents are expected to decide what to do about it and take action on their own. That is an enticing prospect, but a new body of survey evidence suggests the industry is not quite there yet.
The headline number is sobering. Across the organizations surveyed, only about 48 percent of insights generated by analytics agents lead to actions that actually improve the business. That is up from roughly 37 percent a year earlier, which shows real momentum, but it also means more than half of agent output is still failing to move the needle. For companies that have invested heavily in these systems, that gap translates directly into disappointing returns.
Descriptive Insights Are Easy, Prescriptive Ones Are Hard
Most respondents are fairly happy with how their agents describe the world. Charts, summaries, and anomaly flags are generally considered accurate and useful. The trouble begins when agents move from describing to recommending. Prescriptive insights, the ones that tell a team which price to change, which customer to contact, or which supplier to replace, receive far more mixed reviews. A confident recommendation built on incomplete information is worse than no recommendation at all, because it can trigger real decisions with real costs.
The underlying causes are not mysterious. Executives point first to a lack of business context, cited by 55 percent of respondents, and then to fragmented or inaccessible data, cited by 45 percent. An agent can be technically sophisticated and still produce poor advice if it does not understand what a metric means for a particular business, or if the data it needs is scattered across systems it cannot reach. Organizations that have solved those access problems tend to report noticeably better outcomes.
Too Many Insights, Not Enough Judgment
There is also a volume problem. At a majority of surveyed organizations, analytics agents generate more insights than people can reasonably review. Only about a quarter of companies say they can evaluate every agent insight, and most rely on case-by-case human review that is labor-intensive and slow. When the bottleneck is human attention, the sensible fix is not to generate more output but to rank, filter, and govern it. An agent that produces ten strong recommendations a day is far more valuable than one that produces five hundred mediocre ones.
This is where governance and semantics come in. Sixty percent of the companies in the survey are prioritizing stronger AI and data governance, and 45 percent are building a semantic layer into their data architecture. A semantic layer gives agents a shared, consistent definition of terms like revenue, churn, or active customer, so that two agents looking at the same business do not reach contradictory conclusions. Experts consistently describe it as the foundation that high-performing agents depend on.
What Better Insights Mean for Buyers
For IntentBuy readers, this story is less about internal dashboards and more about where technology adoption is heading. Every enterprise that improves its agentic analytics will shape the products, platforms, and services consumers eventually use, from more responsive retail pricing to smarter customer support and supply chains that adjust before shortages appear. The companies pulling ahead are the ones treating data quality and context as seriously as model capability. Watch for vendors that talk less about how autonomous their agents are and more about how reliably those agents are right, because that reliability is what will ultimately decide which AI-driven tools people trust enough to buy.
