Snowflake and MIT Technology Review Insights released research showing that enterprise deployments of agentic analytics struggle to deliver steady business gains. The study, titled Putting Agentic Analytics to Work, found that only 48% of insights produced by software agents lead to actions that improve company operations [1]. While that share is up from the 37% reported a year earlier, more than half of all automated findings still fail to help firms in practice. Leaders surveyed for the project pointed to missing business context as the chief hurdle holding digital tools back from sound choices [2].
Where Agentic Analytics Works Today
The findings come from a survey of 400 senior data and tech leaders conducted in July and August 2026. Every participant works at a firm making at least $500 million in annual sales, with 36% of those surveyed earning over $10 billion each year [2]. The surveyed firms work in eight industries, with retail, finance, and tech groups doing best in turning findings into cash.
A small group of top firms manages to pull ahead. The report spots an elite tier making up 7% of surveyed firms, named insight leaders, where more than 70% of agent insights produce real business gains [1]. Snowflake data shows that the share reaches 77% among this top group. These top performers give their tools much higher quality scores in every area, showing that agentic analytics can work when companies back tools with clean data systems. A market forecast from Bloomberg predicts that generative AI will reach $1.3 trillion by 2032, putting heavy pressure on boardrooms to show real profit from their data tools [3]. Without dependable outputs, enterprise investments risk stalling before delivering their expected returns. The payoff was clear.
Software agents differ from old BI dashboards that sit idle until a user asks a question. AI bots scan data sets on their own, spot new trends, and suggest next steps without code [3]. Speed alone solves little when bots do not grasp what a company wants to achieve. Without context, fast math produces mere noise.
Why Prescriptive Insights Fall Short
Survey responses show a sharp split between what AI tools can describe and what they can recommend. When rating descriptive insights that review what already happened, 83% of respondents judge agent results as good or top grade [2]. That approval falls to 45% for prescriptive advice that tells staff what to do next. The math was easy. Explaining past trends relies on fixed facts, but advising on future moves calls for deep knowledge of firm goals.
When analytics bots produce weak insights, leaders blame data setup flaws instead of model limits. A full 55% of respondents cite a lack of business context as the top cause of bad output, while 45% point to siloed data [1]. As Carl Perry said in the report, “When agents don’t have enough understanding of the business needs, requirements and priorities, they struggle to determine the true value of the insights they generate.” Raghu Ramakrishnan added that prescriptive advice asks tools to turn predictions into actions, which forces software to handle unwritten rules and shifting team budgets. When software can’t read those unwritten limits, even correct numbers fail to guide bosses in daily decisions. Without deep grounding in company reality, autonomous advice quickly turns into guesswork that executives discard [2]. Context made the difference.
Digital teams face this barrier when solving daily web store issues. Mario Ciabarra of Quantum Metric said a site update on Tuesday can spark a 6% drop in mobile conversion by Wednesday morning [4]. Simple AI assistants recap the drop in metrics, but they can’t say why shoppers left because they lack user signals like rage clicks, dead clicks, and buyer hesitation. Without behavioral data, bots guess. Staff then spend days chasing the bug.
Evaluating Outputs Creates a Bottleneck
Even when tools produce helpful findings, firms drown under the sheer volume of automated tips. At 52% of surveyed organizations, agentic analytics tools generate more findings than human staff can review. Only 24% of surveyed firms check all agent outputs, leaving the vast majority of teams forced to inspect individual ideas one by one when time and staff permit [1]. This review logjam stalls work.
To clear this hurdle, leaders plan to hand review duties to AI tools. In two years, 53% of executives expect automated audits by AI agents to be their main review path, up from 29% today. “Many organizations still rely on expert review and anecdotal testing because evaluating insights’ quality is fundamentally harder than evaluating traditional software,” Bogdan Crivat said. He said reviewers must judge if an answer makes sense in context, rather than only testing if the math is right [2]. Human review takes time.
This shift reshapes what data analysts do each day. Instead of building static charts from scratch, 60% of executives expect guiding teams of AI agents to become a primary analyst job within two years [2]. Analysts now spend more hours testing bot answers, refining logic paths, and feeding fixes back into data pipelines. The research group AIRA said early agent tools aimed only at finishing tasks, leaving firms open to bad advice unless they build strict review checks [6].
Semantic Layers and Governance Priorities
To fix context gaps, tech chiefs pour cash into basic data setups instead of buying larger AI models. The study shows that 60% of surveyed firms prioritize AI and data rules [1]. Another 45% of organizations add a semantic layer to unify business terms across tools. Shared data definitions give bots the logic they need to judge metrics fairly.
Unifying metrics across tools has sparked big tech spending. Tom Czaban of GoodData wrote that the market for semantic layers and knowledge graphs is nearing $1.73 billion as firms link reasoning tools with data warehouses [3]. Real agent setups run through a five-step loop: sensing signals, analyzing shifts, explaining causes, recommending fixes, and acting on results. For instance, when store sales dip, an agent can check 90 days of order logs to find stock snags or price mistakes. Grounding tools in shared business rules lets AI systems interpret revenue, churn, and gross profit margins evenly across every branch. Clear definitions keep answers steady.
Can companies trust self-running tools to trigger actions without human sign-off? Most tech managers prefer keeping a person in the loop for big calls. Setting clear rules means that AI bots suggest price cuts or stock moves, but human managers retain final veto power [3]. Guardrails keep systems safe while teams build trust in bot advice.
Connecting Lakehouses to Agent Analytics Tools
Putting agent analytics into enterprise production calls for plugging autonomous tools directly into shared data stores instead of building isolated silos that cut bots off from live numbers. Databricks shows that lakehouse systems offer a unified home where sales tables and raw buyer feedback sit side by side. AI agents query these live stores in the background, spotting root causes without waiting for batch runs. By linking raw tables with reasoning steps, lakehouses supply agents with the fresh data they need [5].
Frameworks such as Databricks Agent Bricks and AI/BI Genie APIs let builders embed chat analytics straight into daily workflows [5]. Teams also use tools like Google Cloud BigQuery and Amazon Redshift to route structured records to agents. New systems rely on retrieval-augmented generation to root agent answers in live facts rather than static training weights. This grounding stops made-up answers and gives human reviewers an audit trail they can inspect. When users can see how a bot reached its verdict, trust grows fast.
The divide in enterprise data will separate teams that treat context as a core system from those hoping bigger language models will fix flawed logic on their own. As agentic analytics systems take on heavier data tasks, feeding them clean business context remains the key test for tech leaders [1]. Firms that set up sound data rules and clear audit trails will turn digital agents into dependable partners for everyday business choices.
- ONLINE NEWS MIT Technology Review Insights. (2026, November 2). Raising the quality of insights in agentic analytics. [Article Link]
- REPORT Snowflake. (2026). Putting agentic analytics to work: Global executive research. [Article Link]
- WEBSITE Czaban, T. (2025, December 21). Agentic analytics: The complete guide to AI-driven data intelligence. GoodData. [Article Link]
- WEBSITE Ciabarra, M. (2026, February 19). The agentic shift: What actually changed in analytics? Quantum Metric. [Article Link]
- WEBSITE Databricks Staff. (2026, October 4). What is agentic analytics? Databricks. [Article Link]
- ONLINE NEWS AIRA. (2026, November 2). Raising the quality of insights in agentic analytics. [Article Link]
