Learning how to improve agentic analytics insight quality starts with better business context, shared data definitions, and clear action rules. Give agents trusted information, rank their recommendations, and test results before expanding automation. The goal is not more insights. It is more decisions that produce measurable business gains.

Key takeaways

• On average, 48% of agent-generated insights at surveyed companies lead to actions that improve the business.
• Lack of business context reduces insight quality, according to 55% of surveyed executives.
• At 52% of surveyed organizations, agents generate too many insights for teams to evaluate properly and act upon.
• Surveyed companies prioritize stronger AI and data governance and semantic layers to improve agent insights.

Generative AI has opened analytics to more people across organizations. Agentic AI goes further: it can manage workflows and take action. But greater freedom raises the stakes when an agent recommends the wrong move.

A report produced by Insights, MIT Technology Review’s custom content arm, shows both progress and a gap. Organizations are getting more useful insights from agents, but reliable action still depends on better inputs, evaluation, and controls.

Why do fewer than half of agent insights improve the business?

Missing business context, fragmented data, and limited review capacity help explain why many agent insights do not lead to better outcomes.

Survey respondents say an average of 48% of agent insights now lead to actions that improve the business, up from 37% one year earlier. That is an 11-percentage-point gain. Still, more than half do not produce that result.

This does not mean every unused insight is wrong. A finding may be accurate but arrive too late, lack a clear next step, or sit in a review queue.

The report also distinguishes between two types of output:

• Descriptive insights explain what happened, such as a drop in sales.
• Prescriptive insights recommend what to do, such as changing prices or shifting stock.

Most respondents praise descriptive insight quality. Their views on prescriptive quality are mixed. That matters because a sound description does not automatically support a safe recommendation.

How to improve agentic analytics insight quality with business context

Give agents the goals, limits, and operating rules that a human analyst would need before making a recommendation.

Lack of business context is the leading reason for poor insights, cited by 55% of surveyed executives. Fragmented or inaccessible data follows at 45%.

For example, an agent might suggest a discount after spotting slow sales. Without margin targets, stock levels, and contract terms, that advice could hurt profit.

Create a short context brief for each workflow

Before running an agent, define the following:

• Business goal: State the outcome the agent should help improve.
• Constraints: Include budget limits, policy rules, and actions the agent must avoid.
• Data requirements: Identify approved sources and how current they must be.
• Decision owner: Name the person who approves sensitive recommendations.
• Success measure: Choose the metric used to judge the result.

Start with one recurring decision rather than every analytics task. This makes missing context easier to find and correct.

A semantic layer gives agents consistent business definitions

A semantic layer helps agents interpret data through shared definitions of metrics, relationships, and business terms.

For example, sales and finance teams may define revenue differently. If an agent mixes those definitions, it may produce a confident but misleading comparison.

The report finds that 45% of surveyed companies are adding a semantic layer to their data architecture. The experts interviewed for the report also identify it as crucial for high-performing agents.

For teams deciding how to improve agentic analytics insight quality, a useful first step is to standardize the metrics behind one important workflow. Define each metric’s calculation, source, owner, and update schedule. Expand those definitions as testing reveals gaps.

A semantic layer supports consistency, but it does not fix stale or incomplete data by itself. Teams still need data quality checks.

Prioritize insights before they overwhelm reviewers

Rank agent insights by business relevance, evidence quality, and action risk so reviewers can focus on useful recommendations.

At 52% of surveyed organizations, agents generate too many insights to evaluate properly and act upon. Only 24% can evaluate all agent insights. Most review them case by case, through a process that remains human-intensive.

Ask each agent to attach a brief decision record to its recommendation:

• Evidence: Which data supports the finding?
• Expected benefit: What could improve, and what assumptions support that estimate?
• Risk: What could go wrong if the organization acts?
• Next step: Who should review or carry out the action?

Combine duplicate findings and filter out recommendations that lack evidence or a clear decision owner. Treat an agent’s confidence score as a signal to test, not proof of accuracy.

Governance should control which actions agents can take

Governance should set access limits, approval rules, and records of agent actions before automation expands.

Stronger AI and data governance is a priority for 60% of surveyed companies. Practical controls can include read-only access during testing, human approval for high-impact decisions, and logs that connect each action to its supporting evidence.

For reversible, low-risk tasks, teams can test limited automation. For pricing, payments, or customer-facing changes, stronger review may be needed. Define when to stop a workflow and how to undo an action where possible.

Accountability also deserves attention beyond technical controls. Our related article on potential risks of an AI company liability shield explores the broader liability debate.

A focused pilot can reveal quality gaps quickly

A small pilot can show which changes improve recommendations before a company expands agent access or authority.

A practical approach to how to improve agentic analytics insight quality is to test one workflow against a clear baseline. This suggested sequence is an implementation plan, not a result measured by the report:

• First, review recent recommendations and record why they succeeded, failed, or went unused.
• Next, fix missing context, unclear metric definitions, and data access gaps.
• Then, test new recommendations with human review before agents act.
• Finally, compare business outcomes, review time, and harmful or incorrect actions against the baseline.

The report’s “insight leaders” offer a useful benchmark: more than 70% of their agent insights lead to actions that improve the business. They also rate insight quality more highly, especially for prescriptive outputs. These findings show an association, not proof that any single change guarantees success.

For your team, how to improve agentic analytics insight quality comes down to making recommendations easier to trust, review, and measure. Start with one workflow, strengthen its context and controls, and expand automation only when business results justify it.

(Source: https://www.technologyreview.com/2026/11/02/1147660/raising-the-quality-of-insights-in-agentic-analytics/)

For more news: Click Here

FAQ