Shaping AI Decisions with the Right Questions

Mar 14

By Kateryna Babenko, dedicated to implementing smart, proactive support systems that go beyond resolving tickets by preventing issues before they arise.

“The way we frame our questions determines the insights we’ll uncover.”

I noticed many in the Support Driven community and on LinkedIn using AI to gain insights in a conversational way. When Pylon launched its 'Ask AI' feature, it highlighted the need to guide people in asking the right analytical questions to get meaningful, actionable insights—not just surface-level answers. I want to provide an analytical perspective to ensure they get meaningful, actionable insights rather than just surface-level answers.

Data is abundant, but clarity is scarce. To extract actionable insights, analytical questions must bridge the gap between business needs and AI's capabilities. Poorly framed questions can lead to wasted resources, misaligned efforts, or irrelevant results. By contrast, well-formed analytical questions provide insights that directly impact business strategies.

Here’s an example of a common scenario in customer support:

Broad Business Question: Why is our first-response resolution rate dropping?

Effective Analytical Question: What are the patterns in support ticket categories, agent response times, and customer satisfaction scores that correlate with lower first-response resolution rates over the past quarter?

The second question narrows the scope, aligns with measurable data, and directly informs actionable next steps (refining workflows, training agents, improving FAQs, etc.). But first, let me help you differentiate among four levels/types of analytics.

The Four Levels of Analytics

Ideally (in my world), data-literate users should be able to categorize business questions as descriptive, diagnostic, predictive, or prescriptive.

Descriptive Analytics: What happened? This is about summarizing and visualizing past events. You can identify recurring issues and prioritize improvements, such as improving FAQs or updating content for an AI conversational bot.

Diagnostic Analytics: Why did it happen? Here, we explore the origins of a trend. By analyzing variables like ticket volume, agent workloads, or tool performance, diagnostic analytics identifies root causes, enabling targeted action points.

Predictive Analytics: What will happen? This level focuses on forecasting trends and outcomes. Predictive models help support teams anticipate challenges, enabling proactive outreach or resource allocation to reduce escalations.

Prescriptive Analytics: What should we do next? The most advanced level; recommends actions to optimize outcomes. Prescriptive analytics provides actionable recommendations and guides decisions like staffing adjustments or ticket routing.

When applied effectively, these four levels form a powerful system for transforming data into insights and insights into action, I just promise.

But let's move to the smart way of forming analytical questions.

The SMART Way to Ask Questions

The SMART framework (Specific, Measurable, Actionable, Relevant, Time-Bound) is a powerful tool for forming analytical questions.

Business Need: Improve satisfaction among premium customers (used here as an example customer segment, as all users are premium by default).
Business Question: Why are premium customers reporting lower satisfaction scores?
Analytical Question: “What are the patterns in support interactions - such as ticket response time, escalation rates, and issue types - that most impact satisfaction scores for premium customers in the last six months?”

This question is:

Specific: Focuses on premium customers and their interactions.
Measurable: Uses metrics like response time and satisfaction scores.
Actionable: Guides improvements in agent training or workflows.
Relevant: Addresses a high-priority business goal.
Time-Bound: Anchored in a six-month timeframe.

What I’ve Learned About Asking Questions

“When time, data, and expertise are at stake, asking the right questions becomes crucial. ”

Over the years, I’ve seen how people often struggle to frame analytical questions effectively or even define what trends they want and badly need to track. I was no different when I started.

Here are a few lessons I’ve gathered:

Start with the Customer. Always begin with the customer’s pain points. If customers are reporting delays in ticket resolution, the right question might be: “What ticket categories have the longest resolution times, and how does agent expertise impact these timelines?”
Go Beyond the Obvious Sometimes, the initial business question scratches the surface. For example: Instead of asking "How can we reduce ticket volumes?", ask "What recurring issues in support tickets could be prevented by improving product documentation, AI conversational bot, or onboarding resources?"
Data-Informed, Not Data-Limited Great questions bridge business priorities with data capabilities. They don’t start with, “What data do we have?” but rather, “What do we need to know?” AI does magic on questions that challenge assumptions and seek clarity.

Teams that master the art of asking will find themselves not only gaining insights but also fostering a culture of informed decision-making. This is what I wish for you, my friends.

About the author

Hi, I’m Kateryna Babenko! A Data & Implementation Expert at Katico.

I specialize in customer service tool configuration, automation, and AI-ready content strategies. My focus is on helping businesses optimize their support systems to improve efficiency, streamline workflows, and enhance customer experiences.

Let’s connect!