Instead of forcing users into a chatbot experience, AI should be integrated where it actually enhances workflows - helping analysts iterate faster without the pressure of crafting the perfect prompt. We broke down why a chatbot-first approach doesn’t work for analytics and show how a better alternative looks like.
Have you ever found yourself rehearsing what you’re going to say to Siri, Google, or any voice system? If so, you’re not alone. I often struggle to phrase my question correctly on the first try.
“Hey Siri, send a text message to Andrew asking him to pick up milk at the supermarket…” (pause)”OK, you would like me to send a text message to Andr—””And coffee!”
Crap! Siri missed it. Now, you have to wait seven excruciating seconds and start over.
This struggle with precise phrasing is also a common issue with chatbot interfaces in BI tools. Many product teams get this wrong when integrating LLMs into their analytics platforms.
Unlike general-purpose AI, where vague questions can be interpreted loosely, analytics demands precision: dates must be exact, cohort segmentations need careful structuring, and lifetime value calculations require proper constraints. This creates pressure to phrase questions correctly—often making chat-based interfaces feel frustrating and inefficient.
For better or worse, analysis requires iteration. Analysts need to fumble through different questions and approaches before landing on clear insights.
But fumbling through prompts in a chatbot is harder than iterating in a UI or even writing SQL. Sometimes, explaining your entire thought process to an AI takes longer than just doing the work yourself. This was a challenge a decade ago when ThoughtSpot pushed for a natural language interface—it worked decently, but never truly took off.
Rather than prioritizing a full chat experience, technical users tend to prefer AI ** features that are thoughtfully integrated into existing workflows, helping users answer questions as they go. These constraints make AI more usable, ensuring it enhances, rather than hinders, the analytics experience.
Some analytics tools take a chatbot-first approach to AI because it feels familiar - just ask a question. While this lowers the barrier to entry, it often overlooks the reality of day-to-day analytics work, where precision and control matter.
A more effective approach is what Omni provides: starting with curated, trusted definitions and then iterating on analysis with the help of AI, rather than forcing rigid chatbot interactions. AI features in Omni enhance workflows instead of replacing them, making iteration faster and more intuitive.
AI can’t solve the hard parts of BI.
You need humans to lay the groundwork. The core challenge is getting the data foundation right. And AI can’t magically do that for us. The salt mines of data work – data engineering, data transformation, and data modeling – require human knowledge about how a business works. This foundation has to be set up before we can plop these new, shiny APIs on top. If we can’t even count customers accurately, how can we expect bots to forecast high-value customers?
Colin Zima, Omni Co-Founder*
Ultimately, AI in BI should be about helping users produce great content more efficiently, not adding friction to an already complex process. Instead of creating rigid chatbot experiences, tools should focus on integrating AI where it genuinely improves productivity and accelerates insight generation.
🚀 Want to see how AI can truly elevate your analytics workflow?
Get in touch with Shearwater to explore smarter, more effective approaches to self-service analytics.
