Before AI started creeping into workplaces (and data), data was managed by engineers with specialized skills. Then Excel came along and gave people a bit more independence. Eventually, BI (business intelligence) tools and data-wrangling platforms removed the need to copy-paste your sanity away every Monday morning.
In our data and analytics platform Coupler.io, pre-AI flow meant getting data from different sources, automating exports, and cleaning things up (which, honestly, was already a lot of work). However, once everyone finally had access to data, a new question arose: what does all this data actually mean?
People don’t collect data just to admire it. They want to make decisions, and AI made that faster. This is where conversational analytics comes in: analytics that finally match the way we actually think.
Below, I have gathered the main highlights from my recent podcast with Code Story, including differences between traditional and conversational analytics, how we embedded it in Coupler.io, and what it means for users.
Traditional vs Conversational analytics
The difference is simple. With traditional analytics, you are expected to know what to look for and which questions to ask. Conversational analytics meets you where you are. Let’s compare them in terms of conversation.
Traditional analytics = inner monologue
A dashboard may show you something, but often you don’t know who created it or what the numbers truly mean. If you want different insights, you have to:
- Tweak filters created by an engineer,
- Build a pivot table yourself, or
- Ask a data analyst to update the report.
Conversational analytics = productive dialogue
You can ask questions directly in natural language:
- “Why did my conversion rate drop?”
- “What happened to my landing page results?”
The AI provides context-rich insights. It knows your role, the reports you work with, and the reason you’re asking.
This is why the Coupler.io team decided to embed conversational analytics into the product (spoiler: it was not an easy task).
Key challenges of integrating conversational analytics
Our team faced three main challenges while integrating an AI agent for conversational analytics.
1. Context limits
One of the main challenges was the limits of the context window. Even though AI can remember more than humans, it struggles with hundreds of thousands or millions of rows. Coupler.io had to find a way to efficiently feed large datasets into the model.
The solution was simple but effective:
- Provide a schema describing all columns, data types, and metadata.
- Supply 5–20 sample rows so the AI understands the data.
- AI generates SQL queries, which they run on the full dataset on our side.
2. User perception
Beyond technical challenges, we faced challenges related to product and user perception. Not everyone knows what AI is or trusts it, among marketers, finance teams, and e-commerce managers. Thus, educating users and demonstrating clear value were major focuses.
3. Reliability factor
Ensuring the AI delivers reliable insights is critical. Their current approach focuses on prompt engineering.
Key safeguards include:
- Prompts with clear constraints (e.g., “don’t make up numbers”).
- Complete awareness of the data schema to prevent hallucinations.
- Tailor-made prompts for each type of insight.
Even with these measures, human oversight is essential. AI acts as a co-pilot, generating ideas. Yet, humans must decide which are actionable.
How the new AI feature changed UX at Coupler.io
AI insights are directly integrated into Coupler’s native dashboards. Clicking it sends aggregated dashboard data to the AI along with a custom-made prompt.Within 20–30 seconds (faster for smaller datasets, longer for larger ones), the AI returns with:
- Key findings
- Trend information
- Actionable recommendations, such as the top three priorities to focus on.
This democratization of data represents significant changes in user experience:
1. Data is no longer siloed
This shift transforms who can interact with organizational data. As data access is no longer limited to trained analysts, stakeholders, and team members from different departments, and even small business owners can now explore and understand data in human-readable terms.
2. The barrier between information and action is drastically reduced
Perhaps most importantly, the role of a “data user” is now more active and exploratory. Instead of passively receiving reports, users can engage with data conversationally. Plainly speaking, they can “talk” to their data. This means: asking follow-up questions, testing hypotheses, and iterating on their understanding.
How conversational analytics changes data tools
Conversational analytics is not meant to replace existing tools. Instead, it complements them (at least in the near future 🙃). Dashboards remain essential for operational purposes: monitoring KPIs, receiving alerts when metrics cross thresholds, and providing at-a-glance visualization through graphs, charts, or office display screens.
Thus, in the near future, dashboards will continue to play a key role, especially for monitoring and operational oversight. Conversational analytics acts as an additional layer, enhancing discovery, interpretation, and decision-making rather than replacing traditional BI.
Numbers now answer questions, not just sit on dashboards
A few years ago, nobody expected their data tools to talk back. You opened dashboards, built reports, filtered columns, and maybe (if you were particularly brave) ran some SQL. However, everything changed the moment AI evolved from pattern recognition to language understanding.
This transformation happened because the way humans think has always been conversational – fluid, nonlinear, full of follow-up questions and sudden shifts in curiosity. And once AI became capable of understanding that natural thinking pattern, something entirely new became possible: conversational analytics.
For more on this topic, tune in to the Code Story podcast.
