Stefan Preusler, CEO LeapLytics
Sometime last year, I had one of those moments where you think: this can’t be right. Our team had just answered the exact same question for the third time in a single week – how to license our Power BI visuals when a company has both creators and pure viewers. Same question. Third time. On a Friday afternoon when nobody really wanted to be at their desk anymore.
The Problem Wasn’t the Question – It Was the Timing
Our customers come from different time zones. A large portion of our users are based in South America, mainly Argentina and Brazil. They write to us at midnight our time. And by the time we respond, they’re already asleep. This loop of time zone gaps and repeating questions cost us more hours than I’d like to admit.
The first idea was simple: build an FAQ page. We did. Nobody read it. Or at least not the right people at the right time. I can’t really blame them – I also prefer to just type a question into a search bar rather than scroll through documentation.
The second attempt was an off-the-shelf chatbot tool – embed it, write a few template responses, done. That didn’t work either. The answers were too static, too generic. The moment someone phrased their question slightly differently than the template expected – nothing. Silence. Or worse: an answer that completely missed the point.
The Turning Point: RAG
That’s when we got serious about RAG – Retrieval-Augmented Generation. Sounds technical, but the core idea is simple: instead of hardcoding answers into the bot, you give it access to your own documents, product descriptions, support tickets, FAQs – and it retrieves the relevant information itself before responding.
That was the moment things clicked for us.
We started systematically collecting our most frequent support topics. Not based on gut feeling, but by actually asking our customers: What was your first question when you started using our product? What problem cost you the most time? Some of the answers surprised us – things we considered self-explanatory clearly weren’t.
We fed this content into the chatbot’s knowledge base. And the key part: we can extend it dynamically. New product launches, new recurring question – we add it to the base, and the bot knows it from that point on. No rebuilding from scratch, no IT tickets, no waiting.
The Language Problem – And How We Solved It
Here’s a detail I underestimated: a lot of our product data, documentation, and internal descriptions are in English. But our customers in South America write in Spanish. And they rightfully expect a response in Spanish.
That sounds like a small problem. It wasn’t. A bot that gets asked something in Spanish and replies in English isn’t support – it’s frustration.
The solution was configuring the bot to detect the user’s language and respond in that language – even when the underlying information is in English. That now works reliably. Our customer in Buenos Aires gets their answer in Spanish, even when our team is asleep.
What the Bot Actually Does Today
Three months after going live, we’re seeing that roughly 60–70% of incoming support requests are fully resolved by the bot – without any human involvement. The remaining questions still land in our inbox, but with one crucial difference: the bot has already captured the context, categorized the request, and we immediately see what it’s about.
But there’s another effect I didn’t anticipate: the chatbot helps customers clarify their own questions. Sometimes you don’t fully know what your problem is – you type something in, the bot asks a follow-up, and suddenly you realize: ah, that’s actually what I meant. That wasn’t a planned feature. It just happened.
What I’d Want You to Take Away
If you have a small team that keeps answering the same support questions over and over – don’t start with technology. Start by collecting and understanding those questions. Then look at whether a RAG-based approach makes sense for you.
The bot isn’t a replacement for human support. But it gives us back the time we need to deal with genuinely complex problems – and to actually sleep through the night.
Stefan Preusler is the founder and CEO of LeapLytics, a software company specializing in Power BI visuals and data visualization. He builds products that make data processes simpler and more accessible for businesses.
