The Time the Chatbot Said 'No' — And How I Found the Right Words
When building the public landing page chatbot for Arise & Shine Transporters, I hit a wall. The chatbot refused to answer questions about pricing, fuel efficiency, and delivery timelines — no matter how many times I retrained it. This was a turning point that taught me the value of listening closely to what the AI actually needed to understand.
app{ "The Time the Chatbot Said 'No' — And How I Found the Right Words", "When building the public landing page chatbot for Arise & Shine Transporters, I hit a wall. The chatbot refused to answer questions about pricing, fuel efficiency, and delivery timelines — no matter how many times I retrained it. This was a turning point that taught me the value of listening closely to what the AI actually needed to understand.", "The chatbot was meant to handle FAQs about the logistics platform for sand and aggregate delivery in East Africa. It was built using the OpenAI SDK and trained on the company's product documentation. At first, it answered basic questions like 'What is your service area?' or 'How do I place an order?' with ease. But when users asked about fuel efficiency, delivery timelines, or dynamic pricing, it would simply say 'I don't know' — even after multiple retraining sessions with the AI collaborators Claude and Codex.",
"The problem wasn't with the AI model itself. It was with the training data. The chatbot had been trained on generic product documentation, not on the real-world context in which the tool was being used. It couldn't understand the nuances of how distance affected pricing, or why a driver might be flagged for taking an unusual route. The AI wasn't being asked the right questions, and it wasn't being taught the right answers.",
"This realization was humbling. I had assumed that feeding the AI enough data would be enough. But it wasn't. The chatbot wasn't just missing information — it was missing context. I had to go back to the users, the drivers, the logistics managers, and ask them directly: 'What are the most common questions people ask about your service?' Only then could I retrain the AI on the actual language and concerns of the people it was meant to serve.",
"After that, the chatbot started working. It could answer questions about pricing models, fuel efficiency, and even explain why a delivery might be delayed. It wasn't perfect — but it was useful. And that was enough for now. The lesson was clear: AI doesn't just need data. It needs context, and it needs to be trained on the right questions — not just the right answers.",
"Building AI-powered tools for everyday life in Kenya isn't just about writing code. It's about listening, learning, and being willing to start over when the first attempt doesn't work. Sometimes, the hardest part of building with AI isn't the technology itself. It's the moment you have to admit that you've made a mistake — and then figure out how to fix it.",
"I've made many mistakes in the process of building these tools. But each one has taught me something. And that's the beauty of it — the mistakes aren't failures. They're part of the journey. And if you're reading this, you're not alone in your struggles. We're all learning as we go.",
"In the end, the chatbot didn't just answer questions. It became a bridge between the users and the tool. And that's what I hope for every AI-powered product I build — not just to function, but to feel like it belongs to the people it serves.",
