When the chatbot has no answer, the question should not disappear
A good question deserves an answer. A question a chatbot cannot answer deserves something else: to reach the place where the answer is.
That is the part most chatbots skip. They answer whether they know or not. At broberg.ai we built it the other way round. Aidan, our own AI guide on broberg.ai, and the other chatbots we build for ourselves and for clients, have several routes for a question. Which route they take depends on the question.
1. The answer is in what it already knows
The first route is the simplest. Aidan starts every conversation with a thorough instruction: who it is, what the house does, which rules it must follow, and what is on the site right now. Ask what Trail is, or how we work, and it can answer straight away.
That sounds trivial, but it covers the vast majority of questions. And it is cheap, because nothing has to be looked up.
2. The answer is in a second brain
The second route goes to Trail, our knowledge base. Trail is not a search engine over documents. It is a second brain, where sources have been read and compiled into lasting knowledge, with a pointer to exactly where each claim comes from.
When Aidan gets a question its instruction does not cover, it looks it up in broberg.ai's own brain in Trail. It answers from what is actually written there, not from what a language model guesses. We use the same pattern for clients: their chatbot gets its own brain, filled with their own manuals, prices and experience.
3. The answer is in another system
The third route is what turns a chatbot into more than a conversation partner. Aidan can be connected to a tool through MCP, or to an API, and so act inside another system.
The best example is what happens when Aidan does not have the answer. Instead of guessing, it offers to pass the question on to a person. With one click the conversation becomes a case in HelpDesk, our support platform, with the whole conversation attached, so nobody has to ask the same thing twice. The visitor gets a reference, and the answer comes from one of us.
That is what we mean by escalation: the question does not disappear because the robot could not answer it. It just moves to the right place.
4. The answer is a price
The fourth route is for the questions about money. "What does it cost?" rarely has one answer, because the price depends on what has to be built.
Here Aidan can pass the request on to a quote generator, or to a more advanced form that asks the follow-up questions a price needs: scope, integrations, languages, how many users. The answers can become a price estimate fully automatically or partly automatically, depending on how standardised the job is.
One line we do not move
There is one rule we have fixed, and it holds however good the automation gets: a regular quote for a technical solution we deliver is never sent without a person having been through it.
A price estimate can come from a machine. A quote is a promise, and promises we make ourselves. The automation prepares it. A person reads it, corrects it and stands behind it.
Why it matters
A chatbot that always answers is easy to build. A chatbot that knows when not to, and who to hand the word to instead, is what makes it something you can trust.
That is the difference we build by, on broberg.ai and for our clients.
Want to hear what escalation could look like for you? → Let's talk.
