What happens when the AI agent doesn't know the answer?

What happens when the AI agent doesn't know the answer?

It doesn't guess, it hands the conversation over. The AI agent says it will check, drops the conversation on your plate with a transcript and a summary via WhatsApp or email, and for emergencies the voicebot calls the on-duty engineer straight away. Nobody is left with half an answer.

This is the first question in almost every intro call, and rightly so: by now everyone has met a chatbot that cheerfully made something up. The honest answer is that a good agent isn't built to know everything. It's built to know where its knowledge stops, and to hand over cleanly at that point. What the chat agents do handle themselves is on the agents page.

What the customer notices at that moment

No error message, no dead-end menu. In chat: "Let me check that with a colleague, can I take your number?" On the phone: a transfer, or a callback note with name, number and the question itself. The conversation ends with an agreement about what happens next, not with an apology.

How does an AI agent know it doesn't know something?

Because it may only answer from a knowledge base we train on your business. If the answer isn't in there, it falls back to a handover instead of falling back on its own language model. On top of that it escalates on signals: emergency words, an angry tone, the same question three times, or quoting an amount.

That last one matters most and is the dullest: prices. An agent that volunteers a figure for a job it hasn't seen ends up costing you money or goodwill. Fixed rates (call-out charge, hourly rate, a standard service visit) are fair game because they're in your knowledge base. Anything bespoke goes to the quote agent or to you.

The four escalation rules we switch on as standard

• Unknown answer → handover with transcript, no improvisation.
• Emergency words (leak, no hot water, gas smell, breakdown) → transfer straight to the on-duty engineer.
• Bespoke price or discount → a site visit or intake, never a figure off the cuff.
• Two failed attempts → bring in a human instead of a third round of questions.

What does one lost conversation cost?

In the installation trade an emergency job easily runs from €300 to €800. So one caller who drops out because they got no answer costs more than a month of FAQ agent (€79) or voicemail bot (€79). That's exactly why the handover matters more than a bot with an answer for everything.

Seen that way, an agent that passes things on neatly isn't half a solution, it's the whole point. Prices updated: July 2026.

| Product | Setup | Per month | | --- | --- | --- | | FAQ agent: an answer to every question | €0 | €79 | | Appointment agent: books into your calendar | €250 | €99 | | Lead agent: intake with photo and video | €250 | €129 | | Voicebot (AI receptionist) with emergency triage | €250 | €249 incl. 300 min. |

Run the numbers yourself on the pricing page; how the phone products differ from one another is covered in voicebot, voicemail bot or outbound bot.

Can my customer always reach a human?

Yes. Every agent and every voicebot has a way out to a human: a 'rather speak to a person' button in chat, and on the phone a direct transfer to the on-duty engineer. You decide in advance who that is, inside and outside working hours, and which words count as an emergency straight away.

That isn't a concession to the technology but a design choice. A bot you can't escape from loses in one conversation the trust that a hundred good conversations built up. So that exit is in there from day one, at night too: the route is then usually a callback note rather than a transfer, unless you run your own emergency service.

How does the agent get better each month at what it didn't know?

Every handover is free research. Once a month we go through the conversation logs looking at one thing: which questions led to a handover? Those answers go into the knowledge base, and the same question no longer has to reach you the month after. That's included in the rent.

In practice that first-month list looks nearly identical at every installation and construction business: what does a call-out cost, do you cover my area, how fast can you come, do you take on small jobs, and do you offer a warranty. Those five are handled after one round. What's left after that is the interesting part, because it's specific to the business: one installer turned out to mostly get questions about heat-pump subsidies, another about access to a shared building's plant room.

That difference is exactly why the knowledge base matters more than the model underneath it. An otherwise identical agent gives a handover at one company and an answer at another, purely because last month's logs were processed. More examples of what these agents pick up in practice are in what an AI agent does for your construction business.

Curious how many of your incoming questions an agent would handle without a handover? In a free 30–45 minute AI workflow scan we go through your most common questions and tell you honestly which ones would still land on your desk.