A friend who runs a small accounting practice rang me last quarter, mid-panic.
The AI assistant she had bought from another vendor had spent the previous three weeks happily emailing the wrong VAT figure to the wrong client. It did not error. It did not alert. It did not break. It just quietly applied the wrong template to the wrong customer, in correct British English, formatted properly, sent reliably, every time.
By the time it was caught, eleven clients had received eleven incorrect figures. The practice had to send eleven apologies, eleven corrections, and explain to two of them in person why a four-figure VAT amount had been off by a factor of ten.
This is the worst kind of failure. Not the loud one. The quiet one. The one that confidently does the wrong thing for three weeks before anyone notices.
The principle that prevents this is called failing closed. Every AI system we put live at CueBot follows it, whether that is the assistant on your website or the voice agent answering your phone. Here is what it means and why every system you commission, from us or anyone else, should be held to it.
What failing closed means
When a system encounters a condition it cannot handle confidently, it has two choices.
Fail open means: do something anyway. Take the best guess. Make the call. Quote the price. Book the slot. Hope it is right.
Fail closed means: stop. Hand off to a human. Take a message cleanly. Refuse to act unless you are sure.
Most AI systems, by default, fail open. They are configured to "be helpful". They prefer a confident answer to no answer. This is fine for low-stakes work like suggesting a help article. It is catastrophic for anything that touches money, bookings, or a customer's trust in you.
CueBot's default is the opposite. Every system we put live has explicit failure modes that route to a human, with the relevant context, the moment confidence drops below a threshold. The threshold is set per job:
- The assistant on your site: only answers directly when it is confident from your own approved content. Otherwise it captures the enquiry and hands it to a person rather than guessing.
- Prices and quotes: never invented. If the answer is not in what you have given us, the assistant says it will get someone to confirm, and passes it on.
- Bookings: only confirmed against your real diary. The system never double-books and never promises a slot it cannot see is free.
- The voice agent: qualifies and books what it is sure of, and warm-hands anything outside that straight to a human, with the call summary attached.
This sounds restrictive. It is. The restrictiveness is the point.
What this costs you
A system that fails closed handles a smaller percentage of cases end-to-end than a system that fails open. Concretely: an open system might resolve 85% of customer queries without a human. A closed system might resolve 70%.
That gap is the cost of safety. It is real. We do not pretend otherwise.
What you get in exchange:
- No silent regressions. When the system stops working correctly, it stops doing anything, loudly. You know within an hour, not three weeks.
- No expensive false positives. The thing that quietly emailed eleven wrong figures cannot happen. The system would have stopped at the first one and asked for a human.
- No erosion of customer trust. Customers can forgive "someone will come back to you within the hour". They cannot forgive being quoted the wrong price, booked into a slot that was taken, or addressed as someone else.
- No compliance liabilities. UK GDPR, sector-specific obligations, the rules your business actually lives under, all have hard edges. A system that fails closed never crosses them. A system that fails open might cross several.
An open system might resolve 85% of cases. A closed system resolves 70%. The 15% gap is the cost of safety. It is the right price to pay.
How we build it in
There are four mechanics that make failing closed work in practice. Every CueBot build uses at least three of them.
Confidence thresholds, not boolean rules. When the assistant reads a customer's question, it does not return "answer X" - it returns "answer X, with this much confidence". Below the threshold, it stops and hands off. The threshold is tunable per job, and we document it.
A human in the loop on anything that matters. Anything that quotes money, confirms a booking, or makes a promise on your behalf can be set to pause for a person. The system can draft and prepare. It does not have to send. The pause adds seconds. It removes a whole class of expensive accident.
Loud handoffs, not silent ones. When the system refuses to act, it does not just sit there. It captures the lead, flags it, and puts it in front of a human, so refusal is visible and gets resolved in minutes rather than missed for weeks. David sees the ones that matter.
Idempotency on everything. If the same enquiry or event arrives twice, the system handles it once. The consequence is that retries and duplicates never cause double-actions. No customer ever gets two confirmations for the same booking.
The question to ask any vendor
Before you sign with anyone putting AI in front of your customers, ask this:
"When the system is unsure, what does it do?"
If the answer is "it does its best" or "it uses a fallback" or anything that sounds like the AI will keep going regardless, run.
If the answer is "it stops, hands off to a person with the full context, and picks up once someone has looked", you are talking to people who understand that one quiet error in front of a customer costs more than a thousand routine successes save.
We are not the only people who think this way. We are not even the strictest. But the principle is real, and worth more than any specific technology choice. Pick a partner who lives by it.
The eleven-figure friend, by the way, switched. Her assistant now fails closed. She has not had an apology email to send in six months. If you want to see how that feels, start with a free AI Growth Audit.