What happens if AI can't answer a customer enquiry?
If AI can't answer a customer enquiry, it doesn't guess. It flags the message for a person on your team and sends nothing on your behalf. The system is built to recognise the edge of what it knows, not to paper over it with a plausible-sounding reply.
Why this is the first question worth asking
Anyone considering automating their inbox worries about the same thing first: what if it gets something wrong and a customer never finds out? A properly built system doesn't reply unless it has a real basis for the reply, so a confident wrong answer isn't the main risk. The real risk is what happens to the enquiries it can't handle, because that's where a badly built automation quietly drops things.
How the system knows what it doesn't know
An enquiry-triage setup reads every message as it arrives and classifies it: a routine question with a known answer, a new enquiry, an existing client with something specific, or something that doesn't fit a recognised pattern at all. The reply only gets drafted when the message matches a category the system has real information for, from your actual prices, stock, services, or availability.
A complaint, a bespoke pricing question, a message whose tone suggests a relationship is at risk, or simply a question nobody configured an answer for all go to a person instead of getting a reply. No draft goes out where there's nothing solid to draft from.
What "flagged" actually means in practice
A flagged enquiry doesn't sit at the bottom of an unsorted inbox waiting to be noticed. It's surfaced specifically, with the reason it was flagged attached, so whoever picks it up isn't starting cold. They're not re-reading a full inbox to find the one message that needed them; they're looking at the short list the system couldn't safely handle, already separated from everything it could.
That's a different job from "triage the whole inbox." It's closer to "review the exceptions," which takes a fraction of the time and loses nothing in the process — the straightforward enquiries are already answered, and the ones that needed a person are already waiting for one.
What this means for approval
Every reply the system drafts for a routine enquiry still sits in an approval queue before anything goes out; nothing is sent automatically, flagged or not. The build gets tested against real enquiries before it goes live, and the categories it's trusted to answer are agreed upfront rather than guessed at by the system itself over time.
Dooco runs this exact setup on its own inbound enquiries: every message to the business is read, classified, and either answered from a drafted reply or routed to a person, since July 2026. The enquiries this piece describes as "flagged" are the ones that have come through dooco's own inbox and gone straight to a person rather than getting an automated answer.
The honest limit
This only works as well as the categories it's given. An enquiry-triage system can't handle a question nobody thought to configure it for — it routes that question to a person instead, which is the correct outcome, but it means the setup needs revisiting as a business's products, prices, or services change. That upkeep is part of what a Retainer covers once a system is live.
If what's holding you back from automating your inbox is the worry that something will slip through unanswered and unnoticed, that's the part this is built to prevent. See how the inbound enquiry triage Pattern works at dooco.