Customer service 24/7,
with nobody watching the phone.
Most messages a business receives arrive when nobody can answer them: at night, at the weekend, or while you are serving someone else. An assistant connected to your data replies on the spot, in the customer’s language, and logs who asked and what they wanted.
What does automated service actually solve?
It solves simultaneity, not politeness. Nobody can serve a customer at the counter and quote a job over WhatsApp at the same time, and whoever writes at eleven at night has gone cold by morning. An assistant answers instantly, always, without taking the human touch away from what genuinely needs it.
What gets lost most is not the difficult customers: it is the easy questions nobody answered in time. Price, opening hours, whether you do a given thing, whether there is a slot this week.
What does it answer, and with what information?
It answers with your data, not generalities: your catalogue, your services, your published prices, your delivery or booking terms. It is given access to the real information and only answers within it.
The rule that makes it usable is the opposite of what people expect: what it does not know, it does not invent. When a question falls outside what it has, it says so and hands the conversation to a person, with the full context already written down.
Where does it answer?
| Channel | What it does | What it needs |
|---|---|---|
| Website chat | Serves the visitor, answers from the real catalogue and captures the contact details of serious enquiries | A snippet of code on the site |
| Replies instantly, understands the request and logs it; the team picks it up with full context | A number and a WhatsApp Business account | |
| Instagram and social | Answers direct messages with the same information as everywhere else | Connecting the account |
| Forms and email | Sorts what comes in and only flags what needs immediate attention | Nothing new: it connects to what you already use |
Languages, where half of these systems fall over
An assistant built on language models understands and replies in whatever language it is written to, without building a separate flow per language. That matters in any business with international customers: they write in their language and get the answer in theirs, with nobody translating by hand.
The detail usually forgotten is the record: the conversation is stored in the original language and summarised for the team, so whoever picks it up does not have to decipher it.
What it does not do, and it is worth saying
It does not close complex sales and it does not negotiate. It does not replace the judgement of somebody who knows the trade. And it must not treat a figure it does not have as fact: better it says “a person will confirm this” than invent a delivery date.
Nor is there any point putting it to work before the information it will use is in order. An assistant connected to messy data answers confidently with things that are not true, which is worse than not answering.
How to test it before committing to anything
You start with one channel — usually web chat, which is the most contained — with a small set of real questions, leave it running for a few days and read every conversation. That shows immediately what people actually ask, which is almost never what you imagined.
Nothing here is bought blind: every piece can be seen running before you decide.
Watch it working
Forty seconds of a real assistant answering on a website, analysing the visitor’s business live and admitting what it does not know.
Frequently asked questions
Can customers tell they are talking to a machine?
They can tell when the system replies in stock phrases or repeats the same message. They can tell far less when it answers with real information about the business and admits what it does not know. Our advice is not to disguise it: an assistant that answers well, replies instantly and hands over to a person when it should makes a better impression than one pretending to be human that gets caught out by the third question.
What happens if someone asks something it does not know?
It says so and hands the conversation to a person, with everything already written down. That is the most important rule in the whole setup: we would rather it hands over too often than answers too much. An invented figure about a deadline or a price costs far more than a handed-over question.
Do we have to change software to put this in?
No. It is built on top of what you already use: your existing website, your WhatsApp number, your management system. What it needs is access to the information it has to answer with, not a replacement for your tools.
How long until it is working?
Web chat is usually answering within days. WhatsApp depends on business account verification, which Meta controls and we do not. What takes the time is not the technology: it is getting the information it will answer with into order.
Is there a record of who wrote in?
Yes. Every conversation leaves the contact and what they wanted somewhere it can be picked back up, with nobody typing anything by hand. That record is half the value: without it, the conversation is handled and then lost.
You might also like
Articles from our blog related to your sector.