24/7 chatbot: what night replies give you, and the risks

A message read in the morning is already late

A customer writes at 23:40: «Can I change my order?» The answer arrives at 09:30. By then the customer has either bought elsewhere or written the same question on a second channel, and the team now answers one person twice. The trouble with night messages is not that they go unanswered. It is that they arrive in the morning as one pile — no context, the urgent mixed in with the routine.

A 24/7 chatbot closes that gap on one condition: what it may answer at night, what it must not promise and what it hands over in the morning are written down first. This article is about writing those rules. Keeping a night-time sales lead warm is a different job, covered in how to keep after-hours leads.

Why a night request is different

A customer writing at night and one writing at noon may ask the same question, but they are not in the same position. A slow reply during the day still tells the customer someone is on shift. Silence at night reads as «nobody is there», and the customer either gives up or starts sending more messages.

• A large share of night messages are information questions — hours, address, price, delivery terms. The knowledge base answers them safely.
• Some are status questions: «where is my order?». The answer depends on live data.
• A smaller share need a decision: a return, an exception, compensation, a complaint. Nobody who can decide is awake.
• And a rare but critical share are urgent: safety, health, money being lost.

Everything a 24/7 chatbot is worth comes from telling these four apart. The first gets a full answer, the second an answer if the data is there, the third a promise that it will be handled in the morning, and the fourth a person, now.

The answer boundary: what can be said at night

It is practical to write the boundary as three levels. Every type of request falls into one of them, and the bot has to know which.

1. 1Full answer — no person neededOpening hours, address, product details, approved prices, the text of delivery and return rules. If the answer sits word for word in the knowledge base, the bot gives it at night with the same accuracy as by day.
2. 2Partial answer and a morning handoverReturn requests, order changes, individual quotes, contract questions. The bot takes the request, collects what is needed — order number, reason, how to reach the customer — and names a time: «Your request is logged; a manager will contact you by 10:00 tomorrow».
3. 3Immediate escalationSafety, health, a payment error, open anger, a legal threat. Here the bot does not try to answer. The person on call is notified, and the customer is told honestly that the matter has gone to a person.

Example: a return question at 23:40

The flow below shows a typical second-level request. Note that the bot does not approve the return; it only gets the request ready for the morning.

1. 1Customer«The shoes I bought yesterday are too small, I want to return them.»
2. 2Bot — rule and question«Returns are possible within 14 days if the item is unused. Send your order number and tell me whether you want another size or a refund.»
3. 3Customer«No. 48213. One size up, please.»
4. 4Bot — record and time«Noted: order 48213, size exchange. A manager will check stock and message you by 10:00 tomorrow.»
5. 5Morning — on the manager's screenOrder number, reason, the outcome wanted and a one-sentence summary of the conversation. The manager starts from the solution without asking the customer anything again.

The bot does little in this flow, yet it halves the morning's work: the manager does not ask for the details again, and among dozens of overnight requests the urgent one is visible at once.

An escalation rule for the night

Escalation by day is simple: an agent is on shift. At night three questions need answers in advance — who is notified, through which channel, and what happens if that person does not respond.

• A rota: who keeps their phone on at night, and who it is this week
• A notification channel the person will actually see — often a Telegram group or a direct message
• An urgent list: which words and situations trigger an immediate alert
• A no-response rule: if the person on call does not react within 15–30 minutes, a second person
• The sentence the customer sees: honest that the matter was handed over, without promising an hour

The full catalogue of escalation cases — which are immediate, which within the session and which are only flagged — is in chatbot escalation rules. At night, keeping only the first group active is usually enough.

Where it creates risk

A 24/7 chatbot is useful, but it can do harm in four places, and they are worth knowing beforehand.

• Stale information: a promotion has ended and the bot quotes the old price at night. Correcting it in the morning is unpleasant for customer and team alike.
• A wrong promise: the bot says «delivered tomorrow» and the warehouse is closed at the weekend.
• A missed emergency: the customer describes a serious problem mildly and the bot treats it as routine.
• The morning pile: if every night request lands in «morning handover», the bot simply moves the queue to the morning and adds no value.

The last risk is the one most often missed. If most night requests fall into the second level, the knowledge base is too thin or the rules are written too cautiously.

How to measure night mode

«The bot replied» does not show that night mode works. Five numbers do, and they need tracking separately for night and day, because daytime figures hide a night-time problem.

1. 1Split by levelWhat share of night requests was answered fully, handed to the morning or escalated. The morning share should fall month by month.
2. 2First reply time on morning handoversIf the bot said «by 10:00», how many requests actually got a reply by then. A broken promise is worse than an unanswered message.
3. 3Escalation reaction timeTime from the urgent alert to the first action by the person on call. Without this figure the escalation rule exists only on paper.
4. 4Repeat-message rateCustomers who got a night reply and wrote the same question again in the morning. A high rate means the reply was a formality.
5. 5Wrong answersAn old price, a wrong promise, a missed emergency. They are few, but each one should be read and fed back into the knowledge base as a fix.

Audit checklist

Check these before switching night mode on, and once a month after that.

• At least fifty night messages from the last month have been read and sorted by type
• Each type has a written level — full, partial, escalation
• There is a list of promises the bot must not make at night
• The rota and alert channel have been tested — a test message actually arrived
• The morning handover shows the order number, reason and summary
• Promotions and price changes reach the knowledge base the same day
• The first half-hour of the morning shift is reserved for overnight requests

Conclusion: night replies should shrink the queue, not move it

What a 24/7 chatbot gives a business is not the line «we answer at night too». The real benefit shows up in the morning: overnight requests arrive sorted by type, completed with the details needed, and the urgent ones already dealt with. That only happens when the answer boundary, the escalation rule and a regular audit are all in place.

In Vexvon, night mode runs on the same AI engine across every channel — see what the AI chatbot does for the details. The rest of this section is collected under AI chatbot strategy. If you would like to write these boundaries against your own night messages, ask us for a demo and assessment.

Frequently asked questions

1. 1Which businesses is a 24/7 chatbot suited to?Any business that receives messages at night and at weekends but does not staff those hours: online shops, clinics, training centres, service companies. If most night messages need an urgent decision, the bot works as a filter, not a replacement.
2. 2Does a 24/7 chatbot fully replace the agent?No. It closes information questions, prepares the rest for the morning and passes emergencies to a person. Work that needs a decision stays with people.
3. 3What is needed to set it up?A current knowledge base, a list of promises not to make at night, a rota and the format of the morning handover. The technical connection is the easiest of the four.