After-hours customer questions are the enquiries that reach your website, inbox or messaging channels while the team is offline and then sit unanswered until morning. The soundest approach is to answer frequent, factual questions instantly from an approved knowledge base, and to log anything that needs a person, with contact details, for a callback at the start of the next working day.
Key takeaways
- A large share of after-hours questions concern delivery, returns, prices, stock and appointments, whose answers are already written down; these can be answered automatically.
- Automated answers should rely only on information the company has approved; where the answer is unknown, the result should be a handover or a callback request, not a guess.
- Every unanswered question should land on one callback list with a name and contact details, and be assigned to an owner in the morning.
- In an NBER study, support agents with access to an AI assistant resolved 14% more issues per hour on average.
- What is not measured does not improve: track night-time conversations, handovers and callbacks every week.
Customers ask the moment they notice a need, and that moment is often after dinner or at the weekend. This guide sets out which questions can be answered automatically, which should stay with a person, and how to make sure a request that arrives overnight is still handled the next morning.
Why do after-hours questions go unanswered, and what does it cost?
In many small and mid-sized firms, the live chat window on the website switches to "We're offline" when the working day ends. The visitor does not bother to write an email, moves to a competitor or simply forgets to come back. What is lost is not just a message but the customer who was closest to buying.
The second cost lands in the morning. Messages that piled up overnight arrive all at once; the same "when will my parcel arrive?" question is answered ten times while an urgent return or a large quote request waits in the queue. A traditional chatbot can offer only fixed menus here and struggles with free-text questions. We compare decision-tree bots with language-model assistants in our customer service chatbot guide.
There is measured evidence that AI can change this picture. In the NBER working paper "Generative AI at Work", covering 5,179 customer support agents, access to a generative AI assistant raised issues resolved per hour by 14% on average, and by 34% for novice agents. Looking ahead, Gartner predicts that by 2029 agentic AI will resolve 80% of common customer service issues without human intervention (Gartner press release, 2025). That is a forecast, but the direction is clear.
Which after-hours questions can be answered automatically?
Not every question suits automation. To sort them, ask whether the answer exists in a written, approved source, and how costly a wrong answer would be.
| Question type | Example | Right approach after hours |
|---|---|---|
| Information | Delivery times, return terms, opening hours | Instant answer from the knowledge base |
| Product and price | "Does this part fit my car, is it in stock?" | Answer from the catalogue with price and stock, link to the product page |
| Action | Adding to basket, booking an appointment | Through a defined site action, with the visitor's confirmation |
| Individual case | Disputed order, invoice correction | Take contact details and log a callback request |
| Sensitive or risky topic | Medical interpretation, legal opinion, complaint | No guessing; hand over to a person or log a callback |
In most businesses the first three rows cover the bulk of night-time questions; check this against your own logs. In the last two, the goal is not to answer but to capture the request without losing the customer. We cover the risk of wrong or invented answers in reducing AI hallucinations, and how to ground answers in sources in RAG for enterprise LLMs.
How do you set up an after-hours support flow?
Treat after-hours support as a small process rather than a tool installation. The steps below work even for a team of a few people.
- Collect the last three months of questions. Group live chat logs, emails and messages by topic and list the 20 most frequent questions.
- Write an approved knowledge base. Give each frequent question a short, current, single correct answer, and assign an owner to information that changes, such as delivery, returns and promotions. The rules that returns answers should rest on are gathered in online store returns process.
- Connect product data. Make sure price and stock answers come from the catalogue or shop data, not from hand-written text.
- Write the handover rule. Decide which topics are never answered automatically and what happens while the team is offline.
- Set up the callback list. Every unanswered conversation should land on one list with a name and a phone number or email address, and the list should have a named owner each morning.
- Be honest with visitors. Show clearly that the assistant is AI, when the team will respond and how personal data is handled.
- Measure weekly. Track night-time conversations, handovers and callbacks, and add unanswered questions to the knowledge base.
If you want to handle enquiries from messaging apps the same way, see our article on the WhatsApp Business chatbot.
What should automated answers get right?
The first rule is that the assistant should not talk about what it does not know. "I'm not certain; let me pass you to the team" is always cheaper than a wrong return policy. This is the customer-service form of the human-in-the-loop principle.
The second rule is that an assistant which takes actions must get explicit consent. Steps such as adding to the basket or submitting a booking form should not run until the visitor confirms. Payment and card fields are places the assistant should never see at all.
The third rule concerns personal data. Names, phone numbers and conversation text collected at night are personal data too; a privacy notice should be visible in the chat window and a retention period should be set. We discuss this in detail in AI and KVKK personal data.
How does SmartOperator make this easier?
SmartOperator is the AI customer service agent in Digital Bridge's Smart360 family, added to a website with two lines of code. It talks to visitors on the site at any hour, basing its answers on the company's knowledge base, product catalogue and the actions the site has defined. When a topic needs the team, it records the visitor's contact details as a callback request.
Take a concrete case: a visitor lands on a car parts website at 11 pm on a Saturday and asks, "Will these brake pads fit my car, and how much are they?" SmartOperator finds the product in the catalogue and shows it as a card with price, stock status and a link to the product page; the name and price on the card are not written by the AI but come straight from the data. If the visitor wants it, the product is added to the basket through an action the site has defined; for important steps such as placing an order or submitting a form, a "Confirm" button appears in the chat and nothing runs until the visitor confirms. The agent prepares the basket and directs the visitor to checkout; on the payment page it reads nothing and runs no actions.
The same night, another visitor writes in to dispute a return. With no matching entry in the knowledge base, SmartOperator does not guess; when nobody is online it shows a contact form, and the request appears on the "Callbacks" screen in the panel. On Monday morning the team works through the list using the New, Called and Completed statuses, and if a notification address is set, the request also arrives by email.
The team stays in control. In the panel, the "How did it answer?" section beneath each reply shows which knowledge base entry or product was used. Values typed into form fields are not read on any page, the privacy notice link the company enters in the panel is shown in the chat window, and the company chooses how long conversations are kept, from 7 days to 10 years. The clinic template writes night-time appointment requests to the shared calendar during the conversation; for the wider appointment process, see hospital queue and appointment management.
Before going live, you can test the agent on your real site without customers seeing it: in hidden mode, the chat bubble appears only to someone who opens the page with the test parameter. For the full feature set, visit the SmartOperator product page.
How does Digital Bridge help with after-hours support?
Digital Bridge treats after-hours support as a customer communication process, not just a tool installation. We first build an inventory of frequent questions from past conversations, write the knowledge base with your team and agree the handover rules. We then set up the catalogue feed or the shop database connection and define the actions specific to your site.
If you need broader language processing, such as an assistant running on your own channels, our NLP and chatbot solutions come in. Our approach to AI across the Smart360 family is described on the Smart360 artificial intelligence page. For how AI supports your human agents, read agent assist AI for customer service; for linking night-time enquiries to your pipeline, see our CRM and sales pipeline guide.
What is the next step for answering customer questions after hours?
This week, pull the last three months of questions and count how many arrived outside working hours. That figure shows more clearly than anything else what automated answers would be worth to you. Our other AI guides are collected on the AI topic page.
When you are ready, explore SmartOperator on the product page and get started on the SmartOperator purchase page. If you would like to plan the knowledge base, catalogue and actions together, contact us.