How Do You Answer Customers Overnight Without Hiring Anyone?

Someone has a question at eleven at night and you are asleep. By morning they have either bought from you, bought from someone else, or forgotten they wanted it.

You can fix that without hiring, and the first step costs nothing. This page covers what automation genuinely handles, what happens to the part it does not, and the two pricing models that make the headline figure meaningless. About seven minutes.

Write the answers before you buy anything

Most overnight questions are the same eight questions. Shipping times, returns, sizing, whether it works with the thing they already own, what happens if it breaks.

An assistant will turn your product documentation into a comprehensive FAQ in one pass — twenty questions with real answers, grouped into getting started, billing, troubleshooting and the rest. Give it your product description, your last fifty support emails if you have them, and ask for answers complete enough that nobody needs to follow up.

This does two things at once. It removes volume, because people find the answer themselves. And it produces the corpus a bot needs, because a chatbot is only as good as the documentation behind it. Skipping it and buying automation first is the most common way this goes wrong.

The same applies to your reply templates. Five or six well-written ones — refund request, shipping delay, feature request, billing question, complaint — cover most of what you type manually, and they are a twenty-minute job with the prompting approach from earlier in this series.

What "handles it automatically" means

Comparison coverage puts automated resolution for small-business setups at roughly 55 to 65%. Treat that as a rough shape rather than a promise — it depends almost entirely on how good your documentation is, which is the variable you control.

The useful way to read it is the other way round. Around a third of conversations will not be resolved, and that third is the entire product decision. What happens to them?

Good handling: the bot says plainly that it cannot answer, takes an email address, tells the customer when a person will reply, and creates a ticket. The customer is mildly disappointed and still a customer.

Bad handling: the bot loops, offers three irrelevant articles, or answers confidently and wrongly. Now the customer is annoyed at your brand rather than at the software, and a wrong answer about a refund policy at 2am is worse than silence.

The pricing nobody quotes

Both major tools charge for the AI separately from the product, and they do it in completely different ways.

More than double the headline. And the add-on is metered by conversation, so a good month for your business is a more expensive month for your support bill — which is a reasonable model, but not one most people price for.

Neither model is worse. Flat tiers are easier to budget and waste capacity in a quiet month; per-outcome billing is fairer and unpredictable. Pick the one whose failure mode you mind less.

Set it up so the failures are graceful

1. Generate the FAQ from what you already have

   You’ll have: Twenty real questions with answers complete enough to end the conversation. · about An afternoon

   Feed your product documentation and your recent support emails to an assistant and ask for grouped questions and full answers. Then edit — this is a first draft of your support knowledge, not a finished one, and the parts only you know are the parts that matter.

   Publish it. Volume drops before any bot exists, and you now have something to point automation at.

2. Decide what the bot is not allowed to answer

   You’ll have: A written list of subjects that always reach a person. · about 20 minutes

   Refunds beyond your stated policy. Anything about a specific order going wrong. Complaints. Legal or safety questions. Anything where being confidently wrong costs more than being slow.

   Configure those as escalations before you go live, not after the first incident. This is the step that decides whether automation helps your brand or damages it.

3. Write the handoff message yourself

   You’ll have: A failure message you would be happy for a customer to receive. · about 10 minutes

   The default is some version of "I'm sorry, I didn't understand that", which tells the customer nothing and leaves them stuck.

   Replace it. Say that a person will answer, say roughly when, take an email address, and confirm the message was received. A customer who is told "someone will reply by 9am" is fine. A customer who is looped is not.

4. Read the transcripts every week for a month

   You’ll have: A list of the questions your documentation does not answer. · about 15 minutes a week

   Every failed conversation is a gap in your FAQ. Read them, add the missing answers, and the resolution rate climbs without changing any setting.

   This is the part that makes the subscription worth having, and the part almost nobody does. After a month it becomes a monthly check rather than a weekly one.

Before adding this to your stack, question one from the tool selection guide is the one that matters: if you get three overnight questions a week, a published FAQ and an autoresponder is the honest answer, and $68 a month is not.

Questions people ask

How much does an AI chatbot cost?

More than the headline, because the AI is billed separately on both major tools. Tidio is $29 for the base plan plus $39 for 50 AI conversations, so about $68 to start. Intercom is $29 per seat plus $0.99 per resolution with a 50-resolution minimum, so about $78.50.

Can a chatbot handle customer service on its own?

It can handle the routine majority. Comparison coverage puts automated resolution around 55 to 65% for small-business setups, which means roughly a third of conversations still need a person. Plan for that third — it is where the brand risk sits, not in the part that works.

What happens when the chatbot cannot answer?

Whatever you configured, which is why it is worth configuring deliberately. The good version admits it cannot help, takes contact details, promises a reply by a specific time and creates a ticket. The default version usually just says it did not understand.

Do I need a chatbot or just a better FAQ?

Write the FAQ first regardless. It reduces the questions that reach you at all, costs nothing but an afternoon, and is the material any bot draws on. If volume is still costing you sales after that, automate — but a bot in front of thin documentation has nothing to work with.

Is Tidio worth it for a small business?

It is one of the more sensibly sized options, provided you budget the real figure of about $68 a month rather than the advertised $29. The metered tiers make it predictable, and it climbs steeply — $289 a month at 500 conversations — so check the tier your traffic will actually reach.

What percentage of questions can AI answer?

Reported figures sit around 55 to 65% for small businesses, but the number depends almost entirely on how complete your documentation is. That is the encouraging part: it is the variable you control, and reading failed transcripts weekly raises it without changing any setting.