Use case
A support bot that answers the repeat questions, and passes on the rest
Most support inboxes are the same twenty questions asked a hundred ways. A support bot answers those from your own help content, at any hour, and hands everything else to a person with the details already collected.
What the bot does, and what stays with a person
The bot
- Answers common questions from your help articles, in plain words
- Walks people through set-up steps one at a time
- Sends the right link for password resets, invoices and exports
- Collects the account and the problem before handing over, so nobody repeats themselves
A person
- Complaints, and anyone upset or confused
- Billing disputes and refund decisions
- Questions the bot cannot answer from your content
- Anyone who asks for a person, straight away
Try it: Tally for Tally Books
Bookkeeping software for sole traders (fictional). This is the designer’s test chat running the example plan. Ask anything, or pick a suggestion.
Tally · Tally Books
Bookkeeping software for sole traders (fictional). Example plan, test chat
Hi, I’m Tally, the automated help assistant. Ask me about bank feeds, invoices, tax returns or your account.
Greeting · says it is automated
How do I connect my bank account?
Go to Banking, choose Add account and pick your bank. You’ll be sent to your bank’s app to approve the connection. It takes about two minutes.
Answer from the plan · matched “How do I connect my bank account?”
Can I talk to a person?
I’ll pass you to a person. Next step: live chat with our team. Our team is available Monday to Friday, 9am to 5:30pm.
Hands over to a person
Out of hours it says: “Our support team is offline. Leave your email and we will reply first thing on the next working day.”
Answers come only from this example’s plan, in your browser. Open this plan in the designer
The example plan
- Goal
- 40% of support chats answered without a person by month three, and first replies under one minute.
- Jobs at launch
- Answer common questions, Product help and how-tos, Account and billing questions, Prices and packages
- Channels
- Website chat, Inside our app, Email
- How it answers
- AI from your documents
- Hands over to
- Live chat with our team, Monday to Friday, 9am to 5:30pm
- Answers written
- 10 questions
Start with the questions you already get
The best source for a support bot is your own inbox. Export the last three months of emails or chats, group them by what the customer wanted, and count. The top twenty groups usually cover most of the volume, and they are the bot’s first job.
Missing content is the usual reason self-service fails. In Gartner’s global survey, customers could not find relevant content in 43% of failed self-service attempts (opens in a new tab). So write each answer in the words customers use, add the other ways they ask, and keep each answer short enough to read in a chat bubble.
- One question, one answer: split long help articles into the questions they answer
- Add three or four other phrasings per question, taken from real messages
- Link to the full article for detail, rather than pasting it all in
- Give every answer an owner and a review date
Make the handover part of the design
People mind bots that block the way to a person. In a 2026 survey for Citizens Advice, 40% of those who tried to reach a human after meeting a chatbot found it difficult or impossible (opens in a new tab). A good support bot offers a person the moment someone asks, and says when the team is in.
Decide the triggers in advance: a request for a person, a complaint, a payment problem, or two answers in a row the bot is unsure about. Pass the conversation to your team with it, so the customer does not start again.
Measure problems solved, not chats “contained”
A chat that ends without a handover is not always a solved problem. Some platforms count silence as success: Intercom’s Fin, for example, records an “assumed resolution” when the customer does not reply for 24 hours (opens in a new tab), and charges for it.
Ask “Did that answer your question?” after each answer, track how many people come back about the same thing within a week, and read a sample of conversations every month. The designer’s measure step sets these targets for you.
Watch for: confident wrong answers about money
If the bot tells a customer the wrong price, refund rule or contract term, your business is responsible for it, as it would be for a member of staff. Keep answers about money in written, checked text, not generated on the fly, and re-test them whenever a policy changes.
Questions people ask
How many questions should a support bot answer at launch?
Enough to cover most of your volume, usually the top 15 to 30 questions. More answers that nobody asks make the bot harder to keep right. Add more from the questions it misses after launch.
Should it use AI or a script?
For a help centre with a lot of content, AI that answers only from your articles works well, with a script for anything involving money or accounts. The designer’s channels step helps you choose, and the guide on AI or scripted bots explains the trade-offs.
Will customers mind talking to a bot?
Many will, if it gets in the way. Say it is a bot in the first message, keep answers short and accurate, and make a person easy to reach. People accept a bot that solves their problem quickly.
Keep going
Make this plan your own
Open the example in the designer, change it to fit your business, test it and export it. Free, no sign-up.
Sources
- Gartner, Only 14% of customer service issues are fully resolved in self-service, 19 August 2024 (opens in a new tab)
- Citizens Advice, Behind digital walls, September 2026 (opens in a new tab)
- Intercom Fin help centre, Fin AI Agent outcomes, updated 17 September 2026 (opens in a new tab)
- CMA, Complying with consumer law when using AI agents, 9 March 2026 (updated 30 October 2026) (opens in a new tab)
Links checked 15 November 2026.