1. Group questions by the answer they need
Start with recent questions your team already has permission to review. Group different wordings that need the same business answer. ‘What do I need for the class?’ and ‘Are materials included?’ may belong to one preparation topic; ‘Can I change my booking?’ may require a different policy and a person’s help.
Separate recurring information from individual cases. Opening hours and published service scope are information questions. A missing order, payment issue, exception request, or complaint may depend on customer records or judgment. Removing personal details from an example does not make every case suitable for AI handling.
- Record the question topic and the approved source, rather than copying private conversations.
- Identify who in the business can confirm each policy.
- Note where a customer needs a person, even if part of the answer is repeatable.
2. Write an answer customers can act on
Begin with the direct answer, then include conditions and a next step. For a hypothetical service business, a preparation answer might identify included materials, required clothing, and where to ask about accessibility needs. The example is useful because it resolves a decision; a vague invitation to ‘get in touch for details’ leaves the original question unanswered.
Use plain language and the terms your customers use. Put exceptions near the main rule. Avoid maintaining one answer in a document and a contradictory version on the website. If an answer changes, identify the authoritative source before updating other copies.
3. Improve the website before adding automation
Place the answer near the decision it supports. Preparation details belong near the service description or booking information. Returns conditions should be discoverable from the relevant shopping or policy pages. A concise FAQ can collect common questions, while each answer links to the full policy when detail matters.
You do not have to choose between a traditional FAQ and an AI chatbot. A good FAQ gives customers and search engines a stable answer to read. A chatbot can help visitors ask in their own words and locate relevant information. Both depend on the same clear, maintained business content.
4. Build a focused business knowledge library
In KivoHelp, you can add an authorized public website, supported documents, and manual business information or FAQs. Review imported text for missing conditions, old dates, and extraction problems. A knowledge library should contain information appropriate for a visitor to see, because relevant excerpts may be used in answers and citations.
Write a small set of test questions for each topic. Include the normal case, a missing-information question, and an exception that should go to a person. AI testing requires an available model on an eligible workspace; Free AI is currently disabled. Free can still be used to organize knowledge and prepare appearance settings.
5. Keep exceptions with your team
A useful first answer should not pretend to solve an issue it cannot inspect. An assistant can explain the published exchange process while leaving approval and an individual order review to the business. It should not claim to reserve stock, make a booking, change an order, or issue a refund.
Set a clear follow-up route. Starter uses the configured business support email. Eligible ticket workflows let a visitor explicitly submit a request, receive a KV ticket number, verify their email, and continue replying in a secure portal. Team replies may come later. Asking the assistant a question alone does not contact anyone or submit a ticket.
6. Review gaps and keep policies current
Give the source library an owner and review it after a policy, service, or product change. Check whether customers are still asking the same question because the answer is absent, hard to find, or unclear. Review saved conversations, unanswered-question insights, and reply ratings where available without treating those signals as proof of accuracy.
KivoHelp supports manual website sync and optional weekly syncing. A failed scan retains the last successful content; editing imported text pauses automatic replacement until syncing is allowed again. Check the important changes yourself. Record whether reviewed examples become clearer over time, rather than promising a numerical reduction in tickets or assuming fewer questions always means better support.
This process can begin before a live chatbot is installed. Free setup is available, Free AI is disabled, and Free does not include website installation or new tickets. Public paid plans are proposed and paid checkout is unavailable. Choose any future plan based on actual questions, source needs, and follow-up requirements.