A useful chatbot does not try to answer every question. It handles a defined set of common requests, makes the next step clear, and hands off when context, judgement, or sensitivity requires a person. That is a better starting point for a Surrey business than adding a generic bot and hoping it creates value.
Good first jobs for a chatbot
Start with questions that are frequent, well documented, and low risk: service coverage, opening hours, basic process guidance, enquiry capture, and routing. Build a clear handoff to a person when the visitor is ready or the request falls outside the approved knowledge.
- Answer repeatable service questions.
- Collect enough context for a useful follow-up.
- Route urgent or complex conversations to a human.
What a chatbot should not guess
A bot should not invent prices, promise availability, provide legal or medical advice, or make decisions from incomplete context. Its knowledge should have an owner and a review date.
- Define approved answers and escalation rules.
- Log unanswered questions for improvement.
- Tell visitors when they are interacting with automation.
Design the conversation around the business
Digital Allo’s AI chatbot service focuses on useful, brand-aligned conversations and human handoff. A short pilot tied to one customer journey makes it easier to measure helpfulness and correct gaps.
- Choose one audience and one journey.
- Write fallback messages before launch.
- Review transcripts for accuracy, tone, and missed intent.
Good and poor jobs for a chatbot
| Good first jobs | Keep with a person |
|---|---|
| Opening hours, location and service area | Complaints and disputes |
| Which service fits a common situation | Quotes that depend on an inspection or assessment |
| Collecting enquiry details for follow-up | Medical, legal or financial advice |
| Booking or callback requests | Anything involving passwords, payment details or health records |
| Pointing to the right page or form | Unusual requests outside the approved knowledge |
What a good handover looks like
Visitor: “Can you look at our server this week? It keeps restarting.”
Better reply: “I can’t diagnose the server here, but I can get a technician to contact you. What’s the best phone number, and is the business affected right now?”
Poor reply: “Server restarts are usually caused by overheating. Try cleaning the fans.” The bot has guessed at a technical cause and given advice it cannot stand behind.
Before you launch
- Write the 20 to 30 questions customers ask most, with approved answers.
- Decide what the bot says when it does not know.
- Tell visitors they are chatting with an automated assistant.
- Do not ask for information you do not need, and say how enquiry details are used.
- Test with real questions from staff before going live.
A practical starting point
- Pick a narrow, high-volume use case.
- Create an owner-approved knowledge source.
- Measure completed next steps, not conversation count.
Frequently asked questions
Will a chatbot replace customer service?
It can reduce routine work, but people remain important for judgement, exceptions, trust, and complex situations.
How do we keep answers accurate?
Use a controlled source, assign ownership, test common questions, and review conversations regularly.
Does a chatbot need to say it is automated?
It should be clear to visitors that they are talking to an automated assistant and how to reach a person. That builds trust and avoids misunderstandings.
Further reading
- Office of the Privacy Commissioner of Canada: Principles for responsible generative AI
- Government of Canada: Guide on the use of generative AI
How Digital Allo can help
Digital Allo designs AI chatbots around a defined set of questions, approved answers and a clear handover to your team. Tell us which questions take up your team’s time and we will suggest a pilot.
