01 / What changed
Agents, not scripted chatbots
Older chatbots followed decision trees and broke the moment a customer phrased something unexpectedly. Current AI agents read the question, look up your own information, and reply in natural language. They can also take actions: check an order, book a slot, create a support ticket.
Used well, they answer the routine 60 to 80 percent of enquiries instantly, any time of day. Used badly, they frustrate exactly the customers you most want to keep.
02 / Start with your data
An agent is only as good as what it knows
Feed it your real answers: pricing pages, service descriptions, policies, past replies to common emails. Keep that information in one place and update it when things change. Most bad AI answers are not the model’s fault. They come from outdated or missing information.
03 / Keep it human
Five rules for the personal touch
- Say it is an assistant. Customers do not mind AI. They mind being misled.
- Write its voice like your best team member. Warm, short, specific.
- Use the customer’s context. Their name, their order, the page they came from.
- Never trap people. A clear way to reach a person, visible at every step.
- Hand over with the full conversation, so the customer never has to repeat themselves.
04 / When to hand over
Let people handle the moments that matter
Complaints, refunds, sensitive personal situations, big purchases and anyone who asks for a human should go to a person, quickly. Set the agent to recognise frustration, such as repeated questions or angry words, and offer a handover before the customer has to ask.
05 / Measure it
Know if it is working
- Share of enquiries resolved without a person.
- Customer rating after AI conversations.
- Questions the agent could not answer, reviewed weekly to fill gaps.
- Response time before and after.
Reviewing a sample of conversations each week is the fastest way to improve it, and to catch mistakes before customers notice.
06 / Where to start
One channel, one job
Begin with the website chat or your contact inbox, and one clear job, such as answering pricing and availability questions. Expand once that works. A small agent that is right every time builds more trust than a big one that is right most of the time.
Next step
Thinking about an AI assistant?
We build assistants trained on your own content, with a clean handover to your team. See AI chatbots and assistants.
