Messaging channels
WhatsApp Business API, an SMS provider, web chat or Instagram, depending on where your customers are.
AI automation
AI customer service for small business: how it works
An AI agent answers your customers straight away, at any hour, using your business information: prices, hours, services, policies and order status. It understands messages however people write them, resolves what has a clear answer and hands over to a person when the conversation needs judgement. Your team stops answering the same question all day.
How it's usually done by hand
In most small businesses, messages are answered by whoever is free: the front desk, the owner, sometimes nobody. The same questions come in dozens of times a day, and each one interrupts someone in the middle of something else.
Outside business hours, customers get no reply. Since many people contact several businesses at once, whoever replies first usually wins. Old menu bots ('press 1, press 2') help little, because customers don't write the way the menu expects.
Example: a business receiving 60 messages a day, at about 3 minutes of attention each, spends 3 hours a day on messages. If half are repeated questions, an agent handles about 1.5 hours a day, around 30 hours a month, and answers at night too.
How the automation works, step by step
- 1
Knowledge base
Everything the agent is allowed to say is gathered in one place: services, prices, hours, policies, FAQs and your brand's tone.
- 2
Understanding the message
AI understands what the customer wants even with typos, voice-note transcripts or several questions in one message.
- 3
Answer or action
For clear questions, it answers. For simple actions, like booking or checking an order, it looks up your systems.
- 4
Handover to a person
Complaints, negotiation, sensitive topics or questions outside the knowledge base go to your team, with a conversation summary.
- 5
Continuous improvement
Questions the agent couldn't answer become a weekly list for completing the knowledge base.
What goes into the build
AI model
Understands messages and writes answers in your tone, within what's in the knowledge base.
Knowledge base
A document or spreadsheet with everything the agent can say. It's the part that defines quality most.
Integrations
Calendar, CRM, store platform or business system, to look up and record information.
Inbox for your team
Where staff pick up handed-over conversations and see everything in one place.
To start today, without hiring anyone
- List the 20 questions you get most. Look through recent conversations to build the list.
- Write the ideal answer to each one, short and in your brand's voice.
- Save the most common ones as quick replies in your messaging app or help desk.
- Set an away message for out-of-hours, saying when the person will hear back.
- Keep that list: it's the knowledge base for your future agent.
A prompt to get ahead today
Prompt: build your customer service knowledge base
Use it to organize your answers before switching on any automation.
You are a customer service specialist. My business: [what it does, city, customers] Tone of voice: [friendly / formal / playful] Hours: [hours] I'll paste the questions I get most often: [PASTE THE QUESTIONS] Provide: 1) The questions grouped by topic. 2) A ready answer for each group, up to 4 lines. 3) Which questions should always go to a person, and why. 4) Information that's missing to answer well ([CHECK]). 5) A welcome message with 3 numbered options.
What usually goes wrong
- Launching the agent with an incomplete knowledge base. It will guess or keep saying it doesn't know.
- Hiding that it's a virtual assistant. Customers accept it when they know, and get annoyed when they find out.
- Having no clear handover rule. An angry customer stuck with a bot is the worst outcome.
See also
Frequently asked questions
What's the difference between a menu chatbot and an AI agent?
A menu chatbot follows a fixed script: press 1 for this, 2 for that. If the customer writes something off-script, it gets stuck. An AI agent understands the message as it comes, with typos or several questions at once, and answers from your knowledge base. It also recognizes when to bring in a person. For businesses with varied questions, the difference is big.
Can the agent make up an answer?
If it's poorly configured, it can. That's why it's instructed to answer only from the knowledge base and to say it will check when it doesn't know. In the first weeks, it's worth reviewing conversations and filling gaps in the knowledge base. A well-built agent prefers handing over to a person to guessing.
Which channel should we use?
The one your customers already use. In the US, text and web chat are common. In the UK, Australia and many other markets, WhatsApp is growing fast alongside text. Instagram matters for consumer brands. The same knowledge base can power every channel, and it usually makes sense to start with the one with the most volume.
Does the agent work on Instagram and our website too?
Yes. The same knowledge base can power the agent on WhatsApp, SMS, Instagram DMs and web chat, in the same tone. Your team follows everything in one inbox. Starting with the busiest channel and expanding later is usually the best approach.
How long does it take to go live?
The technical part is usually quick. What takes longest is organizing the knowledge base: prices, policies, answers and what should go to a person. With that in place, a first agent can run within a few weeks. The first month is for tuning with real conversations.
Free diagnostic
Your AI Diagnostic
Answer a few quick questions and get a diagnostic made for you, showing where AI should go first in your business.
- Takes under 5 minutes
- A reply from Camilly within 1 business day
- 100% free