AI customer service chatbot connected to your systems

Which enquiries a customer service chatbot can handle, and which it can't

Before choosing a tool, take a normal month of enquiries and sort them. They almost always fall into four groups, and each needs a different solution:

• Information: opening hours, delivery terms, returns policy, how to book. Any chatbot that reads your pages and documents can answer these.
• Questions about the customer's own data: 'where is my order?', 'when is the engineer coming?', 'what do I still owe?'. These need the customer identified and a live query to your ERP, booking system or carrier.
• Tasks: changing a booking, opening a return, requesting a simple quote. The chatbot carries out a specific action in your systems, with rules and a log.
• Ones a person should handle: complaints, upset customers, commercial negotiations and anything you mark as sensitive. The right move is a quick hand-over with context.

Intercom, Zendesk, Tidio or a chatbot connected to your systems

If most of your enquiries are informational and you already use a helpdesk, start with its AI agent. Each vendor charges differently: Intercom bills its Fin agent per outcome on top of seats; Zendesk includes a set number of automated resolutions per agent each month and charges for extra ones; Tidio sells its Lyro agent as monthly conversation quotas. Check their pricing pages against your real volumes.

A custom AI chatbot makes sense when answers depend on your data and your rules: an ERP with no connector, your own way of identifying customers, actions inside your systems, or a volume at which paying per resolution gets expensive. It also lets you decide which AI provider is used and where data is processed. If the job is a team member's work across several systems rather than customers' questions, see AI agent development.

WhatsApp chatbot for business: how Meta charges

A chatbot connects to WhatsApp through Meta's WhatsApp Business Platform, not the phone app. Since 1 July 2025 Meta charges per template message rather than per conversation. Templates fall into three categories: marketing, utility and authentication. Marketing templates are always charged. Non-template messages, such as the chatbot's text replies, are free inside the 24-hour customer service window that opens each time the customer writes to you, and utility templates sent inside that window are free too.

In practice, a chatbot answering people who message you works almost entirely inside that window. What you pay for are messages you start outside it: delivery updates, appointment reminders or campaigns. So when we design the chatbot, we decide with you which notifications are worth sending.

Keeping answers accurate and handing over to a person

The chatbot answers only from what its sources return: your approved content and queries to your systems. If it cannot find the answer, it says so and offers a person rather than improvising. When it hands over, your team receives a summary, the identified customer and, where relevant, the order.

If you serve customers in the EU, Article 50 of the EU AI Act has applied since 2 August 2026: AI systems that interact with people must be designed so those people are told they are dealing with an AI, unless that is obvious. The UK has no equivalent general AI law, but saying so up front is good practice anyway, and we build it in. Whether AI providers train on your conversations is covered on our private ChatGPT for business page.

What drives the price and the monthly running costs

An online retailer could start with a website pilot that only handles order tracking, then add WhatsApp or returns once real data shows how many enquiries it resolves. At each demo, every two weeks, you test the chatbot with your customers' own questions; the rest is on how we work. What moves the cost:

• Integrations: how many systems it queries and whether they have an API.
• Actions it can take, and their security rules.
• Channels: website, WhatsApp, app, or several at once.
• Languages and the amount of content to prepare.
• Running costs: AI model usage by conversation volume, WhatsApp template messages, hosting and maintenance. Market references are in the AI cost guide linked below.