Automating customer service
Which questions to hand to AI, which to keep, and how to start in six steps. Per channel, with honest costs and the mistakes worth skipping.
Updated 17 September 2026, by the Heyloha team
Short answer
Automating customer service means recurring customer questions are answered and handled by software, usually an AI agent, while your team keeps the conversations that genuinely need a person. Start with the ten questions you get most often, not the hardest ones.
Gartner expects AI to resolve around 80% of common customer service issues on its own by 2029. For most businesses that is not realistic today, but a large share of repeat questions is: opening hours, delivery times, appointments, order status and prices.
The difference between a project that works and one that annoys customers comes down to three things: answers come from your own information, the AI can finish a task rather than just reply, and a person takes over smoothly when needed.
What should you automate, and what not?
A good rule of thumb: automate what is predictable and has a fixed answer. Keep with people whatever needs judgement, empathy or an exception.
Good to automate
- Frequent questions about opening hours, prices, delivery times and terms
- Booking, moving and confirming appointments
- Looking up order status, return status and stock
- Qualifying leads and passing them to your CRM
- Handling conversations outside office hours
- The first intake on a fault or complaint, so your team starts with context
Better left with your team
- Complaints where the customer is angry or upset
- Exceptions to your policy, such as goodwill gestures
- Negotiating price or terms
- Medical, legal or financial advice
Chatbot or AI agent: what is the difference?
A classic chatbot follows a script: predefined buttons and fixed answers. When a customer phrases the question differently, it gets stuck. An AI agent understands plain language, finds the answer in your information and can complete a task, such as putting an appointment in the calendar.
On the phone there is one more distinction: does the AI only summarise the call after you picked up, or does it answer and handle the call itself? Only the second solves missed calls. More in the guide to comparing AI phone assistants.
Automating customer service in 6 steps
This order avoids the most common mistake: buying a tool first and working out what it should do afterwards.
- 1
List your questions
For one week, note what customers call, email and chat about. In most businesses the bulk is about a handful of topics. Those are your first candidates.
- 2
Choose your channels
Start where the pressure is highest: the phone during busy spells, or chat in the evening. Doing one channel well beats doing five badly.
- 3
Build one knowledge source
Your website, prices, terms and FAQs are the foundation. An AI answering from your content gives correct answers; an AI without a source makes them up.
- 4
Define the handover to a person
Decide when the AI passes a conversation to a colleague and make sure the context travels with it. A customer who has to repeat themselves gives up.
- 5
Connect what a conversation should produce
An appointment in the calendar, a lead in your CRM, a return in your webshop. Without integrations, every conversation ends at 'we will get back to you'.
- 6
Measure, improve, expand
Read conversations back in the first weeks, fill gaps in your knowledge source, and only then add the next channel.
The approach per channel
Each channel asks for something different. A platform that serves all of them from one knowledge source stops customers getting a different answer per channel.
Phone
Often the biggest gain, because a missed call costs revenue straight away. An AI phone assistant answers on its own, including during busy spells and after closing, books appointments and transfers with context.
Email suits sorting and preparing. The AI answers standard questions and lines up the rest, with a summary, for your team. See AI for email.
Website chat
A visitor asking a question on your site is close to buying. An AI chatbot answers straight away and can show a form in the chat to capture a lead or appointment.
Customers expect fast replies here. Mind the cost: from 1 October 2026 Meta charges per service message after the first 1,000 a month. A WhatsApp agent that answers completely in one message keeps that bill low.
Measure what happens, not what stays away
The most common metric is the share of conversations that never reach a person. It is misleading: a conversation also never reaches a person when the customer gives up. A mediocre AI looks fine in that report.
Measure what actually happened instead: how many appointments are in the calendar, how many leads in your CRM, and how many customers came back within a week with the same question. Those numbers do not lie.
Mistakes worth skipping
Starting with the hardest questions
Complex exceptions are the last thing you automate. Start with volume, not complexity.
An AI without its own knowledge source
A general model sounds convincing even when it is wrong. Let the AI answer only from your information, and say honestly when it does not know.
No way out to a person
A customer who gets stuck and cannot reach anyone is a customer you lose. The handover is not an extra; it is the foundation.
Hiding that it is AI
Since 2 August 2026 the EU AI Act requires you to make clear at first contact that someone is talking to AI. It is also sensible: customers who feel misled do not come back.
Set and forget
You learn most in the first weeks. Read conversations back and fill your knowledge source where the AI hesitated.
What does automating customer service cost?
There are roughly three routes. A standard platform charges a fixed monthly fee and is live within a day. Custom work by an agency often costs several thousand euros up front and takes weeks to months. In between sit platforms you set up yourself, with integrations built on request. Want someone to set it up for you? Read how to have an AI assistant built.
Heyloha starts at €149 a month. One credit equals one chat conversation or one call minute, and every channel and unlimited users sit in the same plan. See pricing or use the ROI calculator to work out what it delivers for you.
Do not forget what you already pay: missed calls, overtime and onboarding time. What does customer service cost? works those costs out.
What businesses achieve with it
Van der Linden Makelaardij
675 conversations a month, around 56 hours saved.
Mojo Concerts (De Grote Schijn)
32 hours saved and 70% fewer repeat questions.
Read all customer stories.
When automating is not the best choice
- You get few questions and answer them easily yourself.
- Almost every conversation is bespoke and needs personal judgement.
- Your information is not written down anywhere and changes every week.
Frequently asked questions about automating customer service
How do I start automating my customer service?
For one week, list what customers contact you about and start with the ten questions that come back most. Pick the one channel under the most pressure, make sure the AI answers from your own information, and decide when a colleague takes over.
Will AI replace my customer service staff?
Usually not. AI takes over repeat questions and out-of-hours conversations, so your team has time for complaints, exceptions and advice. Most businesses use it to avoid hiring as volume grows.
How many of my customer questions can I automate?
It varies by business. Gartner expects AI to resolve around 80% of common issues on its own by 2029. Plan more conservatively for a first year and measure it yourself: the share actually completed, not just the share that never reached a person.
Is automating customer service GDPR compliant?
Yes, provided you handle personal data carefully. Check where data is stored, how long it is kept and whether calls are recorded. Heyloha runs on servers in the EU and transcribes phone calls without recording them. See security.
Do I have to tell customers they are talking to AI?
Yes. Since 2 August 2026 the EU AI Act requires you to make clear, at the latest at first contact, that someone is interacting with artificial intelligence. A line in your terms is not enough; the notice belongs visibly or audibly in the conversation.
How long does it take to automate customer service?
With a platform that uses your website as its knowledge source, a first version is often live within a day; Heyloha usually takes about 30 minutes to set up. Custom agency work takes weeks to months, because scripts and integrations have to be built first.
What does automating customer service cost?
Standard platforms charge a monthly fee; custom work often needs thousands of euros up front. Heyloha starts at €149 a month with every channel in one plan, and you try it free for 14 days first.
Start with your ten most asked questions
Put them in Heyloha, let the AI answer them, and see after two weeks what it does for your team.