To automate customer service in an SMB, first find the 20% of questions that make up 80% of your volume (order status, opening hours, returns, appointments) and let an AI chatbot on your website and in your mailbox, or an AI voice agent on the phone, handle those, with clear rules for handing over to a person. At Utomatic an AI chatbot starts at €2,450 one-time + €249 per month and an AI voice agent at €3,950 one-time + €349 per month (call costs of about €0.05 per minute are separate). You are live in 2 to 4 weeks; a 1 to 2 week pilot costs €1,500 to €3,000 and is deducted from the full project.
Customers expect a quick answer, at nine in the evening and on Saturdays too. That used to be something only a large company could deliver. In 2026 a business with five to two hundred employees can do the same with AI customer service. Below you will read how to approach it without customers feeling fobbed off, what it costs and when you are better off not starting.
The customer service challenge for SMBs
As an SMB owner you are stuck with a paradox:
- Customers expect more: they are used to the service of the big Dutch webshops and measure you by the same yardstick.
- But your budget is limited: staffing from 8 AM to 10 PM, seven days a week, is unaffordable.
- Staff is scarce: Statistics Netherlands (CBS) has named staff shortages one of the biggest obstacles for businesses for years.
- Peaks are unpredictable: sometimes 3 calls a day, sometimes 30. After a mailing or an outage the inbox floods.
You know the result: the same fifteen questions every day, the phone during lunch, and on Monday a weekend's worth of e-mail. AI customer service removes exactly that repetitive work. It scales with the peak, is always on, and your staff keep the conversations where a person makes the difference.
What AI customer service looks like in practice
Automating customer service rarely happens in one go. We see three levels; most companies start on the first.
Level 1: the smart FAQ bot
A chatbot that answers your most frequently asked questions in natural language, based on your own documentation: product info, terms, opening hours, return policy. You can read how such a bot works in What is an AI chatbot?; the terms you will come across are in our AI glossary.
Result: the bulk of standard questions is handled without anyone getting involved.
Level 2: the transactional assistant
The chatbot can take actions: look up order status, register a return, reschedule an appointment. For that we connect it to your webshop, CRM or planning system. This is also where aftersales automation begins: the bot sends confirmations itself, asks after a week how things are going and flags customers who have not ordered for months.
Result: customers arrange what they need themselves, when they need it.
Level 3: the AI voice agent
The phone is handled by AI too. A voice agent holds conversations in natural Dutch (and dozens of other languages), schedules appointments, logs callback requests and only transfers when really necessary. For practices and installation companies this is often the first step, because the phone is their busiest channel. See also AI receptionist Netherlands.
Result: 24/7 phone availability without night shifts.
Channel by channel: what AI handles itself and what goes to a person
The question is never "can AI do my customer service", but "which part, on which channel". This overview comes from our own projects and is a starting point, not law.
| Channel | AI handles fully on its own | Must go to a person |
|---|---|---|
| Website chat | Opening hours, delivery time, return policy, product questions, order status (with integration), booking appointments | Emotional complaints, custom quotes, anything outside your documentation |
| E-mail / Outlook | Classifying and routing mail, answering standard questions, reading attachments, drafting a reply | Contractual commitments, legal questions, complaints that call for goodwill |
| Phone | Answering 24/7, intake, making or moving appointments, callback requests, FAQs, transferring | Emergencies, angry customers, complex faults, negotiations |
| Confirmations and reminders, order status, attaching photos of a defect to a ticket, questions outside office hours | Payment disputes, warranty discussions, anything you need to decide on |
Website chat is the easiest starting point: the chatbot runs on your own site and you can review every conversation. E-mail is often the channel with the most hidden work; our Outlook agent reads incoming mail and drafts a reply. In one of our projects that saved 2 hours a day, with 95% correct classification and under a second per e-mail. WhatsApp works well for short, practical messages; see WhatsApp chatbot for business. For webshops these channels come together in our e-commerce approach.
Start here: the 20% of questions that are 80% of the volume
The biggest mistake we see: wanting to automate everything at once. The result is a bot that knows a bit about everything and is good at nothing. First find the handful of topics that make up most of your volume and make those perfect. Here is how to find them in an afternoon:
- Export your mailbox. Take three months of your info@ or support@ address (Outlook: export the folder to CSV, or sort by subject) and tally per topic: order status, return, invoice, appointment, price question, complaint, other.
- Request the call log. Your phone provider or VoIP switchboard gives times and call durations. Have whoever answers tally for a week what people call about. Watch for missed calls outside office hours.
- Look at searches and chat logs on your website. If you already have a chat widget, the questions are literally in there.
- Count and sort. Usually five to eight topics cover 70 to 80% of the volume. That is your pilot scope; the rest goes to a person for now.
Human handover and tone-of-voice rules
Good AI customer service is not defined by how much the bot can do, but by how gracefully it lets go. These are the handover rules we set up by default:
- The customer asks for a person. Always honour it.
- Emotion or complaint. Words like "ridiculous", "never again" or "lawyer": hand over immediately with a summary.
- Money above a threshold. Refunds or goodwill above an amount you choose are decided by a person.
- No answer twice. Then the bot stops trying and creates a ticket.
- Legal, medical or contractual. Out of scope, always.
A handover is only good when it is warm: the colleague receives the summary, the customer details and the reason, so the customer does not have to repeat their story. With a voice agent that means transferring with a short briefing, or a callback request with context in your CRM.
What the bot must never do
- Deviate from your tone: you choose formal or informal address and a few example sentences in your style.
- Promise delivery times or prices it cannot verify in your system.
- Invent discounts, interpret terms or make diagnoses.
- Make things up. "I do not know that, I will pass you on" is a perfectly fine answer.
- Pretend to be human. We will come back to that under the AI Act.
Which KPIs to track
Four numbers are enough to know whether it is working. The target ranges are indications from our projects; your starting point determines what is achievable.
| KPI | What it measures | Target range (indication) |
|---|---|---|
| First-response time | How quickly the customer gets a first substantive answer | AI channels: under a minute. Human follow-up after handover: within 4 working hours |
| Resolution rate | Share of conversations the AI completes on its own | 40 to 50% in month 1, growing to 60 to 70% after 3 months of tuning |
| Handover percentage | Share of conversations that go to a person, and why | 20 to 40% at the start. "Customer asked for a person" is fine; "bot did not understand" should go down |
| CSAT | Customer satisfaction, one question after the conversation (thumbs or 1 to 5) | At least 80% positive or an average of 4 out of 5. Lower? Read the negative conversations and adjust the answers |
In the first month, schedule half an hour a week: reread conversations, flag wrong answers, add to the documentation. That is the real work of automating customer service, and that is where most of the gains are.
Numbers from our own projects
No market research, but results from projects we built ourselves for Dutch SMBs:
The first two numbers come from a voice agent project in which the agent took over intake, in more than 40 languages. The two hours a day from the Outlook agent with 95% classification accuracy. The €180,000 from an aftersales project in which 42% of dormant customers became active again. And in a document project 95% of manual entry disappeared, with 98% OCR accuracy.
What does AI customer service cost? With a worked example
Our prices are on the pricing page; this is the short version. An AI chatbot starts at €2,450 one-time plus €249 per month, including setup, training on your own data, website integration and support. An AI voice agent starts at €3,950 one-time plus €349 per month, with call costs of about €0.05 per minute charged separately. On top of that, expect €50 to €200 per month in third-party usage costs, depending on your volume. A pilot of 1 to 2 weeks costs €1,500 to €3,000 and is deducted from the project. Projects with multiple channels and systems fall under enterprise, from €7,500 and from €499 per month. Everything comes with a 30-day money-back guarantee.
Worked example: webshop with 40 tickets a day
Suppose: a webshop with 40 customer questions a day via chat and e-mail, six days a week. This is a worked example with assumptions, not a promise; replace them with your own numbers.
Starting points (assumptions)
AI chatbot costs, year 1
Effect per month
Two caveats. Those 62 hours a month are rarely cut; in practice they go to returns, reviews and calling customers back, the aftersales work that currently falls by the wayside. And the biggest gain is often not in the hours but in what you no longer miss: the question at 9:30 PM that would otherwise have been a lost order.
GDPR and the EU AI Act: be honest that it is AI
A chatbot or voice agent processes personal data: names, e-mail addresses, order numbers, sometimes a photo of a defective product. That puts you under the GDPR. This is the minimum you arrange:
- Data processing agreement with your supplier and clarity about where the data lives. We run on EU servers.
- Data minimisation: the bot only asks for what it needs.
- Retention period: agree how long conversations are kept.
- No third-party training on your customer conversations, unless you deliberately opt in.
- Update your privacy statement: state that you use AI and for what. The Dutch Data Protection Authority (Autoriteit Persoonsgegevens) publishes practical guidance for businesses.
The transparency duty in the EU AI Act. Since August 2026, people communicating with an AI system must be informed of that, unless it is already obvious from the context. Concretely: your chatbot introduces itself as a digital assistant and your voice agent says at the start of the call that the caller is speaking with an AI assistant. It costs you nothing in conversion and prevents a customer feeling deceived.
Implementation in 4 steps: from zero to live in 2 to 4 weeks
Inventory and pilot scope (day 1 to 3)
The 20/80 exercise. Together we choose the channel and the five to eight topics, and agree handover rules and tone of voice.
Build and train on your data (week 1 to 2)
The bot learns from your documentation and product info. Where needed we connect your webshop, CRM, helpdesk or calendar.
Testing and soft launch (week 2 to 3)
First internally with real questions from your mailbox, then live for part of your visitors or outside office hours.
Fully live and weekly tuning (week 3 to 4)
All traffic through the bot, with a dashboard for the four KPIs. In the first month we review the conversations together weekly and extend the scope.
Still in doubt? Then start with steps 1 and 2 as a 1 to 2 week pilot (€1,500 to €3,000, deducted from the project). You see the bot working on your own questions before you decide further.
Common concerns (and what is true about them)
"My customers want to talk to a human"
Some do, and they get one. But most customers mainly want their problem solved, preferably now. Someone who wants to know at 10 PM whether their parcel arrives tomorrow is happy with a good answer from a bot, not with "we respond within two working days". It is not human or AI: standard questions to AI, so your people have time for the rest.
"AI makes mistakes"
True. That is why you limit the scope, give the bot only your documentation as a source and set hard limits on money, contracts and complaints. A mistake shows up in the dashboard and takes a few minutes to correct in the documentation. A wrong answer from a new employee you only hear about when the customer complains.
"It is too expensive for my business"
A chatbot costs around €5,400 in setup and subscription in the first year, plus usage costs. A full-time service employee costs, as an indication, €35,000 to €50,000 a year including employer costs. If the bot handles a third of the work, the sum is already positive. Tight on budget? Start with the pilot, or with the channel that eats the most time.
When AI customer service is not a fit
We regularly say no, or "not yet", too. In these situations automating customer service is not the first step:
Too little volume
Fewer than a handful of questions a day? Then you will not recoup the investment in hours.
Still worthwhile: a voice agent if you miss many calls outside office hours. Then it is about lost revenue, not hours.
Every question is bespoke
Legal advice, complex B2B quotes, medical questions: the answer is different every time and a mistake is too costly.
Alternative: AI does the intake and preparation, the person gives the answer.
Nothing is written down
If your procedures only exist in one colleague's head, the bot has nothing to learn from.
Do first: the 20/80 exercise and half a page per topic.
The team is not on board
If your team sees the bot as a replacement rather than help, it does not get tuned and fades within a few months.
What works: let the team choose the topics and write the answers themselves.
Frequently asked questions about AI customer service
What does AI customer service cost for an SMB?
At Utomatic an AI chatbot starts at €2,450 one-time plus €249 per month, including setup, training on your own data, website integration and support. An AI voice agent starts at €3,950 one-time plus €349 per month; call costs (about €0.05 per minute) are separate. On top of that, expect €50 to €200 per month in third-party usage costs.
What percentage of customer questions can AI handle on its own?
It depends on how repetitive your questions are. As an indication: where a handful of topics (order status, opening hours, returns, appointments) make up most of the volume, 50 to 70% handled independently after the first months is realistic. For bespoke advice questions it is lower.
Do I have to tell customers they are talking to AI?
Yes. The EU AI Act contains a transparency duty: people must know they are communicating with an AI system, unless that is already obvious from the context. It is also sensible: customers respond more calmly to a bot that says it is a bot and hands over honestly when it does not know something.
How long does it take to automate customer service?
A pilot takes 1 to 2 weeks and costs €1,500 to €3,000, deducted from the full project. The complete process, from inventory to live, usually takes 2 to 4 weeks. The lead time mostly depends on how quickly you supply your frequently asked questions, procedures and system access.
What happens when the AI does not know the answer?
Then it says so and hands over: for a chatbot as a ticket or e-mail to a colleague, with a summary; for a voice agent by transferring or logging a callback request. You set the rules up front, so the bot never guesses on questions about money, contracts or complaints.
Can the AI connect to my webshop, CRM or accounting system?
Usually, yes. Platforms like Shopify, WooCommerce, HubSpot, Pipedrive, Exact and Microsoft 365 have an API that lets the bot look up order status, schedule appointments or create a ticket. Without an integration we start with the questions that can be answered without one and build the rest later. Everything runs on EU servers, GDPR-compliant, with a 30-day money-back guarantee.
Start with AI customer service
Try the chatbot demo first, or book a 30-minute call: we look at your mailbox and call log together and build a pilot in 1 to 2 weeks. Prefer direct contact? Call +31 85 004 1088, send a WhatsApp or book a slot here.
View chatbot demo AI voice agent
