AI chatbot · for website and customer service
Want an AI chatbot? Start with the question, not the bot
An AI chatbot on your website can help customers at night with order status, delivery times and returns. But often a chatbot isn't the first thing you need. Here you'll read when a chatbot works, when an assistant or agent is the smarter choice, and how NinA builds it: with answers from your own knowledge and systems, a person for the exceptions and data in the EU.
- Answers from your own knowledge and systems
- Always a route to a person
- Data in the EU
Chat on the website
livelooks up order 4471 in Shopify
fetches the tracking details from the carrier
reads the returns policy in the knowledge base
creates a second-line ticket in the help desk
01 / Sound familiar?
Why so many chatbots make customers angry rather than happy.
A chatbot is quick to buy and quick to put online. It goes wrong after that: the bot doesn't know the answer, makes one up, or sends the customer round in circles until they pick up the phone. Then you have an extra channel, and no less work.
Going round in circles
“I typed 'speak to someone' three times and got the link to the FAQ three times.”
A made-up answer
“The bot promised a customer free shipping abroad. We don't offer that at all.”
Knows nothing about the order
“It explains perfectly how returns work. But where my parcel is, it doesn't know. And that was exactly my question.”
Nobody keeps it up to date
“Our chatbot still gives last year's opening hours.”
The work is somewhere else
“The questions in the chat are easy. The real work is in the two hundred emails a day that come in behind it.”
Because the competitor had one
“We had to do something with AI too. What exactly the chatbot was supposed to solve, nobody really knew.”
02 / The difference
Chatbot, assistant or agent: what's the difference?
A chatbot talks to your customer. An assistant helps your employee. An agent does work in your systems. All three run on the same kind of AI model, but they solve different problems. So choose the problem first, then the form. The full explanation is in what is an AI agent?
| AI chatbot | AI assistant | AI agent | |
|---|---|---|---|
| What it does | Answers customer questions in a chat window | Helps an employee with writing, searching and summarising | Carries out tasks in your systems: looking up, creating, preparing |
| Who talks to it | Your customer, on your website | Your employee, in their own work | Often nobody: it runs in the background, a person approves |
| Knows your business | Only what you put into it | Yes: products, house style and way of working | Yes, and it looks into your systems |
| Does something in your systems | · | · | ✓ |
| Biggest risk | A wrong answer going straight to the customer | Small: the employee reads everything | An error in your systems, hence a person in the loop |
| Who makes it | Have it built, or a tool with an AI feature | Your own team (build layer 2) | NinA builds (build layers 3 and 4) |
| Suits | Lots of the same questions with one clear answer | Knowledge work where quality and tone matter | High volumes with steps across several systems |
Swipe sideways to see all columns →
The boundaries overlap. A good AI chatbot for customers is often an agent under the hood: it looks in your systems first and then answers.
03 / When it works
When does an AI chatbot work for your business?
An AI chatbot works when three things are in place: lots of the same questions come in, the answer is recorded somewhere and correct, and there's a clear line where a person takes over. If one is missing, the chatbot becomes an extra channel you have to monitor.
1. Lots of repeat questions
2. A knowledge base that's right
3. A clear line with people
01Volume
Do the same questions come back dozens of times every week?
02Knowledge
Is the answer in one place, up to date and correct?
03Systems
Can the bot look up the order, the delivery or the appointment itself?
04Boundary
Has it been agreed when a person takes over, and how quickly?
04 / When not
When you're better off without a chatbot (for now).
Sometimes a chatbot simply isn't the best first step. We'd rather say so now than after the build.
The work is behind the chat
Does most of the work come in through email, forms and phone? Then AI behind your team delivers more than a bot in front of it: preparing answers that an employee checks and sends. That's how AI agents work in customer service.
The knowledge isn't written down
If the answer is mostly in Pete's head, no chatbot can find it. Start with the knowledge base then. You'll benefit from it anyway, even without a bot: new colleagues get up to speed faster and Pete can go on holiday.
Every question is custom
Advice, quotes, complaints with emotion or money involved: that's where a customer wants a person. AI can help that person, with a summary and a draft answer. That's an assistant or agent, not a chatbot.
Nobody is going to maintain it
Products, prices and policies change. A chatbot nobody maintains becomes a risk after a few months. Agree up front who owns it and who updates the knowledge base.
05 / Old bot, new bot
From a chatbot with buttons to a chatbot that knows.
Six moments in a conversation with a customer. On the left the classic chatbot with fixed buttons and scripts, on the right an AI chatbot as we build it: connected to your knowledge and systems, with a person on hand.
Where is my parcel?
Classic chatbot
An 'Order' button and then a link to the tracking page. Work it out yourself.
AI chatbot by NinA
Looks up the order in your online store and the status with the carrier, and says where the parcel is now.
A question that's slightly different
Classic chatbot
There's no button for that question, so: 'Sorry, I don't understand.'
AI chatbot by NinA
Understands the question in plain language and looks for the answer in your own knowledge base, with the source alongside.
An angry customer
Classic chatbot
The same three buttons, until the customer gives up or picks up the phone.
AI chatbot by NinA
Recognises the tone, doesn't try to solve it alone and passes the customer to an employee, with the conversation attached.
A customer from Germany
Classic chatbot
Dutch only, or a separate bot per language that nobody maintains.
AI chatbot by NinA
Answers in the customer's language, from the same knowledge base.
A question about money
Classic chatbot
The bot promises something nobody agreed. Or says nothing at all.
AI chatbot by NinA
If it's about compensation, a discount or an exception, a person decides. The bot says honestly that it's passing it on.
After a month
Classic chatbot
Nobody knows what happens in the chat.
AI chatbot by NinA
You see which questions come up most, which answers are missing and where customers drop off. That way the knowledge base grows along with it.
Our approach
The best chatbot starts behind your team. Not in front of it.
We usually build the AI that helps your employees first: preparing answers that they check and send. For each type of question you then see how often the answer is right first time. Whatever goes well for weeks on end can then move into the chat. That way you know the bot says the right thing before a customer talks to it.
06 / How NinA builds it
How we build an AI chatbot that gets it right.
We don't build standalone chatbots. We build within the AI Partnership, and that always starts with the AI Quick Scan: first know what it delivers, then build. That way you avoid a bot nobody needed.
- 01
AI Quick Scan
1 day, results within a week
We look at where the questions really come in, how many there are and what they cost. Sometimes a chatbot is the best step. Sometimes it's something else, and then we'll say so.
- 02
Knowledge and systems
The first weeks
We put the knowledge in one place and connect the systems that hold the answers: online store, ERP, CRM or help desk. The bot only answers from your own sources.
- 03
Behind the team first
The test phase
The AI prepares answers for your employees. They check, improve and send. That way you see per type of question how often the answer is right first time.
- 04
Live in the chat
When the numbers add up
Questions that have proven to go well, the chatbot answers itself. The rest goes to a person, with the conversation attached. The customer always sees they're talking to AI.
- 05
Measure and maintain
Every month
We look at what's being asked, what's missing and where it goes wrong, and update the knowledge. What's live falls under a maintenance contract, even if the partnership stops.
07 / Safe and honest
Where does your customers' data go, and who's in charge?
A chatbot talks to your customers. So you need to know where their data goes and who is responsible for what the bot says. With us, four things are fixed before anything goes live.
What we build runs in the EU: on your own infrastructure or on servers with a European provider. Azure OpenAI is also possible. We work to ISO 27001 and NEN 7510; certification is in preparation.
The bot answers from your own knowledge base and systems, not from the general memory of a language model. If the answer isn't there, it says so and hands over. That reduces the chance that it makes something up.
Personal data in a chat falls under the GDPR, and the AI Act has rules too. On the legal details, a lawyer has the final say.
01Data in the EU
Your own infrastructure or a European provider. Azure OpenAI is possible.
02Your own sources
Answers from your knowledge base and systems, with the source alongside.
03Human in the loop
Exceptions, complaints and anything involving money go to an employee.
04No lock-in
Code, data and documentation are yours. You take them with you within 30 days if you stop.
Worked with us
08 / In practice
What AI delivers in customer service.
Two clients, the same lesson twice: AI works best with a person involved. One prepares answers for the team, the other gives customers a first answer day and night.
09 / Costs
What does it cost to have an AI chatbot built?
We don't build standalone chatbots for a fixed price. A chatbot that works depends on your knowledge, your systems and your team, and needs maintenance. That's why we build within the AI Partnership, after the Quick Scan. All amounts exclude VAT. Everything side by side is on pricing.
Know first
AI Quick Scan: € 6,000
One-off. One day on site, and within a week a business case per opportunity and a twelve-month roadmap. Also available on its own, and the outcome is yours.
Build
AI Partnership: from € 4,500 per month
Package S, M or L: € 4,500, € 7,500 or € 12,000 per month, minimum 12 months. Guarantee: the partnership as a whole delivers at least 2x the investment in the first year, or we don't start.
Keep it running
Maintenance: from € 245 per month
Bronze € 245 on your own servers. Silver € 445 or Gold € 895 on our platform, on servers in Germany with a European provider. Including AI usage, annually in advance.
10 / Questions
Frequently asked questions about an AI chatbot.
What does it cost to have an AI chatbot built?
At NinA we don't build chatbots as a standalone project, but within the AI Partnership. That starts with the AI Quick Scan at € 6,000 one-off. The partnership costs € 4,500, € 7,500 or € 12,000 per month, for a minimum of twelve months. What's live, we keep running with a maintenance contract from € 245 per month. All amounts exclude VAT. The guarantee: the partnership as a whole delivers at least twice the investment in the first year, or we don't start.
What's the difference between an AI chatbot and an AI agent?
An AI chatbot holds a conversation and answers questions. An AI agent carries out tasks in your systems: it looks up an order, creates a ticket or prepares an answer for an employee. A chatbot talks, an agent acts. A good customer service chatbot is often an agent under the hood, because it looks in your systems before it answers. More in what is an AI agent?
Can I build a chatbot with ChatGPT for my website myself?
For internal use, your team can easily build an assistant with ChatGPT, Claude or Copilot, filled with your own documents. A chatbot that talks to customers on your website takes more: connections to your online store or ERP, agreements on personal data, a line where a person takes over, and someone who keeps the knowledge up to date. That last one is what most home-built chatbots forget.
How do you stop an AI chatbot from making things up?
By having it answer only from your own knowledge base and systems, with the source alongside, and having it say it doesn't know when the answer isn't there. On top of that, we test behind the team first: employees check every answer, so you see per type of question how often it's right. Only what has proven to go well, the bot answers by itself. Exceptions and questions about money always go to a person.
Where does my customers' data stay?
In the EU. What NinA builds runs on your own infrastructure or on servers with a European provider, and Azure OpenAI is possible. We work to ISO 27001 and NEN 7510; certification is in preparation. Code, data and documentation are yours. Personal data in a chat falls under the GDPR.
Do you also build just a chatbot, without a partnership?
No. We no longer do standalone builds, because a chatbot that works needs more than a build project: knowledge in order, integrations, measuring and maintaining. We always start with the AI Quick Scan, and then build within the AI Partnership. If that's too big for now, a workshop can be a good first step: your team then learns to build assistants for their own work.
Do I have to tell customers they're talking to AI?
We always do. It's honest, and a customer then also understands why they're passed to a colleague with a difficult question. Anyone who'd rather speak to a person straight away gets a person. On what the AI Act says about this, and the other legal details, a lawyer has the final say.
When is a chatbot not the best choice?
When most of the work comes in through email, forms or phone, when the knowledge isn't written down anywhere yet, when almost every question is custom or when nobody is going to maintain the bot. Then AI behind your team usually delivers more: preparing answers that an employee checks and sends. You can see how that works in the enterprise customer service case on our cases page.
Further reading
- AI in e-commerce and retailCustomer questions, returns and product copy for online stores, with a person who approves.
- AI AgentsDigital colleagues that do work in your systems, with a human in the loop.
- What is an AI agent?The difference between chatbot, assistant, workflow and agent, without the hype.
First find out whether a chatbot pays off
In a fifteen-minute intro call we look at where your questions come in. Then you know whether a chatbot, an assistant or an agent is the best first step.
01
Intro call
Fifteen minutes, no strings. Where does work get stuck and what do you want to achieve?
02
AI Quick Scan
One day on site. Within a week, a business case and a roadmap.
03
AI Partnership
A dedicated team that builds, trains and shows what it delivers every month.
160+ organizations · Sessions rated 9.3 · Based in Amsterdam