An AI chatbot is worth it when you get the same five questions every day and the answer already exists somewhere in writing. It is not worth it when every question is different, or when answering requires judgement.
A modern chatbot is not a tree of buttons. It reads your own content and answers in sentences — but only within the limits you set.
Not sure whether it fits?
Send us the ten questions you get most and we will tell you honestly whether you need a chatbot or simply a better FAQ page.
The three types — and which suits you
| Type | How it works | Cost | Suits |
|---|---|---|---|
| Rules / menu | Predefined questions with buttons | Low | Few, stable questions |
| AI over your own content | Reads your pages and documents, then answers | Medium | Many variations of the same question |
| Hybrid with human handover | AI for the easy ones, a person for the rest | Medium-high | E-shops and services with volume |
The second type is what people mean today by “AI chatbot”. It does not invent answers — it draws on whatever you give it.
The third is the safest for a business: it keeps a human exactly where one is genuinely needed.
What it can and cannot do
| It can | It cannot |
|---|---|
| Opening hours, addresses, payment methods | Negotiate a price |
| Order status, if connected to your system | Decide on an exception to your policy |
| Product availability and specifications | Handle an angry customer |
| Returns policy and warranties | Give legal or tax advice |
| Collect details and pass them to you | Replace selling in a complex service |
The rule: a chatbot answers what already has an answer. If the answer is not written down anywhere, it will not find it.
And one important distinction: collecting details and pushing them into your CRM is among the most profitable uses — see how a CRM works.
When it is worth it — and when it is not
It is worth it when:
- You get more than 30 messages a week with repeating content.
- Your customers ask outside business hours, in the evenings and at weekends.
- You sell products with many variants and questions about size or compatibility.
- You lose orders because you are slow to reply.
It is not worth it when:
- You get two or three questions a week. A good FAQ page costs a tenth as much.
- You sell a complex service where the first conversation is the sale.
- You have no written policies. The chatbot will mirror the gap.
- There is nobody to pick up what it cannot cover.
The third case is the most common. Before building a chatbot, write the answers — and you often discover that was enough.
What does it cost?
| Component | Indicative range |
|---|---|
| Tool subscription | €20 – €150 a month |
| Setup and knowledge base | €400 – €2,000 |
| Connection to your e-shop or CRM | €300 – €1,500 |
| Maintenance and improvement | €50 – €200 a month |
The payback calculation: if you answer 40 messages a week at three minutes each, that is roughly 8 hours a month. At a €15 blended hourly cost, that is €120 a month — enough to justify a mid-sized build within the first year.
Want the numbers for your case?
Tell us how many messages you get and we will send a cost and payback estimate.
How to set it up properly, in five steps
1. Collect the real questions
Take your last 100 messages from DMs, email and phone. Group them. You will find that 70% are five to eight questions.
2. Write the knowledge base
For each question, a short answer in your own words. This is the actual work — not the technology.
3. Set limits
State explicitly what it does not answer: negotiated pricing, complaints, legal matters. Anything outside goes to a person.
4. Build the human handover
A “talk to a person” button visible from the start. Outside hours, the bot collects details and promises a reply with a stated timeframe.
5. Test with real messages
Run the 100 messages from step 1 through the bot and see how many it answers correctly. Below 80%, work on the knowledge base before going live.
The six mistakes that make a chatbot annoying
- It opens by itself after two seconds and covers the content.
- It hides the human. If it takes five messages to reach a person, the customer leaves.
- It replies “I did not understand” with no alternative.
- It promises things that are not true, because the knowledge base is out of date.
- It does not say it is a bot. Transparency builds trust and is good practice.
- Nobody reads the transcripts. That is where the questions you failed to cover are hiding.
Is it safe for my data?
It depends what you feed it. Three rules cover most situations.
- Do not put customer personal data in the knowledge base. That is for policies and product information, not your customer list.
- Check where conversations are stored and for how long.
- Record the tool in your processing activity register and sign a data processing agreement.
If the chatbot asks for a name and email, the same rules apply as to any form — see our guide to GDPR and your website.
How to measure whether it works
| Indicator | What it shows | A good sign |
|---|---|---|
| Self-resolution rate | How many conversations close without a human | 50%–70% |
| First response time | How quickly the customer gets an answer | Seconds |
| Conversations that became leads | How many contacts it captured | Varies by sector |
| Unanswered questions | What is missing from the knowledge base | Should be falling |
The fourth is the most useful. Every question it could not answer is free market research — and often new content for your website.
In an e-shop, a chatbot also affects checkout: answering shipping and returns questions at the right moment reduces cart abandonment.
Frequently asked questions
Will it replace my customer service?
No. It takes the repetitive work and leaves the difficult cases to people.
How quickly can it be set up?
A simple chatbot in one or two weeks. With e-shop or CRM integration, three to six.
Does it work in Greek?
Yes, modern tools handle Greek comfortably. Test it with real, misspelled questions before going live.
Can I put it on Instagram too?
Yes, most tools connect to Messenger and Instagram. Note that the platforms have their own rules for automated messages.
What if it gives a wrong answer?
That is what limits and human review in the first month are for. After that, you sample-check.
The first step
Before looking at tools, collect your last 50 customer messages and count how many are the same question.
If it is more than half, you have a candidate. If not, write a good FAQ page and count again in a month.
The wider picture, with the other nine uses that pay off, is in our guide to AI automations.
Next step
Book a free 30-minute call with the team at The Dev Alley. We will look at your messages and tell you whether a chatbot is the right answer — or something cheaper is.