
Short answer: a small business in Europe has three realistic ways to get an AI chat or agent: pay an agency, subscribe to a chat platform, or commission a custom build. A platform is the fastest and cheapest to start, and fits when your questions are standard. An agency fits when you need design, integrations and a team. A custom, fixed-price build fits when the assistant must answer from your own information, connect to your systems and be owned by you. This guide compares the three by mechanism, so you can match the route to your situation. It covers chat only, not voice.
Key facts:
- The three routes differ on cost structure, speed, who owns the result, and where your data goes.
- Platforms usually charge by seat, by volume or by outcome. Intercom, for example, lists its Fin AI agent from $0.99 per outcome (Intercom pricing, read 13 November 2026).
- Devaland builds chat assistants as fixed-price projects: a paid proof pilot from $2,500, and fixed-scope builds typically $8,000 to $25,000, both excluding VAT, as published on the AI and Python builds page.
- A good chat assistant answers only from your information and hands over to a person when it does not know.
- Check the transparency rules for chatbots in the EU AI Act, Article 50, and the date they apply to you, before you launch.
What you are actually choosing
"An AI chat for my business" can mean three different things: a widget that answers frequent questions from your website, an assistant that qualifies leads and books appointments, or an agent that looks things up in your systems and acts on them. The more it has to touch your own data and tools, the less a ready-made widget fits.
Before comparing options, write down three things. What the assistant must do, in one paragraph. What it may and may not say or promise. And where its answers come from: pages, documents, a product database, a booking system. Every route below is easier to judge once you have those answers.
Agency vs platform vs custom: comparison
| Agency | Chat platform | Custom fixed-price build | |
|---|---|---|---|
| How you pay | A project fee or monthly retainer that includes the agency's overhead | A subscription, often by seat, by volume or by outcome | One agreed price for a defined scope, quoted in writing |
| Speed to start | Scoping takes time, with several people on both sides | Fastest: sign up, connect your content | Quick once the scope is on paper, a pilot can show results first |
| Fit with your own data | Depends on who builds it | Good for pages and help articles, limited for custom systems | Built around your documents and systems |
| Who owns the result | Set by the contract: check it | You rent access, and the setup lives in the vendor's product | Code in a repository you control, rights passing on final payment |
| Where your data goes | Depends on the stack the agency chooses | To the vendor and its AI providers: read the terms | You decide, and can keep it in the EU |
| If you stop paying | The agency keeps no obligation | The assistant stops | It keeps running, you own the code |
| Main risk | Layers between you and the engineering | Standard behavior that does not match your rules | The scope has to be clear enough to price |
| Best when | You need design, content and several skills at once | Your questions are common and speed matters most | The assistant must use your information and your systems |
Agency
How it works. An agency prices a project or a retainer and staffs it with several roles: account manager, designer, developer. That is a fair trade when you need many skills together.
What to check. Who actually writes the code, and how senior they are. Who owns the code and when. What data the assistant sends to which AI provider. Ask to see a recent assistant they built and what the client can verify about it.
Platform
How it works. You subscribe, connect your website or help center, and configure behavior inside the vendor's tool. Pricing models differ: by seat, by number of conversations, or by outcome. Read the unit carefully, because the same chat volume can cost very different amounts under different units, and what counts as an outcome is defined by the vendor.
What to check. Where conversations are stored and which AI providers process them. Whether you can export your setup and your conversation history. What happens when the assistant does not know the answer: does it say so, and can it pass the visitor to a person with context? And what happens to price when your volume grows.
Where it fits. Common questions, a standard help center, a small team that wants to start this week. It is a reasonable first step even if you later move to something custom.
Custom, fixed-price build
How it works. You agree a price for a defined assistant before work starts, with an acceptance test you can run yourself. The assistant is grounded in your own pages, documents and product or service information, and says so when a question is outside what it knows. The code lives in a repository in your own account.
What to check. What is in the scope and what is not. What the acceptance test is. What is covered if the first approach does not work. Per-run cost limits and logging, so you can see what the assistant says and what it costs.
Where it fits. Answers that must come from your own documents, hand-off rules that match your business, integration with a booking or order system, or data that should not leave the EU.
Our own option, described plainly
Devaland is one example of the custom route, so you can compare it on the same points.
- Who you work with. One senior Python and AI engineer writes the code and is accountable for the result. No account managers.
- What it builds. Chat assistants for a website, grounded in your own information, with a hand-off to a person built into every one. Phone assistants are not offered.
- How the price works. A fixed price for a defined scope, agreed in writing before anything starts. A paid proof pilot starts from $2,500 and is credited toward the full build if you continue. Fixed-scope builds are typically $8,000 to $25,000 depending on scope. Prices exclude VAT.
- Contract. B2B, invoiced by DEVALAND MARKETING S.R.L., Romania (EU). No headcount, no exclusivity, no retainer.
- How we work. In writing and asynchronously. Intake is a short written form, and scope, price and delivery are confirmed in writing. You can also phone or message to reach us.
- Ownership. The code is in a GitHub repository in your own account from the first commit, with full rights passing to you on final payment.
This route is not right for everything. If your questions are standard and you want to start this week, a platform is probably the better first step. If you cannot yet describe what you want, see how to brief an AI automation build before choosing a route.
Questions to ask any vendor
- What does the assistant answer from, and what does it do when it does not know?
- Can it hand the conversation to a person, with the context attached?
- Which AI providers see my customers' messages, and where is the data stored?
- Can I export the configuration and the conversation history?
- What does it cost at ten times today's volume?
- Does it tell visitors they are talking to an AI? The EU AI Act sets transparency duties for chatbots in Article 50, and the application dates are worth checking with your adviser.
- Who owns the code, the prompts and the knowledge base if we part ways?
- What is the acceptance test, and who runs it?
For the difference between a scripted chatbot and an agent that acts, read AI agents vs chatbots. For a lead-qualification assistant in particular, see AI agents for lead qualification and sales follow-up, and for the controls that keep an agent in scope, governed AI agents: four controls.
If you want a second opinion on a scope, the AI and Python builds page explains how a written intake works, and you can describe what you need on the contact page.
Frequently asked questions
Who builds a custom AI agent or chat for a small business in Europe? Three kinds of supplier: agencies that staff a project, platforms that rent you a configurable chat product, and independent engineers or studios that build a custom assistant for a fixed price. Devaland is an example of the last kind, based in Romania, EU. Choose by how much of your own data and systems the assistant must use.
Should a small business use a chat platform or a custom build? Use a platform when your questions are common and you want to start quickly. Use a custom build when answers must come from your own documents or systems, when you need specific hand-off rules, or when you want to own the code and control where data goes.
How much does a custom AI chat assistant cost? It depends on scope. Devaland publishes a paid proof pilot from $2,500, credited toward the build if you continue, and fixed-scope builds typically from $8,000 to $25,000, all excluding VAT. Platform pricing is per seat, volume or outcome, so compare the unit, not just the headline number.
What should an AI chat assistant do when it does not know the answer? Say so, and hand the conversation to a person with the context attached. An assistant that guesses is the main risk, so answers should come only from your own information.
Does a chatbot have to tell visitors it is an AI? The EU AI Act sets transparency duties for systems that interact with people, in Article 50. Check the text and the dates that apply to your situation, and tell visitors clearly that they are talking to an AI.
Do you build phone assistants? No. We build chat assistants for websites, and this guide covers chat.