Your visitors know their problem, but not your offer structure. The offer finder understands the intent, matches it to the right offer and routes to the right page or straight to a booked call - instead of answering questions nobody asks.
The assistant in the bottom right of this page is the same offer finder we build for you - trained on our own portfolio. Describe a project in your own words and see what happens.
Opens the assistant in the bottom right. No sign-up. Only the current session is kept in your browser; the classification itself is logged for quality assurance.
For example:
Anyone selling services that need explaining has a structural translation problem: the prospect thinks in terms of their situation, your website is sorted by your services. The bounce rate lives in between.
Nobody searches for managed operations. They think: our servers run, but nobody is watching them. Navigation cannot make that translation.
Full-text search finds words, not intentions. Without the right term you get either nothing or everything.
Someone still unsure whether you are even the right fit will not fill in a form. They go back to the search engine.
They answer the question that was asked - and, when in doubt, invent a page that does not exist. That costs trust instead of building it.
The visitor describes their project in their own words. No menu tree, no form.
The system maps the intent to your offer structure - based on your content, not free model imagination.
The visitor gets the matching offer, the right page and the next step - up to a booked call or a qualified lead with context.
| Standard chatbot | Offer finder | |
|---|---|---|
| Interaction | Visitor must know what to ask | Visitor describes the intent, the finder guides |
| Result | A text answer | Matching offer, right page, next step |
| Links | Hallucinated or none | Curated catalogue - only existing pages |
| Leads | Form without context | Lead with a summary of the intent |
| Hosting | US SaaS, OpenAI | EU infrastructure, European model (Mistral) |
An example from our own finder - you get the same mechanism with your portfolio.
Visitor
We have three sites with their own servers. The colleague who looked after them is retiring.
The finder recognises
Taking over operations without an internal team, distributed infrastructure, time pressure from a departure - a handover case, not a greenfield build.
The finder routes to
Managed operations as the matching service, the server audit page as a sensible first step, and a booking suggestion - all from the curated catalogue, no invented links.
You receive
A lead with the project summarised, before the first conversation happens.
Instead of a name and a message you get the project in the prospect's own words, the finder's classification and the path they took.
The analytics show which projects visitors describe - often in words that appear nowhere on your site. That is raw material for your copy and your portfolio.
When the finder repeatedly finds nothing suitable, that is not a system failure but a finding about your portfolio.
The finder is embedded as a widget. No relaunch, no CMS rebuild, no migration.
European AI model, operated on WZ-IT infrastructure or on your own - no data to US providers.
The finder answers from your pages and your offer catalogue. What does not exist is never linked.
Labelled as AI, with an honest way out: if the match feels wrong, one click reaches a human - exactly what the EU AI Act transparency duty requires for AI interactions.
It runs here in beta, and we say so under every classification. We know these limits from running it ourselves - you should know them before the decision, not after.
On ambiguous requests it occasionally picks the wrong building block. That is why every classification carries a route to a human and the statement that the classification can be off. A finder without that way out is more dangerous than none.
It does not invent services. If a project matches nothing in your catalogue, it says so instead of proposing something similar. That feels like a gap at first - it is the function.
It classifies and leads to a conversation. Binding figures come from you, not from a language model that does not know the specifics of the case.
It brings the enquirer into the conversation with the right background. A human still holds the first conversation - the finder makes sure it does not start from zero.
Whatever is outdated on your site, the finder will reliably carry forward. It makes your content maintenance more visible, not unnecessary.
Come back tomorrow and you start over. The finder does not recognise a returning visitor and cannot pick up an earlier conversation.
Put three projects into one paragraph and you get a classification that fits one of them and grazes the others. This cannot be configured away; the limit is in the nature of the task.
It classifies on substance, not commercially. Whether a project suits you is your call in the conversation - the finder does not make that judgement for you.
The finder is not always the answer. Here is when it is.
| Full-text search | Contact form | Standard chatbot | Offer finder | |
|---|---|---|---|---|
| Understands a project in the visitor's own words | No, finds words | No, collects text | Partly | Yes |
| Leads to a concrete offering | To a result list | To an inbox | To an answer | To named building blocks |
| Says so when nothing fits | Zero results | No | Rarely - prefers to invent | Yes, with a route to a human |
| Provides context for the first conversation | No | Whatever the visitor typed | The chat history | Project, classification, path |
| Reveals gaps in your own offering | Through search terms with no hits | No | No | Yes, as analytics |
| Effort | Low | Very low | Low | Medium - catalogue and integration |
If your offering is three clear services, the contact form wins. The finder pays off from the point where visitors can no longer tell which of twenty services solves their problem.
The question comes late in the project and should come early. The legal position moved in 2026.
On 12 May 2026 the Higher Regional Court of Hamm ruled (case I-4 UKl 3/25) that a chatbot's output is attributable to the company as its own communication: the bot is a technical means of the operator, not a third party. The case decided an injunction claim under unfair competition law - the operator therefore cannot fall back on the argument that a model chose the wording.
Important for context: the judgment is not final. The court allowed an appeal on points of law to the Federal Court of Justice. Anyone planning today should assume the line holds - but not pretend it has been confirmed by the highest instance. We are not legal advisers; this is orientation for your architecture decision.
The finder classifies and links. It quotes no prices, no deadlines, no assurances - precisely the statements from which claims arise.
What is not on your site is not asserted. That limits the output to what you stand behind anyway.
One click reaches a human. That satisfies the transparency duty and is also where a misunderstanding surfaces before it gets expensive.
If someone claims the bot promised something, you can look it up. Without a log it is one word against another.
The figure comes after a look at your site. These five points determine it:
One-off: building the catalogue, integration, test cases, fitting it into your site design. Ongoing: operations, model and security updates, catalogue maintenance, analysis of the enquiries.
Self-hosting on your side, CRM integration and multilingual setups as extensions - fixed price after the intro call.
Answers to the most important questions
Builder bots answer questions about your website - the visitor has to know what to ask. The offer finder reverses that: it understands the intent, maps it to your offer structure and actively routes to the right page or a booked call. Only pages from your curated catalogue are ever linked.
From your own content: during setup we model your offer map and build the knowledge base from your pages. That is deliberately consulting work - it is exactly where self-service tools fail with complex portfolios.
Yes. The finder runs on European infrastructure with a European model (Mistral), fully self-hosted on your side if preferred. No data flows to US providers.
Qualified requests become leads with context: the finder summarizes the intent and hands it over with contact details - via email, CRM or booking. You see what the visitor wanted before the first conversation.
Typically two to four weeks from kickoff to go-live: model the offer map, build the knowledge base, test the integration, go live. Your website does not need a rebuild - the finder is embedded as a widget.
The finder fits wherever visitors know their problem but not your offer structure: IT service providers, agencies, engineering firms, consultancies, training providers. In the intro call we check honestly whether your portfolio has enough structure - if a simple contact form is all you need, we will say so.
Runs where you want
No risk: worst case, you leave with a clearer understanding of your project than before.


“WZ-IT's advice on our Azure migration was technically sound and completely non-binding right from the intro call - we took away a great deal.”
ml&s speaks about integrating a local AI solution. The other voices cover architecture, data sovereignty and operations - exactly the maturity an AI project needs to reach production.
Whether a specific IT challenge or just an idea - we look forward to the exchange. In a brief conversation, we'll evaluate together if and how your project fits with WZ-IT.
Timo Wevelsiep & Robin Zins
Managing Directors of WZ-IT
