People who visit your website read. They look at what you do, how you do it and whether that fits their question. But they rarely reach out spontaneously. The step from “this sounds interesting” to “I’ll fill in the contact form” is bigger than it seems.
At BlazeForce, we saw this pattern in our own analytics. Good traffic, reasonable time on page, but low conversion to actual contact. The problem was not the offering. It was the threshold.
That was the starting point for an AI intake assistant on our own website: a chatbot that responds to visitors immediately, takes their question seriously and, when there is a concrete request for help, converts the conversation into a qualified contact moment.
From visitor to qualified contact: an AI assistant as an intake channel on your own website. Photo: Unsplash
Why we built it ourselves
The market for chat widgets is not small. There are dozens of SaaS solutions you can add to your website in a few minutes. Yet we chose to build our own WordPress plugin.
The reason is practical. Off-the-shelf widgets are generic, carry a monthly subscription fee and give you little control over the prompt logic, GDPR compliance, data storage and the connection with your own systems. For a firm that helps clients with exactly this kind of custom work, it felt inconsistent to rely on a third party whose internal workings we don’t know.
We also wanted to experience the tool ourselves before introducing it to clients. The best way to learn what works is to run it on your own website and see the results.
How the assistant works
The core of the plugin is a conversation interface available on every page of the website. A visitor can ask a question directly without filling in a form or leaving their name. The assistant responds based on a carefully crafted prompt that describes BlazeForce as a company: what we do, for whom, and which questions we can answer.
Two technical choices make the experience concretely better than a generic chat widget.
The first is page context. The assistant knows which page a visitor is on and tailors its answers accordingly. Someone on the Salesforce services page receives different emphasis than someone on an article about integrations. That makes the conversation more relevant and the threshold lower.
The second is language detection. The website is bilingual. The assistant automatically detects which language the visitor is using and switches along. If someone types in English, the assistant replies in English. If that same person switches to Dutch halfway through, the assistant follows. No configuration required, no dropdown.
From conversation to contact moment
The assistant is not built to converse endlessly. The goal is to quickly understand whether a visitor has a concrete question that BlazeForce can help with and, if so, make the step to contact as small as possible.
When purchase intent or a direct request for help is detected, the interface switches to a contact form. That form asks only the minimum: name, email address and optionally a preferred time for a call. Everything the visitor has already said in the chat does not need to be repeated.
At the moment the visitor submits the form, the plugin automatically generates a concise AI summary of the conversation. That summary is sent along with the lead, so the person following up can see at a glance what was discussed, what question was asked and where the visitor was in their orientation. Not a cold conversation, but a prepared follow-up.
Direct connection to Nextcloud CRM
Leads that come in through the assistant are created directly in our Nextcloud CRM. No email notification disappearing into an inbox, no manual data entry. The lead record contains all relevant fields: name, email, phone number if provided, source page, language and the AI summary of the conversation.
The integration is built on the same Nextcloud connector we deploy for clients. In practice, this means a new lead is visible in our CRM within seconds, complete with the context needed for a targeted first response.
Feedback and quality management
An AI assistant is never perfect from day one. Answers can be too generic, the tone may be slightly off or a question may fall outside the scope of the prompt. To monitor and adjust this, the plugin includes a feedback module.
After each answer, a visitor can indicate whether the answer was useful. That feedback, together with the corresponding question and answer, is stored in the WordPress dashboard. This shows us exactly which questions are answered well and where the assistant falls short. We use those insights to periodically refine the prompt.
The dashboard also shows the basic metrics: the number of conversations, messages, leads and handoffs to direct contact. Simple, but enough to assess whether the tool is doing its job.
GDPR and responsible use
Privacy was a deliberate design choice, not an afterthought. The plugin requests active consent before a lead is saved. The API key for the AI model is stored server-side and is never visible in the browser. Summaries are stored with a configurable retention period. No full transcripts are stored, only compact summaries for quality management, and even that is optional.
Rate limiting at IP level prevents misuse. All input is sanitised before it reaches the server. The consent text is configurable per language, so the GDPR wording matches the website on which the plugin runs.
What it delivers
The most important outcome is not technical. It is that visitors take the step towards contact earlier, because the threshold is lower. Asking a question in a chat window feels different from filling in a form. It is more informal, faster and feels less definitive.
The leads that come in through the assistant are also better qualified than a generic contact form. Because a conversation has already taken place, the person following up knows what the question is, what the visitor has already understood and which service or theme is relevant. That makes the first conversation more concrete and shorter.
For BlazeForce, this is also our own reference. We can show clients who are interested in a similar solution how it works, not as a mockup, but as a live system on our own website.
When is this relevant for other organisations?
An AI intake assistant is not the right instrument for every website. It works best when a service or product requires some explanation before someone makes contact, when visitors regularly ask the same types of questions, and when the conversion from visitor to lead lags behind the quality of the offering.
For organisations already working with Nextcloud, Salesforce or another CRM, the connection is relatively straightforward to build. The plugin is configurable without code changes: the prompt variables, language behaviour, contact flow and privacy texts are all adjustable via a WordPress admin screen.
Interested in a similar solution for your website? BlazeForce builds AI assistants that align with your services, your CRM and your tone of voice. Contact us for a no-obligation conversation.