A chatbot that qualifies leads, invoice extraction, CV screening: real AI use cases with their costs and their limits, no science fiction.
Beyond the buzz, tools that do real work
Artificial intelligence is everywhere in the talk and rarely in the facts. My job isn't to sell dreams: it's to integrate AI building blocks into your website or internal tools that save measurable hours. Everything that follows genuinely exists — most of these examples are running on this very site right now, and you can try them.
An assistant that sells while you sleep
In the bottom right corner of this site, there's a chat bubble. The assistant inside knows my services, my prices and my past projects — and only that: it's designed to never invent a price or answer off-topic. But its real job is sales: when a visitor describes their project, the assistant asks the right questions, offers to take an email address, and records a complete lead file that reaches me by notification. A visitor passing through at 11pm gets an immediate answer in their own language, and I wake up to a qualified lead instead of an anonymous visit in my analytics. That's the difference between a brochure site and a site that works for you.
Your documents enter themselves
Second use case, more discreet and often more profitable: document extraction. You drop in an invoice, a quote or a purchase order — as a PDF or even a photo — and the AI extracts the key information: amounts, dates, references, contact details, ready to be exported to your spreadsheet or your management software. For a business that re-keys twenty supplier invoices a week, the maths is quick. The demo is freely available on this site's automation page: drop in a document and time it.
CV screening that shows its reasoning
Another tool I offer as a demo: you describe a role, you drop in a stack of CVs, and you receive a reasoned ranking — strengths, gaps, a match score for each candidate. Note my position on this: the tool prepares the decision, it doesn't make it. It turns three hours of first-pass screening into ten minutes of reading a summary, and that alone is huge. The final decision remains human, and must stay that way.
What it really costs
It's the question everyone asks, so let's talk numbers. Today's AI models are pay-per-use, and the fast models I use for these tools cost a few cents for dozens of requests. Concretely: a chatbot answering three hundred visitors in a month costs less than a coffee. The real cost is the integration — connecting the AI to your data, defining its scope, building the interface. It's a one-off investment, starting at around €1,500 for a simple building block, and the ongoing usage cost afterwards is marginal.
The guardrails I always put in place
A poorly integrated AI can say anything, and it's your reputation that takes the hit. On every integration, I install the same protections: the assistant only answers within its scope and politely declines the rest, all visitor input is validated before processing, and rate limits stop a joker from running up your bill in a loop. On the data side, documents analysed in my demos are not kept. These precautions don't show in a sales demo — which is precisely why I show them.
When I advise against AI
Sometimes the right answer is a well-designed form, not a chatbot. If you receive three enquiries a week, a qualification AI won't do you any favours: it adds a layer of complexity for a volume that's perfectly manageable by hand. My rule is simple: the integration must pay for itself in hours saved or clients won within six months. If I can't defend it honestly, I'll tell you, and we put that money elsewhere — into SEO or content, for instance, where the return is often more reliable.
Try it for yourself
Rather than taking my word for it: all these demos are freely available on the automation page, and the assistant is waiting for you in the bottom right corner. If one of these tools would save your team hours, describe your situation to me — I'll tell you honestly what's worth it, and what isn't.