Expert answers to technical questions
The chatbot answers product and technical questions in real time, 24 hours a day, including outside the sales team's office hours.
CASE STUDY · AI AGENTS
An AI chatbot that acts like an experienced salesperson
Response speed decides whether a B2B customer stays on the site or leaves for a competitor. For Železiarne Podbrezová we built an AI chatbot that navigates the assortment of steel pipes like an experienced salesperson: available around the clock, in many languages, and from the first conversation ready to turn a visitor into a scored sales inquiry.
Železiarne Podbrezová is a major manufacturer and seller of steel pipes with international reach. They operate in the B2B segment where product specifications are complex and the buying cycle is long: a customer typically needs to verify exact dimensions, material properties and availability before deciding to contact sales.
- Industry
- Metallurgy / B2B manufacturing
- Deployment
- 7 product domains + corporate website
- Model
- Claude Sonnet 4.5
- Cooperation
- since March 2026
Challenge
Inquiries were leaving elsewhere, salespeople got unspecified requests.
Železiarne Podbrezová manage an extensive catalogue of metallurgical material across multiple markets and language versions of their website. Potential clients come looking for specific product information and if they cannot find it immediately, they go elsewhere. Navigating the catalogue requires industry knowledge a website alone cannot offer, so the sales team often received inquiries that were not sufficiently specified and had to spend time on initial clarification instead of closing deals.
ProblemExpectations
An extensive catalogue across markets and languages, navigation required industry knowledge
Searching pipes by technical parameters directly in the chat
Without an immediate answer, visitors left for competitors
An expert answer instantly, 24/7, in the customer's language
Salespeople took over unspecified inquiries and lost time clarifying
Complete, scored inquiries instead of cold contacts
An anonymous visitor without contact details was lost
Turning visitors into scored sales opportunities
How we solved it
Intelligent search across the assortment
Instead of manually browsing the catalogue, the customer can have the chat find relevant steel pipes directly by the parameters they need.
Lead qualification and handover
The chatbot does not wait passively for a contact: it links the conversation with the visitor's contact details and scores how relevant and ready the inquiry is. It first guides the customer through specifying the inquiry and only then collects contact information, so real business interest can be judged objectively.
Agile development and gradual scaling
Development has run since March 2026 and the pilot phase closes in August 2026. The solution evolved iteratively and was gradually rolled out to further language versions and the client's foreign websites, extending coverage to new markets without building parallel support from scratch.
Anthropic, Claude Sonnet 4.5
Almost 81% of visitors who opened the chat actively joined the conversation. Salespeople no longer work with cold contacts: they receive scored inquiries including an overview of what the customer specifically asked for.
Michal GregušINLOGIO
We originally deployed a cheaper, less capable language model. With technical terminology and especially multilingual communication it did not deliver accurate enough answers, so during pilot operation we switched to Claude Sonnet 4.5. One consistent model across all language versions gave the best ratio of answer quality, reliability and speed.
For technical specifications such as dimensions, wall thicknesses and material grades we deliberately did not let the AI guess. We built a separate structured database from the client's materials. When the original data export turned out to contain a wrong value across hundreds of products, it let us locate and fix the error at the source instead of debugging AI behaviour.
with website visitors, 5,324 messages exchanged
.3 s
instead of hours or days of waiting for a salesperson
with a score of 80+ handed to the sales team for immediate processing
Project outputs
- An AI chatbot available 24/7 in many languages, running simultaneously on seven product domains and the corporate website, each with its own authorization; the domain's country sets the language, welcome text and window visuals
- A separate structured database of product data instead of AI guesswork: the chatbot draws exact values of dimensions, wall thicknesses and grades
- A custom analytics dashboard: conversation history, automatic lead scoring (cold/warm/hot), an overview of the most frequently recommended products and their development over time
- Technical attachments (drawings, PDF specifications, Excel sheets) are flagged by the chatbot as a priority hot lead and forwarded to the sales department for manual processing
- An admin console with the client's supplementary instructions under consistency control, plus AI cost monitoring with automatic alerts when limits are exceeded
The chatbot serves foreign customers directly on their local websites in their language and the handover of qualified leads to the sales team runs smoothly. The pilot phase closes in August 2026, to be followed by deeper catalogue integration and further language and market expansions including a planned deployment on the client's subsidiary sites.
Want similar results?
Look at your processes through the eyes of automation. Tell us what slows you down, or book a 30-minute call right away.
Free
Find a use case for your business
Enter what your company does and AI generates 5 concrete ideas where automation saves you the most time. It takes a minute.
Generate 5 AI ideas
Ambitious teams
trust us
"We see Invelity as a team of professionals who are always willing and able to help quickly with solutions across many areas. Today we cannot imagine our digital marketing without this partnership."
Ready to streamline
your workflow?
Tell us about your process. We will help identify what can be automated and where to gain efficiency.
Let's go!