OUR CLIENT
One of our clients is a leading B2B event management company, founded in 2009 and operating across six European markets – including the UK, Germany, Spain, and Poland.
With a portfolio of over 200 industry-leading events spanning sectors such as Technology, Healthcare, HR, Education, Learning, and Transportation & Infrastructure, they are one of the most active players in connecting professionals, communities, and brands across Europe.
Their business model is built around end-to-end event delivery. From the moment a client expresses interest in hosting an event, the company takes the reins – co-creating the concept, curating the guest list, coordinating exhibitors, and assembling a full program of keynotes, panels, and networking sessions. The result is a rich, community-driven experience that brings together decision-makers, thought leaders, and industry professionals under one roof.
THE CHALLENGE
At the core of every event our client runs is a lead-matching system – and optimizing it is where we come in. The goal is simple: connect the right exhibitors with the right attendees, on both sides of the interaction, as quickly and accurately as possible.
How it works on the ground. Every attendee and exhibitor receives a scannable badge. Whenever the two meet on the event floor – whether the interaction is initiated by the attendee or the exhibitor – either party can scan the other’s badge, creating a digital record of the touchpoint. This scanning data becomes one of the key inputs that enriches the matching engine’s output.
Two layers of data. On the exhibitor side, we work with structured company information: industry vertical, product and service offerings, and event presence details. On the attendee side, the data is richer and more granular. Before each event, attendees complete a detailed registration form – sharing their company name, professional role (e.g., nurse, surgeon, or other specialist), whether they hold decision-making authority, and whether they have an active budget. Together, these two layers form the foundation of the matching logic.
A Medallion Architecture pipeline. All incoming data – from registration forms and post-event badge scans – is processed through a three-layer Medallion Architecture. In the Bronze Layer, raw data is stored exactly as received. It then moves to the Silver Layer, where it is cleaned and standardized into a consistent format across all events. Finally, in the Gold Layer, the data is consolidated into global tables for both visitors and exhibitors, enriched with AI-generated profile descriptions and vector embeddings – and made ready for matching.
One of the elements of the project was loading data from external data providers using their API mechanisms. The implementation of this part utilized a metadata-driven approach, which significantly simplified the data loading procedures and pipelines, which were fully parameterized. Thanks to this, adding new data areas that could be delivered via API required only adding new entries to the metadata tables, without interfering with other procedures or pipelines.
The data was incrementally downloaded to a landing layer for initial storage. After preliminary cleaning, it was transferred to the Staging layer, and then to the pre-analytical layer, where the first normalization took place.
Turning noise into signal. A significant portion of the incoming data is unstructured: free-text company descriptions, open-ended form answers, and qualitative responses that resist simple categorization. To bridge this gap, we use a large language model (LLM) to synthesize all available inputs for each party – form answers, role descriptions, company details – into a single, coherent profile. Each profile is then converted into a dense vector embedding, giving the system a numerical representation it can actually compare.
Matching at scale. Once all profiles are embedded, the matching engine applies max cosine similarity to identify the most relevant exhibitor-attendee pairs. Attendees receive a curated list of exhibitors worth visiting. Exhibitors, in turn, get a ranked list of attendees – further enriched by real-time scanning data collected on the event floor.
OUR APPROACH
Before our involvement, the matching engine relied entirely on manually configured mapping files – one set of rules per event, built by hand. The process was slow, resource-intensive, and entirely dependent on deep domain expertise. That dependency was the core problem.
Industry-specific nuances are notoriously difficult to encode. The logic behind a truly good match is often counterintuitive. Consider the veterinary sector: a specialist focused on large animal care would, in most cases, have no interest in pet care products. Yet in smaller rural communities, a single veterinarian may be the only local provider – covering both livestock and domestic animals. That makes them a highly relevant match for both categories. These context-dependent relationships are nearly impossible to anticipate through static rules, and they shift from one region and industry to the next.
As the client’s event portfolio expanded across Technology, Healthcare, HR, Education, and more, manually mapping rules for every event simply stopped being viable. The challenge became twofold: automate the matching logic entirely, and build a system robust enough to generalize across industries – delivering accurate, relevant recommendations without requiring event-by-event manual intervention.
THE RESULTS
The recommendation algorithm helps each attendee make the most of their time at the event – by directing them straight to the exhibitors that are actually relevant to them. Instead of browsing through hundreds of companies with no clear fit, an attendee receives a shortlist of around 10 exhibitors worth visiting. They go directly where the potential is. No wasted time, no guesswork.
This efficiency translates into tangible outcomes on both sides. Relevant interactions lead to real business results: partnerships, transactions, and long-term professional relationships. The event stops being just a gathering and becomes a genuinely productive experience – one that delivers measurable value for attendees and exhibitors alike.
Quality held steady – and that’s the point. Despite the significant reduction in setup time and human oversight, recommendation accuracy remained on par with the previous, manually configured approach. Maintaining match quality while automating a process that was deeply reliant on domain expertise is, in itself, a major milestone. It validates the core architectural bet: that a well-designed embedding-based system can generalize effectively across industries without hand-tuning.
A stronger brand impression. The effective, relevant experience on the event floor has not gone unnoticed. Both attendees and exhibitors have responded positively to the quality of the matching experience. As a result, the overall perception of our client’s technological capabilities – and, by extension, the quality of their events – has grown considerably.
Scaling is next. With the system proven in its current form, the architecture is actively being prepared to onboard new industries and event verticals. The Medallion pipeline and embedding-based matching were designed from the ground up to be industry-agnostic. The goal is straightforward: bring the same quality and automation to the client’s entire portfolio of 200+ events – across all sectors, with minimal manual overhead.
At the end of the day, the key to a successful B2B event is simple: the right people need to find each other.
That’s exactly what we make possible.
Ready to Scale Your AI Projects?
If capacity limits are slowing down your AI and Data Science journey, our expert team is here to help.
Let’s connect – schedule a meeting and see how Machine Learning and AI can unlock new opportunities for your organization.