challenge

Operations managers were making key decisions blindly. "How many people came last night?" – this question had no precise answer. Without an understanding of real traffic, it was impossible to plan staff properly. Marketing budgets were spent without being tied to real demand. The difficulty was not the lack of cameras – they are present in each of the 150+ venues. The problem was that some of them were outdated: low resolution, compression artifacts, unstable lighting. Ready-made market solutions for visitor counting simply did not provide a reliable result on such material. A vicious circle: analytics are impossible without replacing cameras → replacing 150+ cameras is expensive and will take months → analytics are postponed again.

Why now?

!FEST was actively building an analytical ecosystem: dashboards, reports, automation of operational decisions. The lack of traffic data turned from an inconvenience into a real blind spot – the single metric missing for a complete operational picture. Postponing further meant building analytics without a foundation.

Fears and how we addressed them

The main concern was whether the accuracy would be acceptable at all on outdated equipment. More practical questions were also raised: "Won't we end up having to change the cameras anyway – just later?" and "How do we scale the solution to 150+ locations with different lighting, viewing angles, and stream intensity?" We did not bypass these risks – we made them the starting point of the design. Instead of adapting a ready-made algorithm to the existing equipment, we framed the task differently: low-quality video is not a limitation, but an initial condition.

solution

The gloozd team developed its own Computer Vision algorithm from scratch – designed specifically to work with low-quality video. The algorithm stably detects and counts visitors despite compression artifacts, variable lighting, and low resolution. No new hardware equipment – exclusively a software solution that connects to the already existing cameras. The system is deployed across 150+ venues of the holding with a heterogeneous infrastructure. Real-time traffic data is integrated into the general analytical ecosystem of !FEST and displayed in operational dashboards alongside other business indicators.

result

• The visitor counting accuracy reached over 95% under standard conditions – at the level of or higher than solutions requiring new equipment.

• At the same time, not a single camera was replaced: the savings in time and deployment costs turned out to be substantial.

• The operations management of !FEST received systematic and reliable traffic data across the entire network of venues for the first time.

• Decisions regarding staffing, marketing campaigns, and logistics are now based on real figures – not on intuition.