How AI and computer vision transformed rent calculations for 35,000 homes

The challenge: outdated plans, inconsistent data, and time pressure

The shift to the new rent law presented multiple challenges. “We didn’t have exact data on which homes had balconies or how large they were,” said Mark Rossen, IT consultant at Woonin. “Manually inspecting 35,000 properties wasn’t feasible.” Adding to the complexity, Woonin’s floor plans were often outdated and inconsistent. Terminology like “balcony,” “roof terrace,” and “loggia” was used interchangeably, and many plans lacked measurements altogether.

Time pressure added another layer of difficulty. “The law was announced just four days before implementation,” Mark continued. “We didn’t just see this as a technical problem—it was a moral responsibility. As a nonprofit, it’s critical to ensure our tenants aren’t overcharged unfairly.”

The solution: AI and computer vision to the rescue

Passionned Group stepped in with a cutting-edge approach to automate the process of measuring outdoor spaces. Together with Longhow Lam, they developed a four-step AI-driven workflow:

• Converting floor plans to images: Using Python, floor plans in PDF format were converted to image files (PNG), making them easier to analyze with Artificial Intelligence tools.
• Text recognition with OCR: Using Tesseract OCR, the system identified keywords such as “balcony” and “loggia” within the plans.
• Pixel analysis with computer vision: OpenCV, a computer vision library, was deployed to identify balcony contours and count pixels. These were then converted to square meters using reference spaces like bedrooms.
• Cloud computing for speed: A virtual machine in the Google Cloud with 60 cores and 240GB of RAM enabled rapid parallel processing, cutting analysis time from 30 hours to just one hour.

Mark was impressed by the team’s adaptability. “Initially, we used kitchens as a reference point, but their cluttered layouts created issues. Switching to bedrooms was a game-changer, and the team implemented this adjustment seamlessly.”

Figure 1: An annotated floor plan showcasing the AI-calculated balcony area of 5.3 m², using a 13.1 m² bedroom as a reference point for precise measurements.

The solution was refined over three iterations. A pilot run analyzed 500 homes, incorporating feedback to enhance accuracy. Once the process was perfected, the full dataset of 35,000 homes was analyzed.

The results: precision, efficiency, and inspiration

Partnering with Passionned Group brought transformative results for Woonin. “This solution saved us weeks of manual labor,” Mark reported. The AI-driven process produced reliable data to comply with the new rent law, with only a small percentage of plans requiring manual checks due to poor quality. By combining AI results with Woonin’s internal datasets, the team achieved high data quality and fair rent calculations.

We now have data quality that is reliable enough to correctly apply the new rent law.

Longhow Lam emphasized the project’s focus on practical problem-solving. “We used straightforward data science, avoiding unnecessary generative AI hype, to deliver impactful results.” The scalability of cloud computing also impressed Woonin. “The ability to scale capacity so quickly was eye-opening,” Mark said. “It’s inspired us to explore more ways to leverage data analytics.”