JLL: Real estate leader introduces ML services to uncover investment opportunities and deliver rent prediction and valuation

Qwak was brought aboard to enhance JLL's already impressive machine learning operation and enable delivery of complex models faster and more efficiently. The main requirement was to scale model delivery to production and enhance model accuracy without increasing engineering efforts.

JLL, a leader in the industry with a commendable machine learning operation, constantly strives for innovation and improvement. To bolster its capabilities further, they partnered with Qwak. The primary objective of this collaboration was clear: streamline the process of delivering complex models, boost their accuracy, and achieve these advancements without expanding the engineering workload. Dive into this case study to explore the challenges faced, strategies employed, and the outcomes of this exciting partnership.

About JLL

JLL, a global leader in real estate services, pioneers technology-driven innovation to transform the industry. The company invests in diverse real estate assets and serves clients across various sectors, aligning with a long-term commitment to benefit its people, clients, and communities.

Industry

Real Estate, PropTech

Use Case

Real estate investment opportunities
Rent prediction via deep learning and text classification
Portfolio predictions via Automated valuation models (AVM) ‍

Model Frameworks

XGBoost
Pytorch
scikit-learn

Qwak's platform batch inference is significantly faster than other solutions we used in the past. Considering we have hundreds of these a day, the increase in speed allowed us to save hundreds of thousands of dollars.

Or Hiltch
VP Engineering
10X

Faster Batch Inference

A/B

Model Scenario Testing

↓$

100Ks savings yearly

Challenges

  1. Manual Model Training and Tracking: Model training and the process of tracking experiments demanded extensive manual intervention. JLL urgently required an integrated platform for streamlined tracking of model experiments.
  2. Inefficient Batch Inference: Inference batch requests had to be executed hundreds of times daily. Each request's long processing time hampered the pace of development projects.
  3. Lack of Infrastructure for Online Models: As the team ventured into creating new models necessitating online serving, JLL found itself without the requisite infrastructure to support these models on a large scale.

Solutions

  1. Manual Model Training and Tracking: Model training and the process of tracking experiments demanded extensive manual intervention. JLL urgently required an integrated platform for streamlined tracking of model experiments.
  2. Inefficient Batch Inference: Inference batch requests had to be executed hundreds of times daily. Each request's long processing time hampered the pace of development projects.
  3. Lack of Infrastructure for Online Models: As the team ventured into creating new models necessitating online serving, JLL found itself without the requisite infrastructure to support these models on a large scale.