Empowering Commercial Real Estate Professionals with AI

VTS is the commercial real-estate leasing and asset-management platform: 12+ billion square feet of commercial real estate is managed on it. The VTS data-science team had already built promising ML models. Getting them in front of customers was the hard part.

01 The Challenge

Research models were ready; production deployment was not

The data scientists at VTS were fluent in building and validating ML models in research environments. Moving a model into the production platform was a different discipline — one that required weeks of manual work per deployment and specialized infrastructure expertise the team didn’t want to grow inside data science.

The leadership wanted a foundation that let data scientists ship models without depending on a specialist on every deployment. Reduce manual work. Minimize errors. Make iteration cheap.

02 The Approach

Build one model’s path to production. Make that path reusable for every future model.

Provectus reviewed VTS’s ML prototype-to-production process and identified the friction points. Rather than building a one-off deployment for the first model — a leasing-deal-outcome predictor — the team designed a reusable template-based platform that would serve every future model the company shipped.

The first target was concrete: get the leasing model into production on the new platform, using real customer data, within three months.

03 The Build

Two template suites plus a VTS-specific SageMaker SDK

The first template suite standardizes data preparation, processing, and model training on Amazon SageMaker. A custom SDK extends SageMaker for VTS’s environment, reducing repetitive code and hiding infrastructure complexity. Data scientists work on models, not deployment mechanics.

The second template suite connects workflow steps into automated pipelines for model training, batch processing, and real-time serving — integrated with the broader AWS environment. Together, the templates and SDK give data scientists a clear, repeatable path from prototype to production.

Documentation and knowledge-transfer sessions came with the handover. The VTS engineering team owns the platform and extends it.

04 The Results

3x faster deployment. A foundation every future model reuses.

3x

ML model deployment velocity

Measured against the prior manual process

The leasing-deal-outcome model shipped in three months. More consequential: every ML application VTS builds from here starts on the same substrate. The data-science team spends its time on model building, experimentation, and validation — not on deployment plumbing.

VTS is moving closer to its stated mission: becoming the commercial-real-estate industry’s definitive decision-making platform, with AI-powered features shipped at the pace of the product roadmap.

05 What’s Next

The platform carries the next set of AI-powered features

The template-based foundation is the substrate VTS’s data-science team builds against going forward. The engagement with Provectus continues as new AI use cases adopt the pattern.