Investment managers have an excellent opportunity to re-design their investment management practices using AI to achieve investment outcomes that are often influenced by market uncertainty, data proliferation, and rapid advances in finance and technology.
This article will provide a practical guide to developing and deploying machine learning capabilities, highlighting the feature engineering process, model selection trade-offs, and machine learning operations (MLOps) best practices necessary to operationalize AI at scale.
AI AND WORKFLOWS
The traditional process for modeling investment decisions among most real estate professionals relies on linear relationships, such as those between rent growth and vacancy, often overlooking the complexity caused by the heterogeneity of real estate markets and other key factors. However, using AI to identify and leverage pattern recognition from complex, non-linear relationships offer deeper insights into markets.
It is important to note that while more advanced algorithms enhance accuracy, they often reduce interpretability, whereas simpler algorithms improve interpretability at the expense of accuracy. The tradeoff between accuracy and interpretability can be mitigated using (real estate) domain knowledge, feature engineering, and Explainable AI (XAI) techniques.
AI technologies—Machine Learning (ML), Natural Language Processing (NLP), and Generative AI (GenAI)—are transforming commercial real estate by enabling more accurate market predictions, risk assessments, and opportunity identification. ML models, such as classifiers and regressions, enhance credit risk evaluation and scenario analysis by quantifying the impact of economic factors on rent growth and cap rates. With sufficient data and computing power, Deep Learning uncovers patterns in complex datasets like financial news, market data, and social media. NLP extracts sentiment and insights from unstructured text, while GenAI not only interprets data but also generates reports, market updates, and multimedia content, improving communication and decision-making across the investment lifecycle.
In modernizing our CRE life cycle management process, our firm experimented with transitioning a couple of risk-related processes from traditional rule-based approaches for estimating loan default probabilities to an AI-driven risk assessment system. The following section offers a broad overview of our AI-driven framework, which utilizes sophisticated classifier models that combine domain expertise with advanced statistical methods. This approach generates risk signals that reflect broader economic and capital market conditions, significantly enhancing the reliability and accuracy of our underwriting process and providing a competitive edge in the CRE investment landscape.
MODEL DEVELOPMENT WORKFLOW OVERVIEW
The process of deriving an investment strategy from data follows a well-established pattern (as shown in Exhibit 1), it involves the collection, ingestion, and organization of data from capital market indicators, offering memorandums, financial statements, and images into a processed stage that could be turned into signals. A signal refers to meaningful information or patterns extracted from data that could help build Machine Learning models and generate predictions. The generated predictions can be used for decision-making and creating strategies to forecast future trends, detect anomalies, and predict customer behavior, which is especially useful in domains like commercial real estate, where the data ecosystem tends to be fragmented.
The feature engineering process for generating signals utilizes real estate domain knowledge and statistics, often involving advanced mathematical techniques, to transform the processed data. This makes it easier for machine learning algorithms to learn patterns that are useful in modeling the desired outcome. The source data could include loan details, economic factors, and key performance indicators (KPIs) specific to real estate markets.
These are transformed to create new features or to adjust the scale of existing ones, preventing the model from being biased toward features with larger values. It’s crucial to handle data types properly, examine bias in the data, and apply suitable sampling strategies to improve signal quality and model performance. Relying on solid real estate domain knowledge is essential to maintain the right balance between model interpretability and accuracy.
Exhibit 1: Data transformation: from raw data to strategy
Source: Authors
ESTIMATING PROBABILITY OF DEFAULT IN COMMERCIAL REAL ESTATE WITH MACHINE LEARNING
As an example of the modeling framework (Exhibit 2), we utilized a diverse set of signals from inputs, such as debt yield, loan-to-value, unemployment rate, and occupancy, to estimate the probability of default of loans in our commercial mortgage portfolio. To enhance accuracy and mitigate blind spots inherent in any single model, we employed an ensembling technique.
With this approach, our quantitative researchers combined multiple models, each with unique strengths and weaknesses, to contribute distinct insights and integrate their outputs using a meta-learner, providing a more robust and comprehensive estimate for decision-making.
MACHINE LEARNING OPERATIONS
To ensure that our machine learning framework was reliable, scalable, and maintainable in a production environment, our quantitative developers relied on a concept called MLOps. It consists of a set of practices that automate and streamline ML workflows from start to finish, often utilizing cloud infrastructure.
Exhibit 3 illustrates the MLOps workflow, starting with data storage and management as the foundational layer. Then, the modeling process involves model training, validation, and deployment, supported by continuous integration and continuous development (CI/CD) practices. This ensures updates are smoothly integrated and deployed. The workflow also includes an evaluation and feedback loop to keep improving model performance, along with a model registry and prediction serving layer to manage model versions and deliver predictions reliably. MLOps is essential for generating results that are reproducible and auditable, which is vital for maintaining regulatory compliance and building trust.
The use of AI on a scale in developing investment strategies, improving decision-making, and evaluating and managing risks does not come without its challenges. Successfully scaling AI within CRE hinges on careful consideration of several key factors. Our experience indicates key factors include the importance of selecting the right capabilities through systems engineering and strategic alignment, having the proper support for change management, and finding the right partners for technology implementation.
SYSTEMS ENGINEERING AND STRATEGIC ALIGNMENT
The first step in scaling AI in CRE firms is to select the appropriate capabilities aligned with strategic business objectives. Systems engineering plays a critical role in this process, as it involves a structured approach to integrating complex systems and ensuring that all components work well together. Early on, crowdsourcing can be valuable in generating innovative ideas or talking about specific challenges, but systems engineering provides a more structured, comprehensive, and reliable approach to deploying AI solutions. The latter’s focus on integration, quality control, risk management, and compliance make it a superior method for ensuring AI systems are effective, scalable, and aligned with the organization’s strategic objectives.
Strategic alignment ensures that AI initiatives are technologically feasible and relevant to the firm’s long-term objectives. This aligns closely with the Cross Industry Standard for Data Mining (CRISP-DM), which is widely recognized for guiding data-related projects and emphasizes the importance of understanding the business context as the first step.
CHANGE MANAGEMENT
With the 2025 Artificial Intelligence Index report from Stanford University showing that public unease around AI remains widespread, with only 55% of people globally seeing “AI products and services as more beneficial than harmful,” it is essential to recognize that resistance from staff is natural and can hinder AI adoption. Effective change management strategies, including transparent communication about the benefits of AI, employee engagement in the implementation process, and ongoing support from top management, are a must to increase buy-in. Our experience demonstrates that developing AI systems requires a highly collaborative and iterative approach. This approach fundamentally reshapes organizational workflows and operating models. As such, having the right capacity, skills, and alignment is vital to ensure success.
STRATEGIC PARTNERSHIPS
Better collaboration and partnership will be required to increase the ability of technology firms and startups to meet the needs of CRE firms and to close the gap in understanding between the two. This will continue adding to innovation and will foster sharing of research across businesses. The shared learnings will lead to reduced costs and will accelerate development. Partnerships will also enhance communication, giving access to a pool of talent specialized in navigating complex AI projects.
A MAJOR SHIFT
The adoption of AI in Commercial Real Estate marks a major shift from traditional, rule-based investment methods to advanced AI-driven frameworks. By using machine learning, investment managers can understand complex, non-linear market relationships and gain deeper insights from large and scattered datasets. Using advanced techniques, like ensembling and feature engineering guided by strong domain knowledge, enables firms to produce more accurate, reliable, and interpretable predictions of key investment indicators, including loan default risks and rent growth forecasts. Additionally, applying best practices in MLOps ensures that AI solutions are scalable, easy to maintain, and meet regulatory standards, thereby boosting trust and transparency. However, successfully scaling AI in CRE requires careful systems engineering aligned strategically with business goals, effective change management to overcome internal resistance, and strategic partnerships to fill knowledge gaps and foster innovation. By addressing these factors thoroughly, investment management firms can maximize AI’s benefits, leading to better decision-making, increased operational efficiency, and a competitive edge in an evolving CRE landscape.
MORE FROM SUMMIT #19
PLATFORM SPONSOR
ASSOCIATE SPONSOR
Note from the Editor: Issue #19
Mid-Year Pulse: Global Investors on the Economy, Energy, and Housing
Benjamin van Loon | AFIRE
Bend, Not Break: Investing in Real Estate Amid Economic Uncertainty
John Murray + François Trausch + Russell Gannaway + Kirill Zavodov | PIMCO
Test Of Time: How US Real Estate Withstands an Uncertain Investment Market
Riaz Cassum | JLL
Oh Canada: Why International Investors Might Want a Second Look at Canadian Real Estate Markets
Amy Erixon + Long Tang + Daniel Goldberg + Marie-France Benoit | Avison Young
Beyond Oil: Why Gulf Family Offices Are Doubling Down on US Logistics, Data, and Housing
Abbas Hashmi | Saudi Family Holdings
South of the Border: Higher Yields and Growth in Mexican Industrial
Shaun Libou | Raymond James
Migration Myth: NIMBYism and Why Coastal Movers Aren’t Affecting Sunbelt Housing Supply
Donal Warde | Consultant + Ron Bekkerman | Constellation Data Labs
Machine Center: The AI-Driven Transformation of Data Center Investment
Sam Chandan | Chen Institute for Global Real Estate, NYU Stern School of Business
AI in Commercial Real Estate: A Practical Guide for Industry Leaders
Armel Traore Dit Nignan + Shaarvani Kavula | Principal Real Estate
Artificial Control: Are You Ready for AI Management and Oversight?
Marie-Noelle Brisson + Michael Savoie | CyberReady, LLC
Capitalizing on Dynamics: Demographic Mega-Trends Impacting CRE
Stewart Rubin | New York Life Real Estate Investors
Mid-Cap Assets: An Under-Examined Segment in CRE
Asaf Rosenheim | Profimex
The Lifestyle Renter: A Growing Opportunity for Strategic Investment in High-Quality Apartment Communities
Hannah Waldman | The Dermot Company
The Climate Is Speaking: CRE Underwriting for a Future That’s Listening
Ines Diez + Thomas Stanchak | Stoneweg
+ EDITOR’S NOTE
+ ALL ARTICLES
+ PAST ISSUES
+ LEADERSHIP
+ POLICIES
+ GUIDELINES
+ MEDIA KIT
+ CONTACT
NOTES
—
ABOUT THE AUTHOR
Armel Traore Dit Nignan is Head of Real Estate Data and Analytics for Principal Asset Management. Shaarvani Kavula is a Quantitative Developer for Principal Asset Management.
PLATFORM SPONSOR |
YARDI INVESTMENT SUITE
For over two decades, Yardi has developed connected investment management solutions tailored for global real estate investors. Supporting trillions in AUM across a diverse client base, including institutional funds, partnerships, and emerging managers, Yardi enables full visibility and control across the entire investment lifecycle.
From sourcing acquisitions and raising capital to accounting, debt management, performance benchmarking, and investor reporting, Yardi helps streamline complex processes into one integrated platform. Investors gain secure, self-service access to reports, documents, and investment positions, while investment firms benefit from improved collaboration, faster distributions, and scalable operations.
Built on 40 years of real estate technology leadership, Yardi’s end-to-end capabilities empower commercial real estate firms to manage with confidence, transparency, and efficiency.
ASSOCIATE SPONSOR |
HAWTHORNE RESIDENTIAL PARTNERS
Hawthorne Residential Partners is a vertically integrated owner, operator, and developer of Sunbelt Multifamily and Build-to- Rent product.
The firm has acquired or developed 24K+ units since inception and currently manages 64K+ units across eight states. Hawthorne partners with institutions on a single-asset and programmatic basis.