Introduction to machine learning in commercial real estate Part 2: Five components to leverage machine learning advantages
Key highlights
Efficient and cost-effective implementation of machine learning in CRE requires cloud computing, models, data, and capable people, as well as allocation of time and money.
CRE investment firms encounter challenges such as a lack of high-quality data, organizational resistance, and difficulties in hiring qualified staff when developing internal machine learning capabilities. It is crucial to overcome these hurdles to achieve success.
Third-party machine learning providers offer advantages such as experienced data scientists, ongoing model updates, avoidance of unintended biases, and the ability to quickly scale capabilities.
The first part of Introduction to Machine Learning in CRE addressed the features, competitive advantages, and range of uses for machine learning in the commercial real estate sector.
But how can CRE businesses efficiently, productively, and cost-effectively implement machine learning in your organization?
Some businesses build in-house systems, others rely on external providers, and many follow a hybrid approach. In all cases, machine learning requires the following five key components:
Cloud computing
Models
Data
Capable people
Time and money
Uptake and challenges in utilizing machine learning
The Altus global research report, The State of Data Science in CRE Investing found that half of the CRE investment firms surveyed had some in-house data science or analytics capabilities, with 44% of these leveraging machine learning capabilities.
Their most significant challenges with developing internal capabilities were:
Lack of high-quality relevant data to use in inputs (44%)
Difficulty getting the organization to buy into and adopt these capabilities (41%)
Difficulties hiring and retaining qualified staff (39%); in fact, only two in five organizations have staff dedicated to data science and analytics.
1. Cloud computing
While open-source frameworks enable organizations to build machine learning models in-house, machine learning requires huge processing power as many servers work simultaneously on algorithms. And this is data-intensive work – algorithms require a massive amount of high-performance storage.
This is why most companies, whether having in-house capabilities or working with an external provider, rely on cloud computing for machine learning, which does not require a large capital investment. Moreover, the cloud enables a business to experiment with machine learning capabilities and to quickly scale as demand for these capabilities grows.
2. Models
Some CRE businesses prefer to develop their own machine learning models as part of their “secret sauce.“ Others choose to rely on third-party providers to complement their models or as a sole-source option.
When opting for the latter, it’s important to ensure the provider has extensive experience training machine learning models. Since data continually changes, to ensure a model does not degrade, it must also be continuously updated with new data.
Another advantage to using a third-party provider is avoiding unintended biases. For example, a company may subconsciously want an asset to perform better, so their model might be unintentionally tweaked to overweigh certain factors. With years of experience working with a range of organizations, experienced external providers continually strive to ensure that data, frameworks, models, and systems are free of bias.
3. Data
A vast amount of high-quality data is needed to train machine learning models. However, given that a typical real estate fund has fewer than 100 assets, this generally does not yield sufficient data to build a reliable model. For most CRE businesses, this necessitates investing in data subscriptions. Data providers offer data sets featuring investment, leasing, and development intelligence for a range of asset types and markets. This can include public and private data sources as well as traditional and non-traditional, from demographic, economic, market, transaction, property, ownership, tenants, and lenders, to foot traffic and cellphone movement.
Predictive analytics and market insights
Altus Market Insights incorporates market intelligence and predictive analytics into your daily workflows and key decision-making.
4. Capable people
Some CRE businesses will choose to build an in-house data science team, others will entirely outsource, and some will rely on a hybrid arrangement. Among CRE investment firms, Altus’ global research report, The State of Data Science in CRE Investing, found:
38% use external providers to support their data science and analytics program,
the majority purchase both tools/software and expertise/services, and
43% purchase data sources or databases.
Recognizing the ability of data science to boost performance, some large real estate organizations are building internal data science teams that include data strategists and analysts, and software and machine learning engineers.
Among the advantages of developing an internal team is, with time, they develop an intimate understanding of the business and comfort with the culture. And as the team continues developing algorithms and models, machine learning becomes embedded into the business, improving performance and adding value throughout operations.
For smaller organizations, it can be challenging to attract and afford appropriate talent and the cost of developing the technology. Demand for data scientists and artificial intelligence specialists far exceeds supply. There are even fewer of these professionals who have both the technical skills as well as the understanding of commercial real estate to effectively apply machine learning to these particular business challenges.
“Data science teams need both technology and domain expertise – that’s hard to find in commercial real estate.”
Sally Johnstone
Senior Manager
In such instances, collaborating with a trusted partner who can provide the data and technology expertise and immediate scalability can move initiatives forward, months, sometimes years, faster.
External providers continually identify new use cases for machine learning, expand coverage of asset classes, and incorporate more factors influencing asset performance. They can also supply accurate, up-to-date data that is cleaned and aggregated for immediate use.
5. Time and money
Arguably the chief reason more CRE businesses don’t launch data science/artificial intelligence/ machine learning initiatives in-house is because of the large investment of time and money required.
“A new data science project typically requires hundreds of thousands of dollars in investments in new data infrastructure, data sourcing, and data scientists, and takes months to deliver – with potentially uncertain outcomes.”
Sally Johnstone
Senior Manager
According to research by Gartner, only 53% of projects make it from artificial intelligence prototype to production. Evidently, Chief Information Officers and IT leaders find it challenging to scale these because they lack the appropriate tools. A variety of factors will determine an organization’s success in launching a successful machine learning initiative. The article Seven questions to ask when developing data science capabilities for CRE investing offers a set of key questions and suggestions to consider.
How third-party machine learning delivers detailed predictive analytics – within hours
The provider’s machine learning engine, which has been trained by a team of data scientists, ingests client data, which is typically supplemented with third-party data to enrich details.
Deliverables are customized to the needs of individual clients and the composition of their portfolios. Analytics solutions are produced within hours or days. This might include comparisons of how the client’s performance distribution, average growth rate, and growth rate distribution compares to the master data set, providing a basis for group comparison over time. Analysis could also include changes in allocation of risk, factors influencing over- and under-performance, and potential for growth. This can also include analysis of individual properties.
Machine learning models are updated quarterly as new performance information becomes available, along with any updates to the third party’s demographic, economic or other pertinent data sets. The client’s quarterly net operating income, occupancy information and any other relevant data is again fed into the machine learning engine. Portfolio projections are immediately updated.
Machine learning is a process
When making decisions regarding your machine learning future, it may be helpful to keep in mind that artificial intelligence and machine learning are a process, not a once-and-done solution. Since outcomes are only as valuable as data inputs, benefits such as higher revenue and productivity take time to accrue. Those organizations that postpone machine learning innovation will trail competitors in realizing these advantages and a return on their investment.
“Why aren't more commercial real estate companies embracing AI and machine learning? Likely because they’re relying on what they've always done. But as adoption of these technologies and solutions continue, if they don’t participate, they will be stagnating while others are leading.”
David Cockey
COO, Head of Consulting
Machine learning might just be the enabler to help your CRE business assess the current situation, forecast potential, and make well-informed decisions that add value to your real estate assets.
Author
Altus Group
Author
