Unlocking Big Data

The coupling of big data with artificial intelligence (AI) and the internet of things (IoT) has power to disrupt the traditional way of basing real estate decisions on experience, past performance and intuition.

The machine learning component of AI enables multifamily owners and operators to more accurately and quickly pinpoint future risks and opportunities and gain insight into the trends driving performance in markets and at individual properties.

“It’s an evolution. Collected data is beginning to drive artificial intelligence engines. Whereas before, software would capture data for retrospective analysis—things that had happened in the past—the aggregation of massive amounts of data are being used to predict what will be, or should be,” said Jeff Adler, VP of Yardi Systems’ Matrix Products.

“The purpose of using AI is to speed up the analysis timeframe, reduce the repetitive mundane tasks required to make decisions and tease out relationships that would be hard to see when humans are looking at the data. That’s where big data takes a step beyond what traditional analysts have always done, what I have always done, and that is the future of predictive analytics,” he said.

An example is Yardi’s RENT Café, where AI engines deployed across the platform filter and distribute interactions with prospective apartment renters that come into the software via phone, text or email.

Based on how they are read by AI, interactions are either automated or go into a text library or to a chat box—the latter, a function of AI that provides real-time, detailed answers to apartment prospects’ questions, instead of scripted responses. Other examples of data useful in optimizing the resident acquisition process include turn list data and marketing budget data.

Another area in which data is organized, put through an AI engine and converted to recommended outcomes is real estate investment analysis. Yet another is predictive maintenance, where data generated by smart appliances installed in the building can alert management to impending malfunctions and even generate service requests to repair or replace the failing part.

“This is an emerging area and today we are focused on wrapping our arms around data that we have inside the various Yardi products and services. Big data is a five- to seven-year journey and the pieces will be deployed in different practical solutions over time,” said Adler.

Leading the charge

The process of utilizing advanced analytics across a portfolio entails collecting enough data to build accurate algorithms and manually scrubbing the data before it can be used. The datasets of any one owner of an individual property tend not to be large enough to draw generalizable outcomes, so any particular owner has a hard time amassing the amount of information sufficient to make predictive analysis work, said Adler.

For that reason, established software providers who have the ability to amass huge swaths of data from a multitude of sources are leading the analytics charge in the multifamily space.

The ideas behind data mining for predictive analytics were introduced to the apartment industry in 2001 with the revenue management systems that use algorithms to set rents for individual units, making previous methods of setting rents a thing of the past. These systems may be standard today, but they took almost a decade after introduction to catch on.

Similarly, big data has been slow to take hold in the industry where owners still rely on instinct, experience and human touch to glean a course of action from retrospective data.

And, with a number of start-up tech firms entering the market with AI products, many operators are still on the sidelines waiting to see which of them fall by the wayside and which remain standing.

“I think the future of AI and analytics primarily will be driven by the software platforms that have access to huge amounts of data. There’s a bunch of little companies that are trying to do the analytics, but they do not have the underlying plumbing, and I think most of those solutions will eventually go out of business,” said Adler.

Meanwhile, many industry professionals still compile market survey and comparison data the old way.

“We still have our onsite team members complete their own market surveys and regularly shop comps in person and over the phone in similar methods to what has always been done,” said Diane Norbury, senior VP of multifamily operations at Pillar Properties.

“While data providers have become increasingly more valuable in terms of their breadth, scope and frequency of compiling data in a more useful way, there is, in my opinion, no replacement for the daily touch points that our teams have as ‘boots on the ground.’ This allows them to remain in touch with what is happening in the marketplace in a way that can’t be fully captured by third-party sources.

“That said, data providers allow us to validate what we hear and see or dig into areas that previously were difficult for us to reach, providing a much more complete picture. From there, users quickly ask more educated questions and have the tools to develop more valuable strategies for their own operations,” she added

However slow big data is to catch on in multifamily, it’s power is too great for industry professionals to ignore. “AI is not going away. It is going to be something that is very important,” said Adler.

Norbury concurs, saying, “As data scientists continue to find ways to cull this data, scrape it and present it in meaningful ways, it will naturally develop into more reliable and valuable sources for our industry—especially as more users begin to rely upon it for daily decision-making.”

Ownership issues

But surrounding big data are ethical and privacy issues that also cannot be ignored. Data anonymization is the use of one or more techniques that make it more difficult to identify a particular individual or property in stored data. Most privacy protections rely on informed consent for the disclosure and use of individuals’ private data, but because big data is a resource that can be used over and over again and often in ways not originally conceived of at the time the data was collected, erosion of anonymity can occur.

Regulations like the General Data Protection Regulation, a 2016 law on data protection and privacy for all individual citizens of the European Union and European Economic Area, and a recent data privacy law passed in California that is expected to be the first of many in the U.S., set up strict conditions for gaining consent from individuals and place the burden on the data provider to prove that an individual agreed to a certain action.

“It’s very important to us that anything that is individually identifiable remains the property and ownership of our clients and that is a very important core value of ours as we draw on aggregated and anonymized data to create solutions that add value to all of our clients.

“Based upon our relationship with clients and a choice we made, we have created an agreement by which we have the ability to aggregate and anonymize information to create benchmarks and actionable predictions for our clients, but not expose the data individually,” Adler said.

He explains there are substantial costs associated with data cleansing, providing appropriate security for the data and eliminating the inconsistencies that can happen when humans enter data. Data is cleaned monthly at Yardi by an autonomous team of experts who report to no Yardi product line and are the only humans with eyes on the underlying information.

“These are significant efforts by very skilled people who have to know a tremendous amount of information and knowledge about how all of the systems function together,” said Adler.

The data is extracted from clients’ data bases and placed into a centralized database to be normalized and cleansed and to ensure a consistent timeframe. Once assembled, a set of security routines must be run if the information is requested for extraction. Only then can the data be reported out on at a competitive property level or market or submarket or segmentation, but even then, the data must meet certain criteria. A built-in anonymization and aggregation routine requires at least five properties from two or more owners on the return set.

“The way it works in Matrix is you identify a comp set of 15, 20 or 30 properties and then you call out for the data to be returned. As long as there are five or more properties, you will return the results. I won’t and can’t know which five, six, seven or eight of the 15 or 20 came back. That’s the anonymization,” said Adler.

Making it better

Adler said “AI is really a misnomer for a set of technologies, like neural networks and language processing, that interact with each other to create better predictive and proscriptive analysis by applying large sets of data to a problem and running iterations sufficiently frequently to be able to create output and a feedback loop where the machine learns from each of them—with the idea that the more transaction data, the better the machine is able to make predictions over time.

“We’ve taken baby steps—the first was ‘Wouldn’t it be great if we could predict in advance if a property had enough demand to meet financial goals?’ And we said, “What do you need to make that kind of decision?’ You need to know what renewal rates are by floorplan, what the notices to vacate are and what they are likely to be, how much demand is coming or expected to come and put those things together and see if there is a gap. There is a set of algorithms that need to run to get to the point where I can say, ‘I think you have a gap.’ If you run that algorithm enough and back test it, you can say with some confidence there is a prediction of a gap. Now, what do you do with that? There are only three levers one can pull—change the price, change your advertisement activity, or change your resident qualification standards. In the absence of machine learning, and this is what I did as COO of a company, we made those decisions based upon the best information we had at the time and our gut understanding of the relationship between each one of those three levers.

“So, finally we ask, ‘Wouldn’t it be great if a machine could make some of those decisions? Could they make decisions more quickly and better than a human being?’ The first thing we did to derive answers was to exclude the resident quality standards, which were a bit more complex than we could handle and decide if we should increase price or modify advertising. Invariably, what tends to be the case is that as long as you are in a large urban market where there is a lot of demand and your property is a very small percentage of the total stock of housing in that market, it is less expensive to modify your advertising to get more leads to achieve your desired objective. So, we now have that chugging along inside of one of our services.

“We are very early on a five- to seven-year AI journey We are not trying to reinvent the wheel. We are harnessing some of the relationships in our aggregate dataset to make everyday problem solving a little better for our clients,” said Adler.

Benchmarking boost

Pillar Properties is an early adopter of Yardi’s Elevate-branded, AI-enabled intelligence services designed for C-suite and other operational execs. Norbury gives testament to the predictive analytic and benchmarking prowess of AI-enabled software.

“Having benchmark data now allows outliers or concerns to make themselves more readily apparent, so decision making and changes in strategy can be done quickly, with less damage to operational performance when righting a course,”

“Also, it can highlight issues relating to specific areas of a business for more accurate adjustments, rather than applying a shotgun approach and perhaps missing the overall mark. We’ve found that by using benchmarking data, we can quickly assess if a troubled asset has issues with pricing, marketing, expense management or perhaps team members—whereas before it wasn’t always that evident. It has also made our onsite teams much more educated and independent to make successful decisions,” said Norbury.

Benchmarking data also helped Pillar Properties pinpoint the reasons for lagging performance in certain buildings. “For example, we found that one of our communities wasn’t reaching our rent growth goals and was having difficulty with occupancy as well, despite a strong competing sub-market that had high occupancy and rent growth. From the data, we found that historically this asset always outperformed its competitors in rents and rent growth and significantly slowed now that it was aging and there was new product online. We were no longer able to push rents, as we were already at the top of the submarket,” said Norbury.

After a sensitivity analysis extrapolated from the data showed that significantly dropping rents at the property would result in the occupancy level needed to meet or beat cash flow goals and still remain competitive, Pillar Properties changed focus from rent growth to occupancy and recalibrated its budgets.

“We’ve also been able to use benchmarking data to see when our competitors will have expirations and availability in certain floor plans and get ahead of strategizing our own vacancy and marketing plans—we never had tools like this before, and that’s extremely powerful.

“The multifamily and real estate industries are finally catching up with the rest of the world in successfully aggregating and digesting information in a valuable way that can improve efficiencies, operations and overall success. The more participation, the more refined the tools will become for everyone,” said Norbury.

Author Wendy Broffman