Aiming for Recurring, Sustainable Value Creation
by Building a Data Ecosystem
January 20, 2021
SRE Holdings Corporation (hereafter referred to as “SRE”) is driving digital transformation (DX) in the real estate industry by platformizing this real, physical business and making it smart and open. The company is also starting to apply this knowledge to other industries. At the core of these initiatives are their AI-driven cloud and consulting services. What kind of future does SRE see? We talked with director Tomohiro Tsunoda and data scientist Shun Lee to find out.
SRE Holdings Corporation
& Consulting Business Division
SRE Holdings Corporation
Sony, in real estate?
──How did SRE get its start, and what kind of business are you involved in now?
Tomohiro Tsunoda：SRE got its start as Sony Real Estate Corporation in April 2014. The company was positioned as a corporate startup. The idea was to have engineers collaborate with customers with specific needs to rapidly produce technology-driven solutions that could be useful in the real world. Since we thought this type of value creation would only become more important going forward, and technology in many ways had yet to reach the real estate industry, we thought that we could bring a bit of Sony-style innovation into the picture. First, we brought IT and AI to the real estate industry. In this process, we began making forays into finance, which uses property value for evaluation of collateral, and then we expanded gradually into other domains. Based on this growth, the company name was changed to SRE Holdings Corporation in May 2019, and today, we have two mainstay businesses — real estate, and AI-driven consulting services for other industries.
Shun Lee：I think one of the defining features of SRE is that we have team members from the real estate industry and AI engineers working in the same office — in other words an atmosphere of “real world business + technology.” Instead of tech companies with superior technical abilities and AI vendors simply handing over the tech to industry pros, saying “This service and system is convenient, so go ahead and use it,” real estate pros and engineers discuss what to do from the planning stages, and they do this leveraging actual customer-dependent data, allowing for service development and implementation that meets the needs of the industry. Take, for example, the real estate value estimation AI we’ve developed. We don’t simply feed in big data to have it learned and obtain an output. Rather, we have our in-house real estate pros actually use it, then fine tune the algorithms based on their feedback. As we deliver services based on AI, IoT and data analysis out into the world, we directly connect with the customer, listen to their issues, use real data from them to create machine learning models, and then offer precise solutions. I believe this is the feature that is consistent across every realm of our business.
Leveraging open “real world business + technology”
──What kind of process did you go through to grow your AI consulting business?
Tsunoda：In rolling out “real world business + technology” as our service, we have gained the backing of many different customers. However, in terms of spreading innovation more broadly and really changing the world, we realized that even if we were to create strong services on our own, their use would still be limited. We saw it would take a long time to penetrate the entire industry. So, with the idea that providing our technologies and services as a platform would better accelerate big, real-world changes, we collaborated with Yahoo Japan Corporation one year after founding our company, to build a platform for online real estate sales. We had no doubt that the time was coming for application of AI to this field, so it naturally followed that it would be vital to have a platform and collect data. Consequently, we began offering a set of various cloud-based tools for corporations involved in real estate sales, including tools for easy creation of contracts and disclosure statements online, attracting customers online, and easy real estate appraisal using AI.
In 2018, we received an inquiry from a financial institution asking about employing the AI that we had been using in the real estate industry for price estimation, and this prompted us to move ahead in earnest with our AI solution business. Today, this business has been rolled out not only in real estate and finance, but also to a wide range of other customers, including some in the power industry, securities companies, and credit card companies.
At SRE, we believe in creating solid data ecosystems that collects data from platforms and real feedback from customers. We make our real estate sales platform, Ouchi Direct, also available for use by other real estate agents, which means we have considerable volume of past real estate sales data. By working with various real estate companies, we have been rapidly accumulated unique, readily accessible data, which we believe sets us apart from other companies.
──What kind of benefits or synergies are created by Sony’s involvement in the real estate industry?
Lee：Customers purchase typical Sony products, such as a TV, for example, from an electronics store or on the web and set it up themselves. In contrast, SRE is involved in real estate, so every aspect can be incorporated into our services, making it available when building and making renovations. In other words, this allows us to use a different sales method. More specifically, sensors can be embedded in the environment (building) from the start, facilitating various new possibilities for urban development. I believe using our strength of “real world business + technology” makes it possible for Sony to drive digital transformation in real estate. Of course, I believe we can help drive various aspects of real estate DX in real estate transactions as mentioned previously.
Consulting and cloud services
──One of the various issues with AI is bias. What kinds of biases are possible in your business domains and how do you overcome them?
Lee：Bias is an unavoidable part of dealing with AI, and an issue that should continue to be discussed in the future. I think AI bias is likely in various situations, but we are basically dealing with two main types of bias. The first is bias over the passage of time. I think in some cases the AI continues to use old data as-is despite changes in actual situations. Our real estate price estimation engine is updated once a week as a way to counteract this issue. This also applies for the electrical power supply and demand volume prediction engine I am currently involved in developing. In order to address factors such as climate change, aging and deteriorating power generation equipment, as well as the lifestyle changes of occupants of each household, we set an appropriate model-update frequency for each engine. In this way, I believe we can counter bias affected by information freshness by updating models with the passage of time.
The other type is sampling bias. For example, suppose we were to develop AI for a company to recruit personnel based on past hiring data. If the past recruitment data had included a higher proportion of men, the AI would mistakenly assume that males should be prioritized when selecting applicants for recruitment, and this can lead to a feedback loop that repeats and strengthens this kind of unfair bias. This type of problem usually occurs when the human uses the system without a deep understanding of the data and learning model. In the end, despite AI being a form of intelligence, extrapolations and judgements are merely based on past data entered into the system. We believe it is important, therefore, for us to ensure thorough visibility of learning data content and learning models by checking how the data is configured and what the basis for extrapolations and judgements is, making sure data sets are correctly correlated, and then discussing the data, as we monitor data and AI output from many different perspectives.
Tsunoda：It is indeed true that, whenever AI is discussed, the topic of the invisible elements of the process comes up. For example, suppose a financial institution provides tens of thousands of home loans. They need to estimate the asset value of the applicable properties. In order to do this, it is possible to construct a highly precise AI engine using deep learning, although the use of deep learning results in a lower level of explainability. However, our client must be able to explain the process of the estimation, which means we need to opt for a more visible method — in other words we need to use an AI model with a certain level of explainability to help create highly explainable price estimation. Compared with deep learning, however, this method is somewhat less precise. Both precision and explainability are important, but the priority may shift depending on the end use.
──I’ve heard that the SRE AI solutions business includes AI cloud services in addition to AI consulting. How are those two businesses connected?
Tsunoda：As I touched on it a bit earlier, our offerings basically encompass the whole process from planning to operational development. In our AI solution business as well, we not only create AI models; we also develop cloud services and provide maintenance and operations. The phrase, “AI as a Service,” is trending lately. That’s how we run everything at SRE, including real estate assessment tools and services that provide property information to customers on a one-to-one basis. It’s a recurring business, so to speak.
Our other core service of AI consulting is bespoke, since we build an AI model and develop systems with built-in AI to provide solutions for issues faced by each of our customers from different industries. We first implement technical application proof of concept (PoC), then shift to actual development and integrate the solution into their environment, and then receive a monthly licensing fee.
──In what industries do you provide AI consulting, and what issues do you tackle?
Lee：In AI consulting, the first example that comes to mind is the finance industry. With investment banks, for example, we provide a recommendation for an AI engine that uses the stock transaction history of customers who have never purchased bonds to estimate their probability of purchasing bonds and what kind of product they might buy, while also visualizing bond market conditions. Another example is a dynamic pricing AI engine that maximizes profits at lodging facilities such as hotels or Japanese inns. When using a travel website or app to book a hotel, these services offer a variety of plans. The profit at the lodging facility changes depending on the set price. Off-season rates are cheaper and peak-season rates are more expensive, so good decisions are needed to make a profit. If rates are too high, guests will not come, and if they are too low, there won’t be any profit. Traditionally, Japanese inn operators set prices based on experience, which presents challenges such as personnel costs, and the fact that each person’s experience means their ability to make the best decision is not equal. So with this in mind, we built a system that automates pricing.
Tsunoda：Hotel room prices are determined based on their accommodations, such as single or double beds, so dynamic pricing is simple. All you need to do is look at how many competitors there are nearby. However, with Japanese inns, the price structure and profitability changes even for the same room if the guests add dinner plans, stay without meals, or book rooms with an attached outdoor bath. There are also other factors that have to be considered. For example, in a room that accommodates four people, a reservation for two people may be allowed during a certain season, while three or more people may be required to book the room during another season. In other words, there are many parameters to optimize. For this reason, we not only control rates via dynamic pricing in these cases, but also manage the combination of plans available at Japanese inns. That’s why I believe our service is at a completely different level than other AI engines.
Lee：We are also working to solve various other challenges, such as suggesting to credit card companies which people have a high probability of making credit card payments in installments, predicting people with a high probability of investing 500 million yen or more for online brokerage companies by looking at the three-month history of new accounts, and using resumes and SPI data to estimate whether mid-career hires at real estate companies are likely to perform at 120% or above, or stay at 80% one year after joining the company. Come to think of it, one challenge financial institutions such as securities companies face is finding wealthy customers. Some of these institutions had been making such predictions on their own, but combining our real estate price estimation technology with the addresses of customers held by the financial institution made it possible to deliver even higher precision identification of wealthy clients.
Aiming for blue ocean AI
──What kind of AI cloud services do you provide, and to what industries?
Tsunoda：In the AI cloud area, we currently work mainly with real estate and finance, offering real estate AI assessment and financial collateral evaluation. However, I think combining this with our AI consulting could provide more effective results. First we would use AI consulting to identify and extract issues, and then we would create a new AI cloud model and offer it as a recurring service, allowing for an ecosystem that accumulates data leveraging both our AI cloud and AI consulting services. This is the strategy we’re currently considering.
──What sets SRE apart from other AI vendors?
Tsunoda：Normally, it isn’t possible to just create a cloud model for a recurring service, so I think in many cases companies start from an AI consulting perspective. In that regard, SRE starts from AI cloud services because we have a history of providing the cloud service we created as part of Sony Real Estate’s development division to financial institutions as they requested. Compared with other AI vendors, we have a solid base of recurring business, onto which we can add sales in AI consulting, making us well-positioned for a profitable business.
──Could you share your outlook on future initiatives?
Tsunoda：Because we work in real, physical space as a real estate broker, moving forward with smart technologies makes it possible for us to monetize such real businesses. By platformizing AI that can hold up under practical use, we’ve succeeded in scaling and opening it up for use in other industries. We have a cloud smart-service platform used for
AIFLAT, which is an IoT apartment series developed by SRE. Having other developers use this platform in the future will help collect data for scaling, going forward. The next domain we are looking at is the property investment business. We’re currently in the preparation stages to make a foray into the real estate investment trust (REIT) business, where we plan on introducing IT and open transformation, collecting data to expand the business.
People often talk about “red ocean AI” vs. “blue ocean AI.” For example, although something like an AI chatbot may be technologically difficult, it’s easy to imagine what kind of situations it would be used in, and it could be developed pretty much for any situation by someone well-versed in machine learning. But you can’t accommodate requests to introducing AI into the real estate brokerage selling process if you don’t understand the physical aspects of this particular business. Naturally, you also couldn’t make asset management processes, which are an essential part of REIT, AI-driven. In other words, there are barriers to entry into the reality-based AI we do, so it’s impossible to create such a system simply by understanding regular AI. In that sense, we are practicing blue ocean AI. In the future as well, that’s what we want to continue to secure. Once we have a proven track record, it will be easy to introduce our services into other domains, just as we expanded from real estate into finance. This is the kind of course we want to take. We need to create a solid data ecosystem and then develop recurring, sustainable businesses.
Lee：We deal on a daily basis with customer challenges such as the inability to use AI without data. We can’t just say “you can create AI services if you have big data.” Rather, we start by helping customers collect data from real sources with the goal of building an AI service. Going forward, we may be able to process collected data on devices, allowing for real-time feedback. Even in such areas as high-performance, highly responsive edge computing, I believe we can create a scheme that puts us on top thanks to our physical real estate business.
We are looking for full-stack engineers
──In order to make further developments, what kind of staff do you want?
Tsunoda：I think we’re currently at an exciting stage. We’re in an AI startup position within the Sony group, and we find ourselves in an environment where we actually interact with real data. So, I think that people who want to build an actual business using technology to create solutions for the world would be the ideal match for SRE.
Lee：I think full-stack engineers who understand various technological domains and development processes will be able to demonstrate their potential here. These are the kind of people who are able to perform each of the different processes in the development phase. In order to flexibly accommodate diverse needs and build business with a sense of speed, each person will need to fulfill multiple roles, as the situation requires. For that reason, I think people that are knowledgeable or interested in various technical domains would be of great use here. I look forward to welcoming individuals like this to our team.
Tsunoda：We envision what we want to make from the planning stage, build a prototype, develop the actual system, then configure the infrastructure and network for operation in the cloud. Afterward comes collaboration with component technology development areas, which you could say is unique to the Sony Group. We need full-stack personnel that are interested in everything. Instead of a mentality that sticks to just one area, or people who don’t want to handle operations, we’re currently at the stage where we’re looking for people interested in doing everything. Also, after working with us for about five years and when you have come to bear in mind specific services you want to realize using AI, we want you to be able to handle the entire process, from configuring servers to service operation. Maybe you can even think of creating a new AI company. I want to create and deliver new value based on new technology, so I really want to see enthusiastic people who are excited to learn new skills, listen to customer challenges, and be a part of creating a new real estate AI company.
We Are Ready to Eliminate
the Regional Disparities in Education and
Encourage the Individually-Optimized Learning
December 18, 2020