Can Artificial Intelligence Change Property Evaluations Merely by Existing?
Current as of 11/26/20
The digital transformation has had profound effect on the real estate industry, not just by banishing rolodexes from brokers’ desks, or standardizing real estate listings. It also introduced Artificial Intelligence (AI) to an industry that often relies on the human touch, particularly in customer relations and property evaluations. While customer relations management teams in larger brokerages have incorporated digital tools for their tasks, properties, in most cases, are still evaluated manually. Questioning why AI has not taken the job of evaluating property values, by itself a numbers and data driven task, off peoples’ hands, leads to the question of whether AI can be trusted to evaluate real estate reliably. Our take is that AI is up to the job.
First off, it is all about the data. After all, both human and AI evaluators work with datasets that comprise of a multitude of data points. Without data, neither can do this job. Whereas humans learn through motivation, exposure and experience, AI on its own is not motivated to learn. Instead, it can gain knowledge and experience in a process that is known as Machine Learning. Essentially, you put AI through a rigorous training regimen to teach it how to connect the dots in the data you feed it. Because AI never wavers in its attention, it can learn from millions of previous property evaluations in a short amount of time. Humans obviously require considerably longer to attain the same amount of experience. But, at the end of either training process, you have, simply put, experienced property evaluators — just that the AI has learned from a few million evaluations more. However, AI still needs assistance, especially when it comes to recognizing patterns within the existing data.
In our case, we need a large number of examples of existing properties and their corresponding selling prices. After all, the Machine Learning method must understand the data and learn from it in order to be able to predict the prices for upcoming new properties. During the training process, Data Scientists and Engineers have to design and choose features, utilize additional sources of information, compare different algorithms, and evaluate the performance properly. To be clear, AI is not able to learn its tasks independently without prior instruction but stands on the shoulders of those that have taught it everything it knows. This includes the experiences of human property evaluators that labeled the data the AI and its handlers use for training. In essence, using AI for evaluations is not innovating the process, but rather automatizes it.
We also need to keep in mind that AI cannot independently gather and categorize the data it needs to perform a solid evaluation on its own. Though potential sellers supply us with valuable information about their property when they first get in touch with us, the resulting evaluation is preliminary and incomplete, since we only ask for a limited number of data points. Evidently, information and data are key to making accurate evaluations, and that, regardless of whether it is an AI or a human evaluating the property, both require more information. This is where the human touch is still required, and where McMakler’s brokers enter the scene. Our brokers get in touch with prospective sellers, either on site or through a video call during which our brokers share their screen. In either case, our brokers complete a standardized survey with the sellers’ assistance. The survey alone collects more than 300 individual data points, which are then fed to our AI. The evaluation process itself is then just a matter of seconds since our AI is capable to deliver consistent and accurate results in a record amount of time.
All in all, it is safe to assume that, without the ability to recognize the underlying patterns within data, anyone attempting to analyze the available data will be hard-pressed to deliver results. What sets AI apart from humans is their ability to parse information quickly. In our case, we trust AI to reliably evaluate properties, while leaving the tasks that require human interaction to humans. The important part of introducing AI to the evaluation process is, that we, on the one hand, have sufficient amounts of real world examples from which AI can learn from, and on the other hand, use the right algorithm with the right features for this challenging task. AI on its own cannot make these decisions. After all, it takes experience to be able to gather insights from raw data.
