Much of the AI discussion I have seen of late is heavily focused on the automation of content creation. Even within our industry, much of the focus in AI discussions has been on the topic, along with the automation of tasks.

There are some obvious examples where AI is being leveraged within our industry, such as creating blog posts, curating individual email or newsletter content, and even website content and listing descriptions. However, very little time is being spent discussing where and how deeply embedded AI can go beyond simply creating more user engagement to where it is truly generating more business. 

I have been pushing AI for ten years now. In 2014, I started working on AI software that would help deliver email newsletter content individualized to each recipient. I was able to bring that product to market in 2016. At that time, this AI was, and still is, considered very basic, but to this day, it is highly effective and still in use.

One area that I have always struggled with in the real estate industry is being able to accurately convey to an individual sales associate, with high probability, that a person they are working with, or who is active within their CRM, will transact in the next thirty days. I have found over the years that most sales associates, good or bad, will focus on the immediate business, most likely closing in the next few weeks rather than a few months or a year down the line.

When conversing with a Realty Alliance member in 2018, I realized they felt the same way and agreed that it would be amazing if we could accurately predict if and when someone would buy or sell a property. As a result of that conversation, the two of us engaged with the North Carolina State University Institute for Advanced Analytics on a year-long study with five brilliant graduate students to see if we could develop a model, an AI engine, that could do this for us. 

Here was the objective of our engagement: 

[Realty Alliance Member] website sees thousands of new user accounts generated each year. Currently, there is no formal lead scoring mechanism in place, making it difficult for agents to prioritize which leads to follow up with. In response, our practicum team has developed three models to help the business better prioritize and convert web leads: 

1. An early lead identification model that flags high-intent users, graded on the user's first week of activity on the platform. 
2. A lifetime lead scoring model that assigns lead scores to every [website] user, graded on the user's entire history of activity on the platform. 
3. An agent matching model to match high-intent users with agents based on similarity, as determined by the user's search activity.

I won't bore you with all the details of what we went through during the research study except for a few important ones. Firstly, this study involved sharing an insane amount of data from tracking users. All of the data was scrubbed in a double-blind fashion so there would be no way to "cheat" during the study. Secondly, the study also involved sharing transaction data, which was also scrubbed in a double-blind fashion. Thirdly, the data shared included nearly thirty key activities that website visitors could take.

Lastly, during the study, this group of graduate research students spent an entire year working on data analytics models, driven by various AI engines, to see how the different AI model engines were in their accuracy and which key activities were most important in generating transactions. 

At the end of the study, here is what we found: 

In 2016, we had an AI engine in hand that, based on a double-blind study, was 73% accurate in predicting if someone had high intent on buying or selling real estate immediately within the first seven days of their activity. This engine was also 70% accurate in predicting if someone had a high intent on buying or selling real estate immediately as time went on for up to ten weeks of continued activity. 

The most important thing that we learned through this study at that time, and that we still use to this day, is which website and CRM activities and data points are most important that lead to transactions. This is especially beneficial today with the wide variety of technology tools available within the industry for understanding what to focus on relating to AI deployment.

Here are some key takeaway points from the study:

1. We found a direct correlation between the time of day people were active on real estate websites and their likelihood to buy or sell property. This was an interesting find to us, which correlated with the number of visits, length of the visit, and timespan of the activity. 
2. We found an extremely high direct correlation between the types of email marketing that were engaged with and the likelihood of buying or selling property. I had assumed this to be the case, but the interesting part of this correlation is that it could be contradictory to the website activity correlation. This is a major reason why we focus so heavily on email marketing, specifically within our platform, and why our email marketing is so advanced compared to other platforms, including MLS platforms. 
3. We found a direct correlation between cross-device website activity and the likelihood of buying or selling property. This was also a surprise find to the researchers doing the study because we were unique in the web space on the type of cross-platform data we track and match even to what we call unregistered devices to user accounts. This was the one data point that the researchers were most excited to delve into due to its uniqueness. 
4. We found, over time, that the top three most important data points can and do change, even on a per-market basis, at the same time. This makes some sense to those who understand the uniqueness and the hyper-local aspect of the real estate industry, but it is still interesting to see this hold true as it relates to technology in the space as well. 

So, my question for you is this: While you may have augmented current activity with AI, what new things are you doing in your real estate business because of AI?