Predicting Intent in Real Estate with Data-Powered AI
What is your elevator pitch/30-second sound bite for Likely.AI?
We are an artificial intelligence platform that is specifically designed for the real estate industry.
We’ve built this platform by aggregating data, including public record data from ATTOM on 155 million properties nationwide. We also incorporate a 225-million-record demographic dataset that provides information on buying trends and behavioral trends. We also have accumulated a macro market and micro market data set that is proprietary, made up of both local and national influence signals.
We use all this to create different lead-generating products for the real estate industry, the first one being who has the desire to sell. We’ve gotten down to where we’re highly accurate.
In essence we score every U.S. property based on the likelihood of a sale in the near future, and once a property reaches a certain confidence threshold in our deep learning models, we say that is a valuable lead for real estate professionals.
How is Likely.AI Data utilizing ATTOM real estate data?
Other companies — list optimizers — look backwards to predict and it’s not a very viable model. It’s something we were doing back in 2007. These list optimizers, some of whom have rebranded as AI companies, use shallow neural networks with limited layers of neurons. Now we use deep learning. Deep learning has very deep neural, long neural networks. All these layers of neurons allow you to make non-linear non-sequential connections between layers.
Because of all the different property data that ATTOM Data Solutions has, we started looking at other models we could train and what industries would care about those models — predicting the appearance of a notice of default, or predicting the appearance of a trustee sale. If we can predict that a month or two in advance it is highly valuable for real estate investors.
There are a lot models we can create utilizing different pieces of ATTOM data. The challenge is to identify which of these models service the most industries and which industries are willing to pay the most for the models.
We want to be that fourth-generation marketing platform. First generation was direct marketing followed by online leads and then list optimizers who narrowed down a list of 100,000 homes to 20,000 homes that were higher-quality leads based on basic property and ownership characteristics. Our fourth-generation platform, driven by artificial intelligence and machine learning, will narrow down that same list of 100,000 homes to just 1,000 homes that we can predict with a high degree of certainty will represent revenue-generating opportunities for our clients.
When we went out and created our team we really wanted someone with that AI and machine learning expertise … that next level of guy. We got very lucky in finding somebody we had a relationship with and thus being able to go to market very quickly.
How is the marketplace responding to Likely.AI products/services, specifically those integrating ATTOM data?
We started conceptualizing these types of products as early as 2007, before the now-ubiquitous data science nomenclature even existed. We sat on it until technology and the industry caught up — when industries were willing to incorporate data into their growth strategy. Until that happened, the industry was talking about data like high schoolers talk about sex. Everybody was talking about it but nobody was really doing it.
Once we believed market forces had caught up to our vision, we did a soft launch in Q4 2017, providing 10 leads a week per zip code for our beta customers. We got great feedback from this launch. One customer in South Carolina knocked on 10 doors and talked to two people — both wanted to list. A Realtor in Contra Costa County, California got four really solid follow-up leads and one called back with three properties to list from one hour of cold call prospecting.
The key piece of what we do is to identify the people who haven’t done anything about transacting on their home but they are thinking about it — that’s what we’ve found we’re able to predict well.
I knew the predictive modeling was working when my name was highly ranked in the last run for Travis County, Texas, even though my wife and I don’t have any of the regular indicators in terms of years owned, size of home, age of children, etc. But the model was correct in that we are seriously considering a move!
Why did Likely.AI decide to use ATTOM Data Solutions?
We were evaluating ATTOM against another company, and really what it came down to for us was model accuracy and price. At a baseline, we wanted to make sure that our models would function as well or better with ATTOM data and that ATTOM data coverage was comparable or better than the other source. ATTOM passed this initial check with flying colors.
During the evaluation process we also discovered some additional datasets ATTOM offered outside of the standard public record property datasets that would be very beneficial to our models. That was of course icing on the cake.
On the pricing front, our rep John — whom I’ve known for many years and is a great industry resource — worked with us closely to fairly and flexibly price the data licensing agreement so that we were getting the data and terms that fit the specific objectives of Likely.AI.
Lastly it became apparent during the evaluation process that the level of customer service we would be getting from ATTOM far exceeded what we were getting from the other company. When dealing with that other company, we had to wait months for something to get done. We had to sit on our hands for months.
That high level of customer service continued after we signed on as an ATTOM client. As with any data provider, there are always challenges that arise, but ATTOM has been quick to respond to and address those challenges. As long as I have the support of my data provider to solve a problem, that’s what I need. As long as it’s getting resolved in a timely manner. That was something that was a deciding factor for us.
What has been your experience with the data delivery?
The data delivery was actually a pretty smooth process; there weren’t a lot of errors, which is a testament to the robust data management process that ATTOM employs — merging all the various sources together to create one superset of property data.
What has been your experience with the data quality?
Each property in the ATTOM Data Warehouse is assigned a unique ID that persists across all records associated with that property, and that ensures that the data quality and quantity is consistent across all areas.
ATTOM has also been willing to listen to — and work on implementing — our recommendations to improve the data for our specific use case. That type of responsiveness is far above and beyond what we have experienced with others in the industry.
What has been your experience with customer service?
Our rep John carries around two cell phones, and he gave me both cell phone numbers and told me to call him any time. There was nothing that ever fell through the cracks. And that was very important to me being in a detailed-oriented space.
Having someone at the customer support level being able to understand and being able to get us in touch with the right people very quickly is key to our success, and ATTOM has excelled in providing that high level of customer support.
Please contact us if you have questions about the underlying data referenced in this article, or would like to have access to that data in the form of custom reports, API or bulk files.