Introducing Machine Learning to Real Estate Prospecting with DataFlik
Machine learning presents new opportunities for industries like real estate which are tasked with finding qualified clients in a crowded and ever-changing market. However, the large amount of seller data needed to create an effective machine-learning model can be extremely expensive. This is the challenge Dataflik aimed to solve when they approached NineTwoThree to build a machine-learning model that generates prioritized lists of motivated sellers faster and more accurately than a human expert can, saving real estate wholesalers key time and money.
Real Estate Lead Generation Challenges
Real estate wholesalers looking to find new sellers typically spend a significant amount of time and money finding the contact information for potential sellers in specific real estate markets.
This is an important process for real estate agents to go through to find new leads, clean up their existing lists and bad data, and follow up with cold leads. Agents can do this legwork themselves or pay for real estate prospecting, another significant cost.
Once they have an accurate list, agents send out hundreds to thousands of letters to find the right person that is willing or needs to sell their property.
Unfortunately, traditional real estate prospecting and research involve human analysis of historical data to create very broad lists that result in very low conversion rates. Agents typically send out hundreds of mailers and only receive a few responses. Finding motivated sellers whose homes are not listed on the market yet is something of an art, and this method has a high expense but low return.
Growth in PropTech
While the real estate industry has traditionally been slow to adopt new technology, the last few years have seen significant digital disruption. 2022 saw the highest value of PropTech investment on record.
PropTech startups are poised to capitalize on a massive industry that is finally embracing technology.
The biggest trends in real estate and technology adoption are expected to be:
- Automation: CRMs and real estate-based project management software platforms are already automating many of the day-to-day activities of real estate professionals. PropTech will continue to leverage automation to help find and analyze deals, buy and sell properties, and manage rentals.
- Virtual Reality and Metaverse: COVID-19 forced many real estate agents to embrace technology to allow for virtual open houses and remote experiences. With the metaverse market expected to reach $824.53 billion by 2030, expect real estate to continue to embrace VR.
- Big Data and AI: Researching and analyzing property investments, market opportunities, and new sellers used to take months. Companies like Dataflik are allowing real estate agents to gain vital insights from large volumes of data in minutes.
Introducing Machine Learning to Real Estate Prospecting
Dataflik aimed to leverage machine learning to generate more accurate lists of potential sellers by creating models to predict properties that are more likely to sell. Instead of using human analysis of a small number of data points, ML would allow Dataflik to process a large amount of data using a model that could be continuously tuned to increase accuracy.
Even with professional prospecting and skip tracing services, mailing lists will often include houses with low chances of conversion, outdated listings, and other data that is inaccurate or untimely.
DataFlik’s prospecting tool aimed to filter out these results, and use AI to predict houses that aren’t yet going on the market, are going into liquidation, and other potential opportunities that other real estate wholesalers would not be aware of. The end goal was to deliver a prediction model to provide real estate wholesalers with valuable insights into potential prospects before they are publicly listed.
Real Estate Data Used for Machine Learning Model
Creating an effective machine learning model for real estate prospecting requires a large amount of data. Producing fast results which such a large volume and breadth of data is complicated and technologically demanding. There are millions of properties in the US and each property has about 1,700 data points.
Data points relevant to Dataflik’s model included courthouse records, demographics, geographic insights, household interests, life events, purchase behaviors, sports interests, short-term loan shopping, financial stability, and more.
The first challenge we ran into with Dataflik is that the real estate data needed to achieve the desired model is extremely expensive. In order to justify the purchase of this data we needed to prove that the proposed model would be capable of achieving its goal - before we could purchase the data needed to train the model.
Challenges of Developing Machine Learning Models
The challenge of justifying data expenses in machine learning projects is a common one.
The sheer amount of data can be daunting as well. 72% of organizations report that production-level model confidence will require more than 100,000 labeled data items and 10% will require more than 10 million data items.
Since ML is so valuable for processing massive amounts of data, there are many common use cases where a company would be interested in building a model to process this data before investing in the cost of the data itself. Unfortunately, this is also why 33% of AI or ML projects stall during the proof of concept stage.
A “normal” machine learning project starts with setting goals based on guided data analysis sessions with experts to understand how a human expert would make predictions. The next stage involves cleaning and preparing all available data for modeling.
We then create and evaluate several models based on the hypothesis made from meeting with the experts and coming up with the best-performing model. Then the top models are fine-tuned and re-trained until a high level of confidence is reached.
Building a Mini ML Model as Proof of Concept
Dataflik came to NineTwoThree with the goal of building a model as proof of generating an effective motivated sellers mailing list - without developing the full model - to justify the purchase of such data. NineTwoThree created a “mini model” that served as a proof of concept and de-risked the expensive data purchase for Dataflik.
Creating this mini model started with interviewing a human expert in real estate prospecting to help identify the most powerful data points for producing a working baseline model. Traditional prospecting relies on human intuition and a much smaller set of data points to make predictions. This serves as a limited starting point for our baseline model.
Based on the data needed to make this ML model, we requested sample data sets directly from vendors and asked questions to ensure the data was valuable.
Working with an AWS-certified partner can greatly speed up processing time and produce high-quality results quickly and efficiently compared to what could take data scientists months.
Once the mini model proved the superior accuracy and efficiency of using machine learning Dataflik was able to justify investing in the 42 billion data points used to train the full model.
Dataflik’s ML-Powered Real Estate Solution
NineTwoThree created an ML model that generates prioritized lists of motivated sellers faster and more accurately than human experts. Real estate wholesalers gain a tremendous advantage in better targeting sellers to reduce “spray and pray” marketing expenses and close more deals.
“The real estate market becomes more competitive and saturated every day, squeezing margins each year. DataFlik's product performs better each month with our proprietary algorithm driven by our AI and machine learning technology. The system continuously adapts to target the owners most likely to sell as the market changes in real-time.”
The growth of PropTech and AI in the real estate market allows agents and wholesalers to avoid spending unnecessary money chasing unqualified leads. With NineTwoThree’s help, Dataflik overcame the data challenges of training a machine learning model and they are now poised to revolutionize real estate business marketing operations.