One of our most high-profile applications of deep learning is Relitix’s “Switch Risk” flags. Every month our algorithm examines the recent transaction history and market behavior of over 1.2 million agents nationwide and grades each on their likelihood to change brokers in the near future. Our deep learning model compares each agent’s pattern of behavior to those of agents who have moved and assesses the similarity. We award each agent a red, yellow, or green flag with red denoting those agents at highest risk for leaving.
The system works quite well – surprisingly well. How well? We are predicting over 25% of all agent movement before it happens. We looked in our 13 largest MLS’s and found all the $1M+ per year agents who had changed brokerages from February 2022 through February 2023 – nearly 30,000 of them as it turned out. We then looked to see what percentage of those brokerage-switching agents had been red or yellow flagged in the 6 months prior to them making the switch. The answer ranged from a low of 23% in Northstar MLS to a high of nearly 28% in Canopy. The more information the model has, the better job it does in predicting this behavior. When we look at agents doing at least $5M in business our prediction rate increases to over 30% and rises even further for agents above $10M in annual production.
While it remains extremely difficult to predict what a given person might do based on MLS data, this study proves that we can make valuable predictions in aggregate. People are unpredictable, groups are not. Brokers who have access to this sort of crystal ball can compile target-rich lists for recruiting and retention. This is an excellent example of how AI and machine learning can provide an edge for brokerage management in this highly competitive industry.