Relitix's switch risk grade assesses the likelihood of an agent to change brokerages in the next 3 months. Relitix pioneered the use of predictive modeling to assess agent movement likelihood in 2018 and has been continually producing this metric since then.
The model
The switch risk grade is calculated using an AI technique called
deep learning. This technique, at it's most basic level, is a form of superhuman pattern recognition. Our algorithm was shown millions of examples of agents who had either stayed or left their brokerage. The model learned by example what sorts of behavioral patterns were associated with agents who left their brokerages and those who did not. When shown new data the models applied this learning and assesses how similar the behavioral pattern of the new agents is to the patterns it learned were associated with agent movement.
What sort of data is used? We use only those available in the MLS dataset: listings, closings, active listing count, expirations, and so on for an extended period of time. During the development of the model many datasets were tried and rejected as not improving the predictability of the model as measured against a test set.
How accurate is it
During development of predictive models a subset of data is held back from the training process. When the model is trained (taught itself what to look for) it is run against this reserved set of examples to see how well it "generalizes" or performs against new, unseen data. In our tests red-graded agents switched brokerages at 11x the rate of the average agent. Yellow agents switched brokerages at 5x the rate of an average agent. Described differently, a list of 100 red-flagged agents had more than 10x the number of agents who would go on to switch brokerages than a randomly selected list of 100 agents. In 2023 Relitix did a retrospective analysis of agents who switched brokerages in 2022 and 2023. We found that 23%-27% of agents who had changed brokerages during that time had been flagged as yellow or red at least once in the prior 6 months, the variation depending on the MLS.
What causes a red/yellow agent
Because we use a true deep learning model which has trained itself, we don't know for sure. This is the downside of AI: sometimes its a black box. We do know that certain things will almost always flag an agent: someone who strangely stops taking new listings and starts expiring things out as quickly as they can for example. New agents who have a meteoric rise in production wind up flagged too, likely because people start recruiting them heavily. Others, seem a mystery.
How does this compare to "brand X" switch risk
Since Relitix developed the industry's first predictive agent switching model in 2018 a few other companies and franchises have tried their own version of this calculation. Nearly all use a different, more primitive, and easy to implement rules-based approach. This decision-tree based approach is termed
hueristics. They human intuition and hard-coded rules to judge complex predictions like switch likelihood: for example "if an agent is down 25% or more year of year and is in an office shedding agents then they are at risk". While more effective than pure guessing, this approach bypasses the huge gains realized by the fields of machine learning and AI and, in our experience, yields much less accurate results.
How do I use this metric
Switch risk is not an AI crystal ball. People's behavior can change for many reasons, not all (or maybe even most) of them related to engineering a brokerage switch. As with all our metrics, this is another tool in the toolbox. Most red-flagged agents are not going anywhere. If, however, you want to generate a list of agents with a disproportionate number of people looking to move, this is a tool which puts the power of modern AI in your corner. You can use this to create a "target rich environment" of agents very quickly.