Model Rollover Process

The model rollover process allows us to account for recent changes in behavior. Actions from the past year and the impact on the outcomes could and should look different than what occurred three or five years ago. With a model rollover, we'll introduce the latest year of known outcome data to rebuild your models and most likely drop the oldest year of data previously used in training the model.  
This enhances model performance and the understanding of individual student behaviors.
Read more on Othot’s approach to the FAFSA delays SAI transition below.

Data stewards or your data team will add the outcomes to the standard files that are already sent to us.
It can be as simple as providing the outcomes in the same fashion as in prior years, typically through Yes/No columns or dates flagging enrolled, retained, or graduated students.
It's good practice to verify that the data reflects the outcomes validated by your institution, which may involve your core Othot team.

Post-census day, institutions provide the known outcomes for the student populations that we have been predicting on over the last term or year.
The most recent year of data (and associated outcomes) is then used by our data science team to inform and rebuild the models.
This is part of our commitment to you to have the most performant and relevant models.
If there are significant changes to your model based on the most recent year of data, you may receive a request for a model validation meeting prior to updating the model.
Note: this process is intended to capture outcome data and rebuild your models. It's not a time to introduce new data. Additional updates should be planned separately with your Client Success Director and Data Curator.

As soon as your data reflects the final outcomes post-census day, send that data to us, and let your Client Success Director know it's complete. 
We complete the model rollover process in order of data received from our clients. 
While we make every effort to expedite completion of this process, your help is required to make that happen.

Follow the same process as you have throughout the year, whether it's sending files through the platform or uploading through your institution's SFTP site.

This year’s FAFSA delays have impacted institutions in various ways. The approach we will take for each client will largely depend on how many FAFSAs were received for the given year and how filers/non-filers yielded compared to prior years. This review will be part of the rollover process. If these items are relatively consistent with history, there should be no issue in continuing to include all this information in the new models. For clients with significant differences in either of these items, we may consider an alternate approach, such as excluding the 2024 cohort from the impacted phase models (e.g., Admit, Deposit) or excluding the problematic FAFSA features from the impacted phases.
We understand that given the delays in receiving FAFSA information, institutions may have elected to only import a subset of FAFSAs received into their internal systems, focusing on those still relevant in the funnel at a certain point in time. However, to ensure we have the most complete and accurate data for the rollover process, please ensure that all FAFSAs submitted to the institution are imported into your systems and delivered to us via the regular data feeds.

The shift from EFC to SAI has brought additional considerations for the models. For clients who have previously delivered historical SAI data to Othot, no additional steps are necessary as the models have already been updated to reflect this information. For clients who cannot provide historical SAI but can provide it for the 2024 cohort, we will review if incorporating this data into the models is feasible. Since this would require us to build the impacted phases (e.g., Admit, Deposit) using only the 2024 cohort where SAI data is available, this determination will be made alongside considerations for handling any issues introduced by the FAFSA delays.