What is a model retrain?

A trained model is a snapshot of what the history showed up to a cutoff. A retrain moves the cutoff: the model is trained again on everything up to today, so the latest weeks of spend and sales update every channel's estimate. It is not the same as new data arriving; data arrives daily and can be reported on at once. The retrain is when the model's beliefs are updated with it.

Sales and marketing data come in daily through the managed integrations. Once a month the model is retrained on all of it, and each retrain includes monitoring for changes in channel efficiency over time. The recommendations are refreshed, an Odins analyst reviews them, and the customer gets a note on what changed and what we would do differently. The team acts, the results flow back in, and next month's retrain is better for it.

Because roughly two years of history sit behind each new month, the model changes slowly and stays stable; the newest data weighs somewhat more, but one month cannot flip the picture. Every earlier version stays available, so the forecast a model made six months ago can be checked against what has happened since.

Monthly is as often as a budget decision changes, and the model is trained on weekly data, so retraining more often would add nothing and would make the model look busier than it is. What is not monthly: the data, which flows in daily; reporting, which is current whenever it is opened; and scenarios and forecasts, which run any time in the platform or from your own AI assistant. Only the retrain and the reviewed recommendation follow the monthly rhythm.

Creative changes and viral moments are handled the same way: the model assumes a channel works as before until the next retrain shows otherwise, which is the honest position when one good week is not yet a pattern.

• Forecast verification: Forecast first, compare with actuals afterwards. The only honest way to know whether a model is right.
• Posterior: The model's answer: the prior after the data has had its say, with a most likely value and a range.
• Scenario planning: Forecasting what a budget would return before committing it, and comparing several futures side by side.
• Marketing mix modeling (MMM): The statistical method behind everything else on this list: spend in, sales out, every channel on equal terms.

Go one level deeper: Staying accurate. Managed MMM service. Or go back to all terms in the glossary.