Kantar provides three different levels of model maintenance types:
- “Model Reconfigurations”
- “Model Refits”
- “Model Updates”
Model Reconfigurations
During the customer onboarding, the model we deliver is a configuration of different data feeds, which are called variables. Within those variables are a configuration of different columns, which are called dimensions. The selection of combined variables and dimensions are determined and signed off by the customer during the initial Product Scope and Data Collection stages.
Once a model has been delivered, adding or removing an existing dimension or variable requires a Model Reconfiguration. A Model Reconfiguration is defined as a change in product scope and in model granularity. A model reconfiguration will change historical results as it is in effect, a whole new model.
How often should my model get a Reconfiguration?
We do not recommend our customers to reconfigure a model unless significant changes to the business drivers or industry dynamics take effect.
How long does a Reconfiguration take?
A Reconfiguration takes about 15-20 business days once all of the new data has been collected and validated
Model Refits
Changing data in an already existing dataset- such as adding a new publisher or a new dimension value. A Model Refit is defined as the re-estimation of the model parameters and includes new observations on new publishers, and can therefore change the result of future predictions (as the model will now know more than it did before). A refit is not a change to the actual configuration of the model, but rather a change to the data within the same configuration.
How often should my model get a Refit?
We recommend monthly refits if weekly updates, but can also be quarterly or even bi-annually depending on industry dynamics.
How long does a Refit take?
A refit takes up to 10 working days to complete before it is available in the platform.
A model refit based on changes to historical data changes will require additional days, as both new and all values will need to be evaluated.
Model Updates
A Model Update is defined as the continuous addition of new time-period data, and will not include observations on new publishers or the addition or removal of new variables or dimensions. A model update will therefore not change the model itself - but rather evaluate and explain what happened in a newly added period of time.
How often should my model get an Update?
Our standard (recommended) model update cadence is weekly and not less often than monthly.
How long does an Update take?
Once the new time-period data has been validated, the update process takes 3-5 working days to complete before it is available in the platform.
Please refer to your contracts “Model Maintenance Agreement” for the agreed cadences for model maintenance types. If at any time you have questions about or would like to adjust the agreement, please reach out to your Customer Success team or Product Delivery Manager. Keep in mind that all maintenance types need to be scheduled in advance and may include additional costs.