Looking to learn more about common terms we use check out: https://knowledgebase.faraday.io/en/articles/4972409-predictive-modeling-terminology

General questions

What is a model refresh & how often will my model(s) be refreshed?

  • A model refresh is when we prompt the model to re-run and provide an updated analysis and report.
  • Model refreshes take place typically every 1-3 months, but it depends, oftentimes on the volume of new customers/leads that are coming in and at what frequency. Other factors like major business changes such as to the product, price points, marketing efforts can often prompt a refresh as well.
  • Your Account Manager & our Data Science team will monitor this and keep you informed every step of the way.

What can I expect with a model refresh?

  • An updated model report will be provided
  • Your Account Manager & Data Science team will always review the new report and compare to the previous one, identifying any changes or notable items. When all looks good, the new model is activated and can be used

Will you use my 1st party data to build the model?

  • Your 1st party data is used as the training set, where we use the historical action such as the date someone became a customer and train the model to look at those examples.
  • Your 1st party data is not apart of the modeling, where the model makes decisions, we use our FIG data (300+ attributes) for this.

Model reports

What are features of importance?

  • Feature importances are a window into the character of the training profile. Ranked by relative contribution to the predictive power of the model, the top 10-20 of these are the characteristics that most distinguish individuals who have achieved a business outcome from those who have not.

What do the arrows mean on the features of importance on the model report?

  • Arrows on the top features in a model summary show the directionality of the importance. For example, a ⇧ on Household Income suggests that individuals with a higher income are more likely to achieve the business outcome, or that these things are positively correlated.

Why aren’t arrows showing for all of the features of importance in the model report?

  • Arrows only appear for features with a strong correlation with the objective. The correlation may exist, but just not be strong enough. There may not be a correlation, in the statistical sense.
  • Correlation is a measure of the linear dependence of two things. But there are other patterns that exist that the model picks up... these will be predictive but technically not correlated.

What does the % of each feature mean?

  • The percentages listed for each of the features of importance on the model report represent the % that contributes to the overall prediction; think about all of the FIG features totally 100%

Model / Outcome types

Lead generation

Lead generation, customer acquisition or prospecting are used to find more of "something"

Examples include -

  • finding more likely customers
  • finding more likely leads
  • finding more likely customers who purchase 2 or more times

Lead conversion/customer engagement

Lead conversion or customer engagement is where we take a pool of your existing customers (or leads) and score them on their likelihood to take a specific action

Examples include -

  • the likelihood that the lead will convert
  • the likelihood that the customer will buy product A
  • the likelihood that the customer will make a second purchase

Retention

Retention or churn models focus on identifying an existing customer and their likelihood to no longer be a customer. This is typically used in subscription businesses but can be used for financial institutions, and other eCommerce services.

Examples include -

  • likelihood to stop their subscription
  • likelihood to cancel their account (bank/credit union)

Did this answer your question?