Here in Pecan, to start your modeling journey faster and clearer, we created a set of templates for you to choose from.
Let's go over what each template means and which one you should use.
How to choose the right template
As a best practice, we recommended first writing in simple words what we are trying to achieve with our model.
For example:
"I want to know who & how many users will cancel their subscription plan in 6 months from now"
After we write our use case, we can continue to choose the right template for our model.
Tip:
You can also use our on-screen guidance that will help you choose which template is best for you, click here to see more.
Binary or Regression
We offer two model variants: binary and regression. The appropriate model type will be selected automatically when you choose the relevant template that aligns with your use case.
Binary Classification Models
Binary models predict if an input belongs to one of two classes. For instance, they can determine if a customer will cancel their subscription (True) or not (False). They categorize inputs into positive ("1" or "True") and negative ("0" or "False") classes.
To learn more about the probability score for our binary model, click here
Regression Models
Regression models involve predicting a continuous value using input features. For instance, they can estimate the potential revenue a customer might generate based on various input factors. This might result in a range of values, such as $122, $80, $261, and so on.
Binary (True or False) Templates | Regression (Value) Templates |
Retention | Lifetime Value |
Conversion | Revenue Forecast |
Churn | Engagement Decrease |
Subscription Renewal |
|
Upsell |
|
High Value Customers |
|
Chargeback |
|
Event based or Frequency based
Event based models make predictions based on user actions or occurrences, while Frequency based models prioritize timely insights by predicting outcomes at regular intervals, which is especially useful when time is crucial for predictions.
Frequency based
Frequency-based templates are utilized when our business use case is time-dependent, and we intend to conduct predictions at regular intervals of X time units.
For example:
I would like to predict every week, the likelihood of a customer to churn within the next 14 days.
Event based
Event-based templates are utilized when our business use case relies on specific user actions, and we aim to conduct predictions based on those actions.
For example:
I would like to predict 2 days after a transaction, the likelihood of the customer canceling the transaction.
Frequency Based Templates | Event Based Templates |
Retention | Lifetime Value |
Churn | Conversion |
Subscription Renewal | High Value Customers |
Upsell | Chargeback |
Engagement Decrease |
|
Revenue Forecast |
|
Starter templates vs Advanced templates vs Blank
Our Starter and Advanced templates were created for you to start modeling faster, or if you feel comfortable with SQL you can start from blank.
Starter
The starter templates are ideal for your initial journey with Pecan. Within these templates, you'll discover commonly used use cases and straightforward text-based formats for each scenario.
Advanced
Once you've established a model and gained a deeper grasp of Pecan's functionality, we recommend exploring our advanced templates. Similar to the starter selection, these templates offer enhanced customization options, enabling you to align them more closely with your unique use case.
Note:
Each template has its own required tables for it to successfully predict and generate a successful model.
Blank
If you want to create your own model and start with an empty SQL code, you can do so by clicking "Start without a template" at the bottom.
Templates description
Below, you will find Pecan's templates and their recommended use case.
Retention
Binary
Frequency Based
Retention is the act of keeping or maintaining a behavior over time.
Predictive question:
Our Retention template is recommended to use when we want to predict if:
X (Customer) will still be Y (Active / Not Active) in Z (Month/Week/Day) time.
Lifetime Value
Regression
Event Based
Lifetime value helps us understand what will be the total value of something at a certain point in time.
Predictive question:
Our lifetime value (LTV) template is recommended to use when we want to predict:
On day X after the customer's first activity, what will be the total value of that customer on day X.
Conversion
Binary
Event Based
Conversion helps us to predict if something will change its status after a certain amount of time when the change is reflected in another table.
Predictive question:
Our conversion template is recommended to use when we want to predict:
What is the likelihood that a customer on day X after their first activity, will convert within the next X days.
Churn
Binary
Frequency Based
Churn is the act of stopping or suspending a behavior over time.
Predictive question:
Our churn template is recommended to use when we want to predict:
Every X time, what is the likelihood that an active customer will churn within the next X days.
Subscription Renewal
Binary
Frequency Based
Keeping or maintaining a behavior once a time period comes to an end.
Predictive question:
Our Subscription Renewal template is recommended to use when we want to predict:
From X days after the subscription begins and up to X days before it ends, on an X (ex: weekly) basis, what is the likelihood that the customer will renew their subscription.
Upsell
Binary
Frequency Based
Predicting every Month/Week/Day the likelihood a new behavior will happen.
Upselling is the practice of encouraging customers to purchase higher-priced or additional products or services beyond their initial choice. In Pecan, we can analyze the likelihood of a customer performing an upsell.
Predictive question:
Our upsell template is recommended to use when we want to predict:
On an X (Month/Week/Day) basis, the probability that an active customer will perform an upsell activity within the next X days.
High Value Customers
Binary
Event Based
Predicting the likelihood of something to be top performing after a certain amount of time.
The High Value Customers template helps create predictive models to pinpoint customers with the highest potential for exceptional performance. These predictions assist in prioritizing customer focus, highlighting those of high value.
Predictive question:
Our High Value Customers template is recommended to use when we want to predict:
On day X after their first activity, which customers are likely to be in the top X % when reaching day X.
Chargeback
Binary
Event Based
Predicting if there will be a status change for someone/something when the change is reflected in the same table.
Chargebacks occur when a customer requests a transaction to be reversed and funds are returned. Predicting chargebacks helps businesses prevent fraud and prepare for potential problems by identifying risky transactions.
Predictive question:
Our Chargeback template is recommended to use when we want to predict:
On day X after a transaction, what is the likelihood of it becoming a chargeback within the next X days.
Revenue Forecast
Regression
Frequency Based
Regularly predict the value of something for the following X days.
Revenue forecast helps us to predict or estimate how much money a business or organization expects to earn in a specific period.
This helps businesses plan better, make smart decisions, and set goals for growth.
Predictive question:
Our Revenue Forecast template is recommended to use when we want to predict:
On an X (ex: weekly) basis, what will be the sum of the revenue in the next X days.
Engagement Decrease
Regression
Frequency Based
Consistently forecasting the fluctuation of a value for the upcoming X days.
Engagement decrease refers to when the level of interest, interaction, or involvement with a product, service, or platform goes down. Predicting engagement decrease allows us to anticipate and address issues that might lead to reduced customer interest and interaction.
Predictive question:
Our Engagement Decrease template is recommended to use when we want to predict:
On an X basis, what is the likelihood of a customer who had at least X transactions in the last X days reducing their engagement by X % or more within the next X days?