Predict Seller Intent From Subscription Data
Problem: Predict Seller Intent From Subscription Data
You are provided with a dataset containing approximately 5,000 new subscription records from a website-building platform. The primary business objective is to identify factors associated with a new user becoming an active seller and to develop an interpretable predictive model for seller intent.
Dataset Description
The dataset, seller_intent_take_home_dataset.csv, includes the following columns:
subscription_id: Unique identifier for each subscription.subscription_start: Timestamp indicating when the subscription officially began.first_sale: Timestamp indicating when the user made their first sale, if any (nullable).days_in_trial: Integer representing the number of days the user spent in a trial period.subscription_plan: Categorical field for the selected subscription plan.subscription_period: Categorical field for the billing period (e.g., monthly, annual).discount_amount: Numeric value for any discount applied to the subscription.country: Categorical field for the user's country.site_topic: Categorical or free-text field provided by the user during onboarding, describing their site's topic (nullable).site_need: Categorical or free-text field provided by the user during onboarding, describing their reason or need for creating the site (nullable).user_agent: Raw browser user-agent string.
Important Data Issues
- Target Variable Definition: There is no pre-existing
is_sellerlabel. You must define the target variable using thefirst_saleandsubscription_starttimestamps. - Timestamp Inconsistencies: Some records exhibit inconsistent timestamps, including instances where
first_saleappears to occur beforesubscription_start. - Missingness and Long-Tail Categories:
site_topicandsite_needcolumns have substantial missingness (approximately 28% to 32%) and contain many long-tail categories. - Raw User-Agent String: The
user_agentcolumn is a raw string, requiring extraction of useful features such as device type, operating system, and browser.
Deliverables
Your final notebook or report should be concise, suitable for clear presentation, and prioritize data reasoning, feature engineering, and interpretability over complex hyperparameter tuning. Specifically, you must deliver the following:
- Target Variable Definition: Clearly define the target variable and explicitly describe your approach to handling timestamp edge cases.
- Exploratory Data Analysis (EDA): Conduct EDA focusing on data quality, missingness patterns, class balance, and feature distributions.
- Feature Engineering: Engineer meaningful features from categorical fields, timestamps, discount amounts, trial lengths, and the
user_agentstring. - Model Training and Evaluation: Train an interpretable model, such as logistic regression, and evaluate its performance using appropriate classification and ranking metrics.
- Interpretability and Explanation: Explain the primary drivers of predicted seller intent, carefully distinguishing between predictive associations and causal claims.
- Presentation: Prepare a short presentation summarizing your assumptions, modeling choices, results, identified limitations, and recommended next steps.
AI interview prep
Core dilemma
Balancing the robust definition of a business-critical target variable and sophisticated feature engineering from messy, limited real-world data, all while adhering to a strict interpretability requirement for actionable business insights.
3 tradeoffs and 3 stress tests locked
Interview prep
Unlock the interviewer X-Ray rubric, tradeoff matrix, stress tests, clarifying questions, and key concepts.
Predict Seller Intent From Subscription Data
Problem: Predict Seller Intent From Subscription Data
You are provided with a dataset containing approximately 5,000 new subscription records from a website-building platform. The primary business objective is to identify factors associated with a new user becoming an active seller and to develop an interpretable predictive model for seller intent.
Dataset Description
The dataset, seller_intent_take_home_dataset.csv, includes the following columns:
subscription_id: Unique identifier for each subscription.subscription_start: Timestamp indicating when the subscription officially began.first_sale: Timestamp indicating when the user made their first sale, if any (nullable).days_in_trial: Integer representing the number of days the user spent in a trial period.subscription_plan: Categorical field for the selected subscription plan.subscription_period: Categorical field for the billing period (e.g., monthly, annual).discount_amount: Numeric value for any discount applied to the subscription.country: Categorical field for the user's country.site_topic: Categorical or free-text field provided by the user during onboarding, describing their site's topic (nullable).site_need: Categorical or free-text field provided by the user during onboarding, describing their reason or need for creating the site (nullable).user_agent: Raw browser user-agent string.
Important Data Issues
- Target Variable Definition: There is no pre-existing
is_sellerlabel. You must define the target variable using thefirst_saleandsubscription_starttimestamps. - Timestamp Inconsistencies: Some records exhibit inconsistent timestamps, including instances where
first_saleappears to occur beforesubscription_start. - Missingness and Long-Tail Categories:
site_topicandsite_needcolumns have substantial missingness (approximately 28% to 32%) and contain many long-tail categories. - Raw User-Agent String: The
user_agentcolumn is a raw string, requiring extraction of useful features such as device type, operating system, and browser.
Deliverables
Your final notebook or report should be concise, suitable for clear presentation, and prioritize data reasoning, feature engineering, and interpretability over complex hyperparameter tuning. Specifically, you must deliver the following:
- Target Variable Definition: Clearly define the target variable and explicitly describe your approach to handling timestamp edge cases.
- Exploratory Data Analysis (EDA): Conduct EDA focusing on data quality, missingness patterns, class balance, and feature distributions.
- Feature Engineering: Engineer meaningful features from categorical fields, timestamps, discount amounts, trial lengths, and the
user_agentstring. - Model Training and Evaluation: Train an interpretable model, such as logistic regression, and evaluate its performance using appropriate classification and ranking metrics.
- Interpretability and Explanation: Explain the primary drivers of predicted seller intent, carefully distinguishing between predictive associations and causal claims.
- Presentation: Prepare a short presentation summarizing your assumptions, modeling choices, results, identified limitations, and recommended next steps.