Analytics-Ready Marts¶

Marts are the final dbt contracts used by Tableau dashboards, notebooks, and the project writeups. They are intentionally closer to business questions than raw data: market opportunity, demand drivers, category health, customer segments, repeat behavior, and transaction anomaly review.

Layer purpose

The mart layer answers: what clean, analysis-ready tables should Tableau and modeling notebooks consume?

Marketplace Growth¶

Mart	What it provides	Analysis output
`mart_geo_market_opportunity`	Market-level performance, growth, supply, macro, adoption, reliability, and opportunity scoring	Marketplace Analysis

This mart supports the market opportunity story by combining current Nova performance with external country context and supply signals. It is where the project moves from "which markets are large?" to "which markets are attractive growth candidates?"

Demand Drivers and Modeling¶

Mart	What it provides	Analysis output
`mart_geo_market_category_day_features`	Market-category-day feature table with demand, weather, calendar, macro, supply, and service signals	Demand modeling notebook, Local Demand Drivers
`mart_geo_weather_category_effects`	Bad-weather transaction and GMV lift by category and market	Local Demand Drivers
`mart_geo_weather_category_sensitivity`	Rain sensitivity and weather-correlation summary	Local Demand Drivers, market clustering
`mart_geo_category_hourly_demand`	Hourly demand curves by market and category	Local Demand Drivers
`mart_geo_market_cluster_features`	Market-level feature table for strategic clustering	Local Demand Drivers

These marts support the demand-driver narrative: weather matters in some categories, but service supply and market context explain more of the demand picture.

Category and Service Performance¶

Mart	What it provides	Analysis output
`mart_category_performance`	Category-level GMV, amount share, success, failure, refund, rating, and risk-segment counts	Category Performance
`mart_category_market_performance`	Category performance split by market	Category Performance
`mart_marketplace_service`	Service-level amount, reliability, rank, and risk segment	Category Performance
`mart_service_monthly_performance`	Monthly service-health tracking	Category and service monitoring

These marts make category performance operational. They show not only which categories are large, but also where reliability leakage, refunds, failures, and critical services need attention.

Customer Segmentation¶

Mart	What it provides	Analysis output
`mart_customer_overview`	Customer profile, membership, acquisition, lifecycle, transaction count, completed revenue, average order value, and recency fields	Customer Analysis
`mart_customer_segments`	RFM scores and business-friendly segments such as Champions, Loyal Customers, Potential Loyalists, At Risk, and Lost Customers	RFM Analysis

These marts support the customer strategy pages. They separate user count from customer value, showing which groups are worth retaining, upgrading, or reactivating.

Repeat and Risk Analytics¶

Artifact	What it provides	Analysis output
`mart_service_repeat_behavior`	Repeat users, repeat interactions, repeat amount, and repeat dependency by service/category	Customer Repeat & Risk
`ml_transaction_fraud_features`	Transaction-level behavioral features for anomaly detection	Fraud anomaly notebook
`ml_transaction_fraud_scores_sample`	Isolation Forest anomaly scores written from the notebook	Customer Repeat & Risk
`mart_transaction_fraud_tableau`	Tableau-ready transaction context joined to anomaly scores and suspicious flags	Customer Repeat & Risk

The risk workflow combines dbt and notebook modeling. dbt builds the transaction-level feature table, the notebook scores anomalies, and the final mart joins those scores back to business context for review.

Fraud interpretation

The fraud workflow is unsupervised anomaly detection. The output flags suspicious transactions for review; it does not prove confirmed fraud.

Quality Controls¶

Mart tests check unique business grains, required metrics, accepted customer segment labels, and core anomaly-score fields. Custom dbt tests also validate geography grains and reconciliation between market-day, category-day, and feature-mart totals.

Why this matters

The marts are the bridge between engineering and storytelling. They make the Tableau dashboards and project writeups credible because the business metrics come from tested, reusable models rather than one-off dashboard calculations.