Note : This a personal conceptual framework with no relation to existing subscription businesses or organizations. This blogpost aims to solve a business problem with technology and not promote any products or services
Customer-Centric Challenge
Assumption : Happy Cart is a leading retail & grocery chain in Australia. Recently they launched a new membership service named HappyCart More and direct competitor to Amazon Prime & WalmartPlus. The service includes in-store & online benefits such as free deliveries for any purchases, special discounts , fuel discounts and much more. Costing at $120 per year or $10 a month, they include a 15-day trial but they need your credit card.
Happy Cart More have decided to offer a 1-month trial to existing customers at the retailer. To keep CAC or Customer Acquisition Costs low, this offer would be provided over email only & unique codes would be available for only 2 days.
Customer-Centric Solution
For the marketers at Happy Cart More, they were looking for a solution that meets the following objectives :
- Augmented Segments : Rule based segments brings the challenges of restrictions & human interventions from marketing ops team. Augmented Segments powered by Machine Learning/Data Science models would learn from customer purchase patterns and find the next best set of members.
- Personalize Offers based on Recent Purchase : To drive relevance email offers are defined into 4 categories – groceries , snacks , coffee/tea & cleaning products.
- Personalized Offer Links : Instead of inserting coupon code on email, each customer would be provided with a unique URL to get started with the journey.
1. How to find relevant customers using ML ?
To drive optimization, we would want to define probability to subscribe for each customer ID/profile based on purchase behavior. As more subscribers are acquired through this channel it would learn from their behaviors and find similar customers. Since both events – purchase & subscriptions would vary daily data models would compete to be the best fit to define the probability. The complete solution would be owned managed end-to-end by the inhouse data science team at HappyCartPlus.
2. How to manage customer data & ML at scale?
HappyCart uses Snowflake to manage customer data that’s updated daily with the latest customer purchase history. Using Snowflake -Databricks Connector the complete dataset would be available to the Data Science team in their Databricks notebook. Then machine learning training pipelines would be established to explore data & compete models to define the probabilities. After training the probabilities would be updated directly into Snowflake tables for activation.
3. Why Databricks?
Databricks offers the ability to process large amounts of data & deploy machine learning jobs end-to-end. It is built upon the public cloud capabilities of AWS & Azure and powered by Apache Spark. Databricks enables collaboration within wide number of data science teams & others enterprise-wide to ensure best practices & unified approach. Also all of the resources are just one click away with pre-configured clusters.
4. How to send personalized emails from Snowflake?
Amazingly, Snowflake has a native FDA connector with Adobe Campaign. After probabilities have been updated in Snowflake, marketing teams can use the existing schema for the data to be available in Adobe Campaign. Downstream workflows would define the personalized content based on the probability & category defined. Also the queries would run in Snowflake and resulting data passed to Adobe Campaign, so no new resources need to be defined.
The End?