ML Solution for Marketing Services

Client

A provider of innovative marketing campaigns to the financial sector

Industry

Marketing, Financial Services

Services

Data Science, Data Management, Data Analytics, ML

Tech

AWS Services (AWS S3, AWS Glue, AWS SageMaker), MySQL

Challenge

Leads are the lifeblood of any business. No matter the size of the firm or the marketing budget, targeting the right prospects is a key priority for lenders in the financial sector, too, where competition for new customers is fierce. A US-based provider of innovative marketing services to banks and other financial institutions was using the power of ML to take lenders to the leads that matter. Specifically, they offered ML-powered direct mail lists for mail marketing campaigns. The company had an engineering team that maintained a data warehouse storing raw customer data. However, they had to outsource data science services to a vendor for ML-driven lead scoring to rank prospective customers who are most likely to convert. The external vendor’s ML approach to lead scoring was what the client called a black box for them, with the processes from input to output lacking any transparency. The client wanted to bring ML capabilities for marketing in-house and was looking for a trusted partner in ML development. They found us.

Our task was to:

●

Dive deep into the client’s business context to identify key business priorities, risks, and constraints

●

Evaluate their current models and map data

●

Design an end-to-end ML solution in AWS

●

Build a roadmap for further improvements

●

Provide training for the client’s engineering team

●

Prepare comprehensive documentation for ML knowledge transfer

Our approach:

An end-to-end AWS-based ML solution for marketing campaigns that has provided the client with in-house ML capabilities for scoring leads while getting better accuracy than delivered by their previous ML vendor. Our approach to building the solution can be summarized as follows:

●

Evaluation of old ML models to identify metrics for each model

●

Exploratory data analysis

●

ETL processes using AWS Glue to extract and prepare data for two data pipelines: ML model training and data scoring. The automated processes were designed to save the effort and time of the in-house engineering team on data preparation and give the client more operational flexibility

●

ML model training with AWS SageMaker, with dozens of experiments organized; creation of one comprehensive ML model trained using all historical data

●

ML model deployment in production

●

Product improvement roadmap outlining recommendations on enhancing ETL processes, ML model optimization, and using the solution as the basis for building a Software-as-a-Service platform that would allow the company’s clients to score leads on their own, with no engineering skills required

Impact

●

Better lead scoring accuracy

●

Cost savings, with the client now paying only for actual AWS resources consumed instead of engaging an external vendor every couple of weeks for a lead-scoring task

●

Operating model flexibility

●

In-house ML capabilities with complete ML knowledge retention

●

Transparency in ML model training

●

The foundations for transforming the product into a SaaS solution