Reducing Customer Acquisition Cost for a Marketing Company

eCommerce ($20m+ revenue)Second largest mobile loyalty app providerIn North America with over 23 million customers 

This is some text inside of a div block.
This is some text inside of a div block.

About the Customer

eCommerce ($20m+ revenue)

Second largest mobile loyalty app provider

In North America with over 23 million customers 

The Pain point 

  • Runs number of marketing campaigns round the year and requires dedicated data scientists/analysts to build checks to monitor and alert fraud as the campaign gets underway and prevent 
  • Manual intervention to go through dashboards, detect and block fraudsters  
  • 10 mins spent on each fraudster for validation (180+ users flagged on some days) 
  • Type of fraud keeps changing based on data and nature of acquisition campaigns - ads, seo, gift cards, affiliate, invitation perks

What is their starting point / ecosystem / tech stack 

  • Java, Python, Docker, Kubernetes 
  • Streamsets
  • Google Cloud Platform
  • MySql, BigQuery
  • Third party machine learning tools - R, Python

Compelling Event (Trigger point)

  •  Internal initiative to drive towards high-value user acquisition 
  • Lack of understanding in behavioral data from user leaves with less confidence on existing fraud rule engine 
  • Data collection and analysis largely solved by BigQuery + Tableau
  • Initial fraud rule engine largely static and manually patched for every new discovery of fraud 
  • Customer support Workflows for inspecting and flagging fraudsters is done in another interface ; with communication lag between the fraud rule engine and experts

 Predera Benefit

  • Mature end-to-end AI driven platform for fraud 
  • Machine Learning continuous improvement  using self-learning and updation techniques 
  • Business stakeholders can define their own definitions of Fraud and contribute to the improving AI engine 
  • Customer service get direct alerts in their work environment (Slack) from the AI engine and the feedback updates the model

Evaluation

  • Easy to install and setup
  • Business stakeholders can teach AI for Fraud Engines in 8 hours
  • Able to migrate all existing ML models and rules in less than a week 
  • Integrated with data stores on AWS/ GCP - no extra support needed from IT/ data teams 
  • Didn’t need to re-write workflow integration as it directly integrates with Slack and Hipchat for feedback from Customer support 
  • Dedicated expert level support for onboarding 

Business Impact 

  • Business stakeholder -> fraud analysts -> customer support representatives  loop was tightened, reducing the feedback time by  80% 
  • Saved 280 hours of stakeholder productivity per year 
  • Reduced time to resolution of Fraud by 75% from ~20 mins to ~5 mins
  • Handle more fraud queues and volume with less people 

ROI 

“We would need two full-time data scientists with knowledge of marketing and acquisition campaigns for 6 months just to build out our Fraud Engine capability let alone the rest of monitoring and continuous automatic improvement provided by Predera. We could spend $300,000 to do that, or just buy Predera and have a mature Fraud and AI / ROI management platform today”