Case Study

Effective Targeted Campaigns with Machine Intelligence


We worked with a utility provider to improve their payment operations by applying machine learning on their customer database. Our personalized marketing campaign drove a 47% increase in automated debit signups, and a 30% improvement in payment timeliness versus the control group in an A/B test run on a Metro Manila pilot group.

  • 47%
    increase in
    signups vs. control
  • 30%
    improvement in
    payment timeliness
    vs. control

Schedule a Consult

Introduction

Market segmentation

Organizations have long employed market segmentation methods with the aim of understanding how and why the public engages with their brand. Traditional market segmentation are typically based on demographics, location, or attitudes approaches. The challenge with most segmentation models is that they are based on static data, and are at best proxies of underlying customer behavior.

Machine learning, built on the multidisciplinary methods of data science, can help forward-looking organizations uncover behavioral patterns and associations previously hidden behind the overwhelming universe of big data. Armed with these behavioral insights, organizations can deploy highly targeted campaigns to influence behavior.

The Challenge

Optimizing Customer Behavior Using Data from Multiple Sources

In the Philippines, 86% of households do not have a bank account and only 2% have a credit card. This means that whenever the monthly utility bill is due, customer will need to go to a physical payment center to pay in cash or deliver a check. The result is that customers are often late in paying their utility bills with some outstanding payments exceeding 60 days.

Our client sought a data-driven solution that would achieve two objectives:

  1. Improve payment timeliness
  2. Increase signups to their automated payments system


Day Sales Outstanding (DSO), the average number of days needed to collect payment after a service is delivered, is an important metric used by our client to monitor its financial health. Low DSO means fewer days taken by staff to collect money, freeing up company resources for investment, whereas high DSO means more time spent hunting for payments, increasing costs while dragging down employee productivity.

Our client wanted to improve enrollments in Automated Payment Arrangement (APA). APA enables their customers to get their monthly bill auto debited from their bank account or auto charged to their credit card on time, every month. Any customer that signed up would by definition, always pay their bill on time.

At the start of the engagement, some outstanding payments were known to exceed 60 days and only a fraction of existing customers had enrolled in APA. How could the company use its heaps of historical customer and payment channel data to drive better outcomes?

The Solution

Segment-specific Understanding of Customer Payment Behaviors

Data Consolidation and Pre-processing

We consolidated just under 100 GBs of data, everything from socioeconomic and geographic data to payment behavioral data. Our data science team filtered and cleaned it for subsequent analysis.

Behavioral Clustering and Factor Analysis

Our data science team used machine learning clustering models to discover structures and hidden patterns within the client’s data. After a set of highly correlated customer features were identified, they were passed through a clustering algorithm to distill four previously unknown customer segments. A regression analysis of each segment’s features revealed what factors were most likely to contribute to certain outcomes—that is, paying on time or enrolling in APA.

Testing and Validation

Guided by insights from the data, different versions of ad content aimed at improving payment timeliness and APA enrollment rates were quantitatively tested and confirmed.

"We are excited to take machine learning to a bigger scale -- not just in deploying this model to a wider area but also exploring other targeted communications that will be relevant to our customers."


Marketing Head

The Tools

Embracing complexity to
answer unknown questions

Thinking Machines helped the client figure out how to leverage its data to carry out advanced machine learning work:

  • Big data clustering, mainly HDBSCAN and variants
  • Multivariate regression analysis using random forest techniques
  • A/B testing to validate the effectiveness of our targeted recommendations

The Impact

Significant Improvement
in Signups and Payment Timeliness

  • 47%
    increase in
    signups vs. control
  • 30%
    improvement in
    payment timeliness
    vs. control

Our client was able to identify customer segments that revealed the greatest potential for intervention, as well as segment features that would drive down DSO and increase APA enrollment. This enabled them to create ad content which improved payment timeliness and APA enrollment rates with statistically significant results.

Can we help you solve a similar problem?

Get a free consultation