Digital Propensity: Predicting

Customer Intent Smarter

Digital Propensity: Predicting

Customer Intent Smarter

Digital Propensity: Predicting

Customer Intent Smarter

Leveraging machine learning to transform customer targeting, boost personalization, and drive retail growth through real-time insights.

Leveraging machine learning to transform customer targeting, boost personalization, and drive retail growth through real-time insights.

Time

2018-2020

Role

UX/UI Designer

Key Skills

Agile

Research

UX

UI

Introduction

Understanding

Customer

Intent

Understanding customer intent

Understanding customer intent

A machine learning-based framework that tracks demographic traits and behaviour to predict purchases and personalize engagement. Helps deliver timely, relevant offers that boost conversions.

A machine learning-based framework that tracks demographic traits and behaviour to predict purchases and personalize engagement. Helps deliver timely, relevant offers that boost conversions.

A machine learning-based framework that tracks demographic traits and behaviour to predict purchases and personalize engagement. Helps deliver timely, relevant offers that boost conversions.

Challenges

Solving Scalability and Accuracy Gaps

Solving Scalability and Accuracy Gaps

The legacy machine learning pipeline was slow and relied on sampling just 10% of data, resulting in poor accuracy and delays in business processes.

The legacy machine learning pipeline was slow and relied on sampling just 10% of data, resulting in poor accuracy and delays in business processes.

The legacy machine learning pipeline was slow and relied on sampling just 10% of data, resulting in poor accuracy and delays in business processes.

Highlights from Data Discovery and Analysis

Highlights from Data Discovery and Analysis

Limited accuracy due to data sampling

“Sampling only 10% of our data was hurting model performance and our ability to personalize.”

“Sampling only 10% of our data was hurting model performance and our ability to personalize.”

Excessive processing time

“Running a single ML pipeline took over 18 hours—it was holding up everything downstream.”

“Running a single ML pipeline took over 18 hours—it was holding up everything downstream.”

Too complex to manage

“With over 3600 features, tuning the models was both time-consuming and inconsistent.”

“With over 3600 features, tuning the models was both time-consuming and inconsistent.”

Business Challenges & Competitive Gaps

Competitor research

The market was moving toward real-time personalization and omni-channel integration—areas where the client lagged.

The market was moving toward real-time personalization and omni-channel integration—areas where the client lagged.

UX audit findings

Data silos and lack of a unified pipeline were preventing meaningful insights and collaboration across teams.

Data silos and lack of a unified pipeline were preventing meaningful insights and collaboration across teams.

Business pain points

Low model accuracy, slow delivery cycles, and missed opportunities in customer engagement were impacting growth and revenue.

Low model accuracy, slow delivery cycles, and missed opportunities in customer engagement were impacting growth and revenue.

Strategy

Customer Journey Map

To identify the main problems and opportunities, we conducted workshops, examined user data, and mapped the customer journey. This made the team more cohesive and helped establish clear goals for the redesign.

To identify the main problems and opportunities, we conducted workshops, examined user data, and mapped the customer journey. This made the team more cohesive and helped establish clear goals for the redesign.

Data Scientist

Data Scientist

Phases

Phases

🌟 Discover

📊 Evaluate

🛠️ Onboard

🤝 Use & Collaborate

🎯 Optimize

Actions/Needs

Actions/Needs

Needs scalable, fast data modelling

Assesses tools that support full datasets, no samples

Wants drag-and-drop UI for feature engineering

Builds and refines models faster

Reviews model metrics, retrains as needed

Touchpoints

Touchpoints

ML pipeline templates; demo models

Full dataset support; 100% ingestion

Visual ML designer

Auto-versioning, tuning dashboard

Validation stages; performance tracking

Data Engineer

Data Engineer

Phases

Phases

🌟 Discover

📊 Evaluate

🛠️ Onboard

🤝 Use & Collaborate

🎯 Optimize

Actions/Needs

Actions/Needs

Seeks a unified ETL + ML platform

Evaluates scheduling and integration capabilities

Sets up connectors, configures governance tools

Maintains pipeline health + deploys at scale

Automates updates, tests scalability

Touchpoints

Touchpoints

Batch + streaming pipeline showcase

Workflow scheduling, I/O connectors

Built-in monitoring + error handling

Metrics dashboard; alert system

Pipeline versioning + hybrid deployment

Business Analyst

Business Analyst

Phases

Phases

🌟 Discover

📊 Evaluate

🛠️ Onboard

🤝 Use & Collaborate

🎯 Optimize

Actions/Needs

Actions/Needs

Insight into customer behaviour + ROI metrics

Looks for visualization and metric tracking support

Joins onboarding sessions to understand outputs

Monitors campaigns, segments + O/P performance

Requests feature iterations or campaign pivots

Touchpoints

Touchpoints

Prebuilt dashboards + segment-level

KPI dashboards; explainable output

Simplified visual reports

Real-time Tracking + Integrate with tools

Feedback loop enabled via collaboration

Design

Design, Prototyping & Validating

We built and deployed the Digital Propensity pipeline using XStreams, tested multiple ML models, collaborated with stakeholders, and refined features based on real-time performance insights.

We built and deployed the Digital Propensity pipeline using XStreams, tested multiple ML models, collaborated with stakeholders, and refined features based on real-time performance insights.

Impact

Results that Delivered Business Value

Training time reduced from 18+ hours to

Training time reduced from 18+ hours to

Training time reduced from 18+ hours to

3 hours

3 hours

3 hours

Model accuracy improved by

Model accuracy improved by

Model accuracy improved by

14%

14%

14%

Estimated revenue uplift

Estimated revenue uplift

Estimated revenue uplift

~$20M

~$20M

~$20M

Lessons

What I Learned?

Visual simplicity supports complex systems

A clear, intuitive UI (like drag-and-drop pipelines) makes advanced data processes more accessible to non-technical users.

Feedback loops drive better outcomes

Continuous iteration based on real-time user and performance feedback helps improve both design and functionality.

Designing for scalability matters

Interfaces and interactions must handle large data volumes without overwhelming the user.

Create a free website with Framer, the website builder loved by startups, designers and agencies.