This case study explores how TravelTide can improve customer retention through a more personalized rewards strategy.
The business challenge was to move away from generic offers and design a framework that assigns each eligible customer to the reward perk most aligned with their historical behavior.
The final outcome of the project is a rule-based behavioral targeting framework that maps users to one of five actionable perk categories, supported by exploratory machine learning analysis for comparison.
TravelTide wants to improve retention by offering more relevant rewards to existing customers.
Instead of relying on broad demographic personas, this project focuses on behavior-based segmentation and aims to answer one key question:
Which available perk is most relevant for each customer based on their past travel behavior?
The project followed this analytical pipeline:
- Join raw session, user, flight, and hotel data
- Filter the active customer cohort
- Validate and clean key booking-related fields
- Aggregate session behavior into user-level features
- Build perk-specific rule-based scoring logic
- Compare the rule-based framework against ML-based clustering approaches
- Generate final customer segments and business recommendations
The final segmentation framework was built around the five perks TravelTide can actually offer:
- Exclusive Discounts
- Free Checked Bag
- No Cancellation Fees
- Free Hotel Meal
- 1 Free Hotel Night with Flight
For each perk, a small set of relevant behavioral features was selected.
These features were converted into percentile ranks and combined into perk-fit scores.
Each user was then assigned to the perk with the highest score.
Unsupervised learning methods were also tested for comparison:
- K-Means
- DBSCAN
These methods were useful for exploratory pattern discovery, but they did not provide the same level of business alignment, interpretability, or stability as the rule-based framework.
- Hotel discounts were associated with stronger conversion than flight discounts
- Page-click activity was a strong signal of booking intent
- Cancellations appeared only in bundled flight + hotel bookings
- Converted sessions often included both flight and hotel activity
The clearest user-level differentiators were:
- Engagement
- Travel variety
- Discount responsiveness
Average trip distance was less useful as a segmentation driver.
The final rule-based framework assigns each eligible user to one of five mutually exclusive segments:
- Price-Sensitive Users → Exclusive Discounts
- Baggage-Heavy Travelers → Free Checked Bag
- Flexibility-Seeking Travelers → No Cancellation Fees
- Hotel-Oriented Users → Free Hotel Meal
- Bundled Travelers → 1 Free Hotel Night with Flight
Segment sizes were relatively balanced, which supports practical deployment in a marketing campaign.
Although machine learning helped explore the structure of the user base, the final recommendation favored the rule-based framework because it offered:
- stronger business alignment
- clearer interpretability
- more stable user-to-perk assignment
- direct readiness for campaign deployment
In this case, the most effective segmentation approach was not the most complex one, but the one most clearly connected to business action.
traveltide-segmentation-retention/
│
├── data/
│ ├── processed/
│ │ ├── sessions_filtered_cleaned.csv
│ │ └── user_agg_features.csv
│ └── outputs/
│ ├── user_perk_assignment.csv
│ ├── segment_summary.csv
│ └── segment_profile_summary.csv
│
├── docs/
│ ├── traveltide_executive_summary.pdf
│ ├── traveltide_detailed_report.pdf
│ └── traveltide_precision_targeting_presentation.pdf
│
├── images/
│ ├── workflow_overview.png
│ ├── user_level_signals.png
│ ├── final_segments_and_perks.png
│ ├── methodology_comparison.png
│ └── source_schema_overview.png
│
├── notebooks/
│ └── traveltide_segmentation_retention_final.ipynb
│
├── sql/
│ ├── 01_create_sessions_joined.sql
│ ├── 02_create_sessions_filtered.sql
│ └── 03_create_user_agg_features.sql
│
├── README.md
├── requirements.txt
└── .gitignore
- sessions_filtered_cleaned.csv — cleaned and filtered session-level dataset used for downstream aggregation
- user_agg_features.csv — aggregated user-level behavioral feature table
- user_perk_assignment.csv — final rule-based perk assignment for each eligible user
- segment_summary.csv — size and share of each final segment
- segment_profile_summary.csv — average feature profile of each segment
The project uses a mixed workflow:
- SQL / Spark SQL was used for data joining, cohort filtering, and user-level feature aggregation
- Python / pandas was used for data cleaning adjustments, rule-based scoring, segmentation logic, ML comparison, and final exports
The SQL files in this repository follow the Spark SQL / Databricks SQL dialect.
Initial data retrieval was performed in the Databricks course environment using the TravelTide case-study database.
Processed datasets and final project outputs are included in this repository for review.
- Python 3.12
- pandas
- NumPy
- Spark SQL / Databricks SQL
- Matplotlib
- scikit-learn
The recommended next step for TravelTide is to deploy the five rule-based segments as the primary targeting framework for a personalized rewards campaign.
Performance should be validated through A/B testing using:
- email open rate
- perk sign-up rate
- incremental ROI versus a non-segmented control group
The final project presentation is included in the docs/ folder as a PDF.
Portfolio project developed as part of a customer segmentation and retention case study focused on behavioral analytics, perk assignment, and business-oriented segmentation design.