🇷🇺 Русская версия | 🇬🇧 English version
Hackathon: Dataton 2.0 (Central University)
Team: aCUtone!
Timeline: July 2024
Result: Data-driven marketing strategy to reduce churn and attract family segment
JFood, a food delivery startup by entrepreneur Jacob, experienced a sharp decline in sales after just 2 months of operations. Despite running 5 marketing campaigns, customer retention dropped significantly in late June - early July 2024. The task was to:
- Identify the root cause of customer churn
- Analyze 3 months of data (3,000 customers)
- Develop an actionable strategy within $600K budget
- Prepare recommendations for investor pitch
Our team conducted comprehensive data analysis and discovered that the marketing strategy failed to target the right audience — specifically, educated adults aged 18-50 with families and children.
Customer churn pattern:
- Peak churn occurred in weeks 26-27 (late June / early July 2024)
- Primary churn segment: adults 18-50 years old with higher education
- Both customers with and without children left the platform
Marketing campaign effectiveness:
- Campaigns resonated with single/widowed users
- Campaigns completely missed married customers with families
- People without children responded 3-4x better to ads than those with children
- Campaign #2 had the lowest success rate
Data quality issues identified:
- 8% outliers removed from order data (negative average checks)
- 11% outliers in online catalog interactions
- 3% outliers in mobile app data
The marketing campaigns were designed for single individuals and failed to address the needs of family customers — the core demographic for a food delivery service. This misalignment caused educated, financially stable customers (prime target audience) to abandon the platform.
- Removed negative values and outliers (8-11% of data)
- Merged three datasets: user data, order data, interaction data
- Created age-based cohorts and segmentation
- User complaints timeline: identified complaint spike in June-July
- Churn analysis: tracked last order dates to identify churned users
- Demographic segmentation: analyzed by age, education, family status, children
- Marketing effectiveness: evaluated 5 campaigns across segments
- Purchasing patterns by age and income
- Marketing campaign success rates by demographic
- Churn distribution across customer segments
Overall Accuracy | 83% marketing mismatch identified
Posts analyzed | 3,000+ customer records
Outliers cleaned | 8-11% of raw data
Churn peak | Weeks 26-27 (late June 2024)
Primary insight: Marketing campaigns targeted singles while product serves families → 60%+ churn in target demographic
Recommended budget reallocation:
- Social media: 900K RUB (family-oriented content)
- Contextual ads: 700K RUB (parent-focused keywords)
- Outdoor advertising: 2.3M RUB (family lifestyle imagery)
- Targeting specialist: 85K RUB
- Exploratory data analysis: cleaned dataset (removed 8-11% outliers), analyzed user complaints timeline
- Customer segmentation: grouped users by age, education, family status, and children presence
- Churn analysis: identified churn peak (weeks 26-27), visualized customer inflow/outflow patterns
- Marketing campaign effectiveness: evaluated 5 campaigns across demographic segments, discovered family segment gap
- Strategic recommendations: co-developed 4-pillar recovery strategy with budget allocation and risk assessment
- Visualization: created age-income distributions, campaign effectiveness charts, churn timelines
- Python: pandas, numpy, matplotlib, seaborn
- Analysis: Jupyter Notebook
- Visualization: matplotlib (scatter plots, histograms, time series)
- Presentation: MS PowerPoint
- Data cleaning: outlier detection, negative value removal
- Segmentation: demographic, behavioral analysis
- Cohort analysis: time-based churn patterns
- Statistical analysis: correlation matrices, distribution analysis
- User demographics (age, gender, education, family status)
- Registration dates
- Children count
- Order history (timestamps, amounts, categories)
- Average check values
- Purchase frequency
- Marketing campaign interactions
- Website/app engagement
- Campaign outcomes (success/failure)
cleaned_userdata.csv — User data after outlier removal and validation
cleaned_orderdata.csv — Order data after cleaning negative values
cleaned_interactiondata.csv — Interaction data after validation
age_interactiondata.csv — Segmented analysis combining user demographics with interaction patterns
- Full analysis notebook — complete EDA, segmentation, and insights
- Final presentation — strategy and recommendations
- 3C Analysis (Company, Customers, Competition)
- SWOT
- Customer segmentation
- Cohort analysis (time-based churn)
- Demographic segmentation
- Marketing attribution analysis
- Outlier detection and data cleaning
- Scatter plots (age vs income, payment ability)
- Histograms (order distributions by demographics)
- Time series (churn patterns, complaint trends)
Dataton 2.0 is a case-based data analytics competition organized by Central University and T-Education platform. Teams received real business datasets (3,000 customer records) and had one day to:
- Conduct exploratory data analysis
- Identify business problems through data
- Develop data-driven solutions
- Present findings in technical presentation + video pitch