Paid, owned, organic, promotion, and ecommerce sources.
Commercial Data Science Case Study
Estimating marketing impact and budget trade-offs for fashion ecommerce.
A UK fashion retailer needs a credible way to connect marketing spend, incrementality evidence, uncertainty, and commercial planning. This lab turns that measurement problem into a working analyst dashboard and reusable Python package.
Econometrics, adstock, saturation, priors, calibration, and uncertainty.
Profit-aware scenarios, constrained optimization, and executive summaries.
Business Context
The retailer wants to understand how paid and owned marketing channels contribute to revenue, new-customer growth, and profit. The team needs more than channel correlation: it needs a defensible workflow for estimating incremental contribution while accounting for seasonality, promotions, carryover, diminishing returns, and experiment evidence.
Measurement Question
How should a fashion ecommerce brand allocate weekly marketing budget across channels while accounting for data quality, uncertainty, experiment calibration, and practical business constraints?
Product Workflow
Validate Source Data
Check weekly marketing schema and upstream connector exports before modeling.
Assemble Weekly Dataset
Map ecommerce, GA4, paid media, CRM, affiliate, influencer, display, and control exports into an MMM-ready weekly table.
Diagnose Readiness
Review source coverage, history length, outcome quality, and channel availability.
Model Incremental Impact
Fit baseline econometrics, MMM-style response curves, Bayesian intervals, and experiment-informed calibration.
Plan Budget Scenarios
Compare gross-margin-aware budget options and constrained allocation recommendations.
Current Outputs
Analyst Dashboard
Revenue, orders, new customers, media spend, channel mix, promotions, and readiness checks.
Modeling Diagnostics
Holdout performance, contribution, ROI, response curves, uncertainty, and Bayesian summaries.
Evidence and Governance
Lift-test uploads, quality scoring, approved-only calibration, and experiment-informed priors.
Connector Workflow
Templates, validation, weekly assembly, source diagnostics, and downloadable assembled CSVs.
Commercial Planning
Profit-aware scenarios, channel constraints, allocation optimization, and planning caveats.
Stakeholder Summary
Automated executive summary language, model-run report, and JSON manifest for review.
Recommendation Governance
Readiness gates for model accuracy, evidence, profit impact, spend movement, and history length.
Technical Stack
Python
Pandas
Statsmodels
Plotly
Streamlit
Pytest
Ruff
uv
GitHub Actions
Data Notes
The included dataset is generated deterministically for portfolio use. It is not ASOS data and does not copy any private brand data. Uploaded files are parsed in memory in the current Streamlit version.