Analytics P6: Campaign Effectiveness Analysis

Client

Wei Liang, Director of Marketing at BrightSmile Dental. Six dental clinics in Chengdu, China.

What you are building

A statistical analysis of whether a CNY 180,000 marketing campaign (WeChat ads, KOL partnerships, referral bonuses) actually drove new patient bookings or whether the observed 22% increase was seasonal. The deliverable includes hypothesis tests with p-values, confidence intervals, channel-level attribution, and board-ready findings.

Tech stack

Python 3.11+ (Miniconda, "analytics" environment)
DuckDB (data loading and profiling)
pandas (data manipulation)
scipy.stats (hypothesis testing)
statsmodels (proportions_ztest)
matplotlib / seaborn (statistical visualizations with confidence intervals)
Metabase (dashboard with statistical results)
Jupyter Notebook
Git / GitHub

File structure

~/dev/analytics/p6/
  materials/
    CLAUDE.md              <- This file
    booking-data.csv       <- Two years of daily booking data (~28,000 rows)
    campaign-calendar.md   <- Campaign timeline and cost breakdown
    data-dictionary.md     <- Column definitions for booking data
    statistical-testing-guide.md <- Hypothesis testing concepts and test selection
  analysis/                <- Your Jupyter notebooks and analysis files

Key materials

booking-data.csv -- 28,000 rows, 6 clinics, 2 years of daily bookings. Columns: date, clinic_id, patient_type, service_category, booking_source, revenue
campaign-calendar.md -- WeChat ads (Oct 1-Dec 31), KOL (Oct 15-Dec 15), Referral (Oct 1-Dec 31). Total: CNY 180,000
data-dictionary.md -- Column definitions, business terminology, data notes
statistical-testing-guide.md -- Hypothesis testing intro, test selection, confidence intervals

Work breakdown

Profile the booking data and identify data quality issues (Gaoxin clinic, attribution gap)
Frame testable hypotheses and select statistical tests (z-test for proportions)
Run the tests with scipy.stats/statsmodels, compute confidence intervals
Build uncertainty-aware visualizations and write findings in uncertainty language
Deliver results to Wei and close the project

Verification targets

Primary hypothesis test: p-value for campaign effect on new patient bookings (z-test for proportions)
Confidence interval for the campaign effect (95% CI)
Channel-level breakdown with per-channel test results
Service category analysis (chi-squared test)
Cost per incremental patient expressed as a range, not a point estimate

Commit convention

Commit after completing each unit's work. Descriptive commit messages: "Profile booking data and identify Gaoxin clinic issue", "Frame hypotheses and select z-test for proportions", etc.