Marketing Campaign ROI & A/B Test Dashboard

Turn raw marketing data into budget decisions — backed by statistics, not guesswork.

---

Live Demo

🚀 View Dashboard →  ·  📂 GitHub Repository

Note: All currency values are converted from INR to USD using an approximate exchange rate of 1 INR = 0.012 USD for presentation purposes.

---

Power BI Dashboard

In addition to the live web dashboard, this project includes a Power BI version — Marketing-Decision-Engine — built on the same dataset to demonstrate BI tool proficiency.

Download the .pbix file — open in Power BI Desktop (free) to explore interactively with filters and slicers.

Note: All currency values are converted from INR to USD using an approximate exchange rate of 1 INR = 0.012 USD for presentation purposes.

---

Problem Statement

Marketing teams routinely split budgets evenly across channels with no data to justify the allocation. Without statistical rigor, "Campaign B performed better" is anecdote — not evidence.

---

Business Impact

Reallocating budget to Email — the top-performing channel at 354.35% ROI — from Google Ads (−10.91% ROI) is projected to improve blended ROI by 255.5 percentage points, backed by statistical A/B testing — not guesswork.

Metric	Value
Total Spend	₹1.74 Cr
Total Revenue	₹2.37 Cr
Best Channel	Email — 354.35% ROI
Worst Channel	Google Ads — −10.91% ROI
ROI Uplift (best vs avg)	+255.5 pp

---

Strategic Recommendation

Reallocating 20% of budget from the lowest-ROI channel (Google Ads, −10.91% ROI) to Email (354.35% ROI) is projected to improve blended ROI by 255.5 percentage points.

---

Key Insight

Campaign B significantly outperforms with a 131.9% conversion uplift (p < 0.0001 — statistically significant at 95% confidence).

---

A/B Test Results

Two-proportion z-test on campaign response flags (n = 2,240 customers):

Variant	Conversion Rate	Winner
Campaign A — Existing	6.43%	—
Campaign B — New	14.91%	★ Yes

• Uplift: +131.9%
• p-value: < 0.0001
• 95% CI on difference: [+0.067, +0.103]
• Verdict: Significant at 95% confidence — roll out Campaign B to 100% of traffic.

---

Tech Stack

Layer	Technology
Data pipeline	Python · pandas · SciPy (z-test) · requests
Frontend	React 18 · Vite 5 · Tailwind CSS v3 · Recharts
Deployment	Vercel (static)

---

How It Works

1. Channel ROI Analysis — Python script aggregates purchases per channel, computes spend vs revenue, ranks by ROI.
2. Statistical A/B Test — Two-proportion z-test compares Campaign A vs B conversion rates; outputs p-value, 95% CI, and significance verdict.
3. Budget Recommendation — Identifies worst and best ROI channels; quantifies projected uplift from reallocation.
4. Static Dashboard — Results bundled into results.json at build time; React dashboard renders charts, A/B card, and recommendation banner with zero backend.

---

Features

• ROI Bar Chart — Per-channel comparison with color-coded bars and tooltips
• A/B Significance Card — Variant conversion rates, p-value, 95% CI, plain-English significance verdict
• Recommendation Banner — Data-driven budget reallocation call-to-action
• Multi-Currency Toggle — Live conversion: INR ₹ · USD $ · GBP £ · EUR € · AED د.إ
• Mobile-Responsive — Fluid layout, no horizontal scroll, charts resize to viewport

---

Architecture

generate_analysis.py  →  results.json  →  Vite build  →  Static HTML/JS  →  Vercel CDN

Analysis is pre-computed once and served statically — instant load, zero cold-start, zero hosting cost. The component contract is a plain JSON import; swapping it for a fetch() call adds a live FastAPI backend whenever real-time data is needed. Deliberate design choice: ship fast, scale later.

---

Setup

# 1. Install Python dependencies
pip install -r requirements.txt

# 2. Generate analysis (downloads dataset automatically, falls back to synthetic if offline)
python generate_analysis.py

# 3. Start the dev server
cd frontend
npm install
npm run dev

Dev server binds to 0.0.0.0 and reads PORT from the environment — Codespaces port-forwarding works out of the box.

Deploy to Vercel: import the repo at vercel.com/new, set Root Directory to frontend, done.

---

Methodology

Statistical significance is determined using a two-proportion z-test — the standard choice when comparing binary conversion rates between two independent groups of equal size. The test outputs a z-statistic and two-sided p-value; significance threshold is α = 0.05 with a 95% confidence interval on the conversion rate difference.