Trend
Monthly revenue and units index
Each series is indexed to its own maximum so revenue and units can share one honest visual scale. The year filter narrows the period.
Verified analytics case study · 2022–2024
Smart Supply Chain Intelligence
Turning 190,740 FMCG transactions into actionable sales, inventory, promotion, and logistics intelligence.
The website uses recomputed CSV values where source PDFs contain inconsistent or ambiguous dashboard aggregations.
01
Project context
From raw transactions to business decisions.
Business problem
FMCG teams must balance demand, inventory availability, promotion effectiveness, delivery speed, and regional allocation. Raw tables rarely show where action is needed.
“Every visual should answer one decision question in under thirty seconds.”
Project objective
Build an auditable decision-support system for business, marketing, inventory, and supply-chain teams using verified metrics from the full dataset.
Business managersMarketing teamInventory managersSupply chain team
- Stack
- Python, Pandas, Power BI logic, HTML, CSS, Vanilla JS, SVG
- Output
- Static, deployable analytics portfolio with interactive filters and reports
Dataset overview
Real data, visible quality checks.
The complete 190,740-row CSV was analyzed in Python. The browser receives optimized aggregates and a representative sample, because loading 20 MB into every visitor’s browser would be an odd punishment.
100/100
Data quality status
No missing values, duplicate rows, or invalid negative values were detected. Outliers were documented rather than quietly deleted.
Dataset schema 14 core variables
| Column | Type | Description | Example |
|---|
Representative sample
Searchable transaction explorer
Executive dashboard
One system. Four stakeholder views.
Filters update the headline metrics and contribution charts from a compact 22,908-row analytical cube, not from invented front-end values.
Mix
Delivery status
Market
Brand contribution
Portfolio
Category revenue leaders
Geography
Regional contribution
Route to market
Channel performance
Deep analysis
Evidence before recommendations.
₹126.61 Cr
Revenue grew in both later years
Revenue increased 4.29% in 2023 and another 2.00% in 2024. Growth remained positive, although it slowed.
Top brandDabur · 10.24%Top categoryTea · 18.14%Best regionWest India · 20.10%
Yearly revenue growth
Revenue concentration
13.47 M
Weekend records carry higher average volume
Sunday averaged 73.44 units per transaction, compared with 69.02 on Monday. This pattern is measurable, not mystical retail folklore.
Daily calendar average12,522.79 unitsHighest weekdaySunday · 73.44Average per record70.64 units
Average units by day of week
25.73%
Portfolio-wide sell-through is modest
Milk shows 83.55% category sell-through, while slower categories create overstock candidates. The alert scores are normalized decision aids, not oracle tablets.
Reorder alerts shownTop 30Overstock alerts shownTop 30Highest category STRMilk · 83.55%
| Brand / SKU | Category | Region | Channel | Units | Fill rate | Sell-through | Risk score | Status |
|---|
WarehouseStock availability
Transport1–5 delivery days
ChannelRetail · Kirana · Online · Supermarket
DecisionFill rate and risk alerts
Average delivery by channel
Fill rate by region
Source reconciliation: the Stage 5 visual reports 297–304 as aggregated chart values. They are not literal delivery days. The raw CSV field ranges from 1 to 5, and the verified average is 3.00 days.
+28.69%
Promotion records show higher average unit sales
The association is consistent across brands and regions, with Dabur showing the strongest brand-level unit lift at 31.39%.
The analysis identifies associations in the available dataset and does not prove causal impact.
Unit lift+28.69%Revenue lift+26.85%Promotion share15.11%
Promotion lift by brand
Promotion lift by region
Statistical analysis
Business meaning, with technical evidence one click away.
Promoted records average about 19 more units than non-promoted records.
A small-to-moderate standardized difference, large enough to matter but not proof of causality.
Higher prices correlate with fewer units sold in this constructed dataset.
Price remains positively correlated with revenue despite lower volume.
Correlation matrix
Price versus units interpretation
Units sold
Price
Welch t-test
Promotion vs non-promotion units
- Null hypothesis
- Mean unit sales are equal.
- t statistic
- 47.026
- p-value
- < 0.001
- Cohen’s d
- 0.353
- Interpretation
- Statistically clear difference with small-to-moderate practical magnitude.
Mann–Whitney U
Distribution-robust comparison
- U statistic
- 2,854,891,262.5
- p-value
- < 0.001
- Assumption
- Independent records; similar distribution interpretation requires caution.
- Limitation
- Synthetic construction and non-random promotion assignment may inflate certainty.
Formula transparency
Verified KPI definitions
- Revenue
- price_unit × units_sold
- Fill rate
- Σ delivered_qty ÷ Σ stock_available
- Sell-through
- Σ units_sold ÷ Σ(stock_available + delivered_qty)
- Promotion lift
- (promo mean ÷ non-promo mean − 1) × 100
Deep insights
Numbers translated into action.
Methodology
An auditable path from CSV to decision.
Source reconciliation
Discrepancies are shown, not hidden.
The Stage 5 report, jury presentation, and complete CSV do not agree on every metric. The final website uses reproducible CSV calculations and preserves reported values as context.
| KPI | Stage 5 | Presentation | CSV recomputed | Final use |
|---|
Recommendations
Prioritized by stakeholder.
Protect category concentration
Tea contributes 18.14% of revenue. Track its availability separately, while building mid-tier categories to reduce dependence.
Evidence: category revenue share · Timeframe: quarterlyUse balanced regional allocation
Regional revenue shares sit tightly around 20%. Avoid dramatic regional shifts without more granular demand evidence.
Evidence: West 20.10%, Central 19.92% · Timeframe: monthlyScale promotions selectively
Promotion records show +28.69% average unit lift, but evaluate margin and causal tests before broad expansion.
Evidence: Welch test and effect size · Timeframe: next campaignPrioritize Dabur experiments
Dabur has the highest observed brand lift at 31.39%. Use controlled holdouts to verify whether the response survives real deployment.
Evidence: promotion by brand · Timeframe: 4–6 weeksSeparate fast and slow category policies
Milk sell-through is 83.55%, while Cheese is 6.05%. A single safety-stock rule would be administratively neat and operationally foolish.
Evidence: category sell-through · Timeframe: immediateReview critical alert combinations
Use reorder and overstock scores as triage. Validate each alert against current inventory snapshots before purchasing or clearance action.
Evidence: top 50 alert tables · Timeframe: weeklyMove fill rate toward 95%
The verified 90.82% fill rate remains below the report’s 95% operating target. Investigate replenishment and supplier reliability.
Evidence: delivered ÷ stock available · Timeframe: 90 daysFocus on critical delivery records
19.88% of records exceed four days. Channel averages hide this tail, so manage the critical share rather than celebrating a three-day mean.
Evidence: delivery status distribution · Timeframe: weeklyData ethics
Responsible analytics is part of the system.
The source presentation frames the work around privacy, transparency, fairness, and accountability. Here those pillars are translated into concrete project controls.
Privacy
No customer names, personal purchase histories, or individual profiling are present.
Risk: future data additions must preserve minimization.Transparency
KPI formulas, source conflicts, statistical tests, and limitations are visible.
Risk: stakeholders may still confuse proxies with operational truth.Fairness
Brands and regions are described through evidence, not blamed through labels.
Risk: synthetic balance can hide real-world inequality.Accountability
Analysis code, cleaning logs, reconciliation, and downloadable source tables support audit.
Risk: actions still require a named business owner.Project report
Source document, preserved with context.
21 pages · PDF
Stage 5 Dashboard Report
Dashboard design, stakeholder reporting, KPI color logic, findings, recommendations, risks, DAX, and schema appendices.
Open complete 33-file analysis library
Portfolio case study
Smart Supply Chain Intelligence
A complete analytical workflow spanning data validation, feature engineering, KPI reconciliation, statistical analysis, operational alerts, dashboard design, and responsible data storytelling.
Challenge
Turn a large transactional dataset and inconsistent report artifacts into a reliable, decision-ready portfolio.
Approach
Recompute every major KPI, document conflicts, export browser-safe aggregates, and design around stakeholder questions.
Result
A static, responsive website with interactive filters, charts, data explorer, audit files, and source reports.
Future work
Add real demand forecasts, controlled promotion experiments, current inventory snapshots, and automated data refresh.