Fashion e-commerce analytics that audits the data before trusting it. 31,047 order lines of Dhaaga & Co., a fictional Pakistani fashion label, rebuilt end to end: data forensics, a returns desk, a size-by-size buy plan, plain-language answers to any question, and a 10-page Power BI report.
Live: https://threadline.twinstackstudio.com
- The story — what the data really says, chapter by chapter, from the fake columns to where the money is lost.
- Data forensics — every one of the 18 columns put on trial: reliable, repaired, aggregate only or made up, with the evidence for each verdict.
- Returns desk — 2,406 returned, refunded and cancelled lines. Record the reason, mark the parcel received, grade it A/B/C back to stock or write-off, refund with a reference; every step has a history that can be undone. CSV export and print.
- Buy plan — how many of each style and size to reorder: net sell rate after returns, times weeks of cover plus supplier lead time, plus safety stock. Sliders, size split, CSV and print.
- Ask — type a business question in English, Urdu or Roman Urdu ("top 10 divisions by net sales", "punjab vs sindh"); Threadline looks it up across all orders and shows the answer, a chart, a short explanation in plain words and the exact query it ran.
- Order inspector — open any order, see every line and why it is risky.
- Products, basket, geography, explorer — the styles that make 80% of sales, items bought together, a Pakistan map by province and division, and a free explorer.
- Reports — an Excel workbook (12 sheets), a printable one-page brief, the Power BI files and the SQL behind every number.
- The layout and type scale from a 320 px phone to a 2560 px screen.
| Buy plan | Data forensics |
|---|---|
 |
 |
| Ask the data | Geography |
|---|---|
 |
 |
| Power BI report | |
|---|---|
 |
 |
 |
 |
- 4 of 18 columns were made up. Channel, age, gender and customer ID were randomised by whoever built the file: 79% of multi-item orders appear on 2+ sales channels at once (one order ID, one date, up to 12 ages). A score built only on those columns does no better than a coin flip.
- The "seasonal trend" was a date bug. Day and month were swapped; the real window is 1–12 April, May and June 2022 (36 days).
- 322 of 1,164 styles (28%) make 80% of net sales. 3-Piece Suits are half the business with the lowest problem rate.
- Rs 36.8 L lost to cancellations and returns (7.75% of lines). Items under Rs 1,000 fail 16.5% of the time vs ~6% above.
- Return risk: the riskiest 10% of lines fail 3.4× more often than average.
- Basket: multi-item orders are worth Rs 3,620 vs Rs 1,730; Shalwar + Dupatta are bought together 8× more often than chance.
- Geography: 7 provinces, 35 divisions; Punjab is 56% of net sales; Karachi, Lahore and Islamabad lead the cities.
flowchart LR
RAW[Raw order workbook<br/>31,047 lines] --> LOC[Pakistan localisation]
LOC --> ETL[Cleaning<br/>17 quality checks · date repair]
ETL --> DB[(Reporting database<br/>star schema)]
DB --> SQL[16 business questions]
DB --> AN[Statistics · basket rules<br/>return risk score]
SQL --> DD[Dashboard data]
AN --> DD
DD --> UI[Web dashboard<br/>story · returns desk · buy plan · ask]
DB --> PBI[Power BI report<br/>10 pages]
Source: a public e-commerce sample workbook, localised to Pakistan (divisions, towns, postal codes, sales channels, product names, PKR) before the audit runs; the mapping keeps the data's structure, so every finding still holds. The store and its customers are fictional. Returns desk notes are kept in the viewer's own browser. Pakistan map: geoBoundaries PAK ADM1 (CC BY 4.0).
This is the public showcase. The source code is kept in a private repository. Try the live demo above, and contact TwinStack Studio for a walkthrough or a similar project.
© 2026 TwinStack Studio. All rights reserved. See LICENSE.