Data

Tools

Indexes

Applications

Experiments

Awesome

An open API service indexing awesome lists of open source software.

https://github.com/samyomb/olist-ecommerce-analytics

Olist e-commerce performance & customer reviews — Python cleaning + BigQuery SQL + Looker Studio dashboard (2017 FY & 2018 YTD) with actionable insights
https://github.com/samyomb/olist-ecommerce-analytics

analytics bigquery brasil customer-experience dashboard data-visualization e-commerce looker-studio olist python review sql

Last synced: 8 months ago
JSON representation

Olist e-commerce performance & customer reviews — Python cleaning + BigQuery SQL + Looker Studio dashboard (2017 FY & 2018 YTD) with actionable insights

Awesome Lists containing this project

README 

          
# Olist E-commerce analytics

Olist e-commerce performance & customer reviews — Python cleaning + BigQuery SQL + Looker Studio dashboard (2017 FY & 2018 YTD) with actionable insights



Live dashboard:

https://lookerstudio.google.com/reporting/9bf77bf9-0da9-4c79-8745-28d9893c480a



## Results



2017 (full year): GMV R$ 7.02M, 45.1K orders, AOV R$ 155.7, Canceled 0.59%, Avg delivery 13 days, Late rate 6.4%, 44,573 reviews (14.7% negative).



2018 YTD (Jan–Aug vs Jan–Aug 2017): GMV R$ 8.55M (+140.9%), Orders 54.0K (+135.1%), AOV R$ 158.4 (+2.5%), Canceled 0.58% (−0.16 pp), Late rate 9.2% (↑ vs 4.0% PY).



Top negative reason: Delivery delay (~49–52% of negatives).



Highest severity reason: Wrong / Not as described (~75–80% rated 1–2/5).



Impact plan (next 30–60 days): cut late rate −1 pp in hotspot states (AL/MA/CE), reduce negative reviews −1 pp (delivery-delay share −5 pp), grow AOV +1–2% on high-ticket categories, keep canceled ≤ 0.6%.



## Why I built this



Understand Olist’s sales performance, category mix, and customer pain points, then turn findings into an actionable plan that Ops, CX and Growth can execute.



## Stack



Python - cleaning (Drop isna, duplicated, columns)



BigQuery – SQL table + view final_data_enriched



Looker Studio – interactive 3-page report (report-level filters)



SQL – CTEs, date logic, KPI calculations



## Data model 



One row per order with:



gmv = price + freight



product_category & customer_state (cleaned)



order_purchase_timestamp, order_delivered_customer_date, order_estimated_delivery_date



delivery_days = date diff (delivered − purchase)



is_late = delivered > estimated



Reviews analyzed by score bands (1–2 / 3 / 4–5) and reason (grouped)



## KPI definitions (BigQuery-style)



### -- GMV

SUM(oi.price + oi.freight_value) AS gmv



### -- Orders

COUNT(DISTINCT o.order_id) AS orders_total



### -- AOV

SAFE_DIVIDE(SUM(oi.price + oi.freight_value), COUNT(DISTINCT o.order_id)) AS aov



### -- Canceled %

SAFE_DIVIDE(COUNTIF(o.order_status = 'canceled'),

            COUNT(DISTINCT o.order_id)) AS canceled_rate



### -- Late rate

SAFE_DIVIDE(COUNTIF(order_delivered_customer_date > order_estimated_delivery_date),

            COUNT(DISTINCT o.order_id)) AS late_rate



### -- Avg delivery time (days)

AVG(DATE_DIFF(order_delivered_customer_date, order_purchase_timestamp, DAY)) AS avg_delivery_days



### -- Negative reviews %

SAFE_DIVIDE(COUNTIF(review_score <= 2), COUNT(*)) AS negative_reviews_pct



## What’s inside the report



### Page 1 – Sales Performance

KPI cards (GMV, Orders, AOV, Canceled %, Avg delivery days, Late rate), GMV & Orders trend (combo), Top-10 states, Late rate by state.



![Page 1 - Sales Dashboard](assets/screenshots/sales_dashboard_2017.png)(https://lookerstudio.google.com/u/0/reporting/9bf77bf9-0da9-4c79-8745-28d9893c480a/page/iFgZF)



### Page 2 – Category Performance

KPI cards, Growth vs PY by category (GMV share & Δ%), Category × State heatmap, Top-5 GMV by category, Top-5 AOV by category.



![Page 2 - Category Heatmap](assets/screenshots/category_2017.png)(https://lookerstudio.google.com/u/0/reporting/9bf77bf9-0da9-4c79-8745-28d9893c480a/page/p_zv98sabpwd)



### Page 3 – Customer Reviews & Reasons

100% stacked review trend (High/Medium/Low), Severity × Prevalence bubble chart, table with Share of all reviews, % negative within reason, % of total negatives.



![Page 3 - Reviews](assets/screenshots/reviews_2017.png)(https://lookerstudio.google.com/u/0/reporting/9bf77bf9-0da9-4c79-8745-28d9893c480a/page/p_iy8wkjepwd)



Filters (report-level, persistent across pages): Date · State · Category.

Button Clear all resets the whole report.



## Key insights (data-driven)



2018 is scaling fast (like-for-like): GMV +140.9%, Orders +135.1%, AOV +2.5%.



Risk: on-time delivery – Late rate up to 9.2% (from 4.0% PY Jan–Aug); hotspots: Alagoas → Maranhão → Ceará.



Customer voice – Delivery delay drives about half of all negative reviews; Wrong/Not as described has very high severity.



Growth levers – High-ticket Computers (AOV ~R$1.2–1.3K) and selected Appliances; GMV leaders Bedding & Bath, Health & Beauty, Sports & Leisure.



Regional play – SP/RJ/MG support premium AOV; target local bundles & financing.



## Actions & expected impact



Ops – Reduce late deliveries (AL/MA/CE): tighten carrier SLAs, add fallback carrier, daily on-time dashboard & alerts, clearer cut-offs, first-mile controls, proactive ETA comms (auto SMS/e-mail when ETA > SLA with new ETA + goodwill).

CX – Fix “Not as described”: double-scan at packout + photo proof; clarify PDP titles/variants/photos.

Growth – Monetize high-AOV & protect volume: bundles + financing for Computers/Appliances (geo-target SP/RJ/MG); for Bedding & Bath / Health & Beauty ensure stock health, price monitoring, cross-sell.



## Success metrics (60 days):



Late rate −1 pp (AL/MA/CE)



Negative reviews −1 pp (delivery-delay share −5 pp)



AOV +1–2% on high-ticket categories



Canceled ≤ 0.6%



## How to run



BigQuery



Build a view/table final_data_enriched with fields above and the KPI logic.



Ensure date fields are DATE/TIMESTAMP and categories/states are cleaned.



Looker Studio



Connect the table & view as a data source.



Use report-level controls (Date, State, Category).



For YoY on charts, set Comparison date range = Previous year.



## Limitations



2016 partial (Sep–Dec). 2018 data available Jan–Aug only; YoY comparisons use Jan–Aug vs Jan–Aug.



Review reasons are grouped; linking every review to an order may vary by dataset completeness.



## Contact



Samy Bouhassoune – Data Analyst

LinkedIn: https://www.linkedin.com/in/samy-bouhassoune · Email: samyy.b@hotmail.fr



## 🇫🇷 Résumé



2017 : R$ 7.02M GMV, 45.1K commandes, AOV R$ 155.7, Annulés 0.59%, Livraison 13 j, Retards 6.4%, 44,573 avis (14.7% négatifs).



2018 YTD (jan–août vs N-1) : GMV +140.9%, Commandes +135.1%, AOV +2.5%, Annulés 0.58% (−0.16 pt), Retards 9.2%.



Priorités : baisser les retards (AL/MA/CE), corriger “non conforme”, pousser les catégories à fort AOV, protéger les best-sellers.