Delivery Reliability Analysis

A public-data supply-chain and business-analysis case study by Lesley Maore.

Read the case study · Read the executed notebook

The business question

Where are orders losing time, and what should the business investigate to make delivery more reliable?

Historical Olist records connect delivery promises, outcomes and customer reviews. The case translates descriptive findings into proposed operational requirements.

What the evidence says

• 6,534 / 96,470 eligible delivered orders arrived late: 6.8%.
• 62.4% of reviewed late orders received 1–2 stars, compared with 9.3% of reviewed on-time orders. Association is not causation.
• On a common valid-stage subset, late orders averaged 27.9 days after carrier handover, versus 8.0 days for on-time orders. The interval does not establish carrier fault.
• Rio de Janeiro has a 12.1% late rate; São Paulo has 4.5% but more late orders in absolute terms. Both rate and volume matter when choosing an investigation.

What I built

• A plain-language portfolio case study with a process map and interactive destination comparison.
• A reproducible Python analysis and an independent SQL reconciliation of the main cohort.
• An executed learning notebook with charts, exclusions and a sensitivity check.
• A business brief with proposed stakeholder roles, an overdue-order queue concept, requirements and acceptance criteria.
• A methodology record with definitions, provenance and limitations.

Reproduce

Download the original Olist dataset and extract its CSV files locally. Do not commit raw_data/ to this public repository.

From this project directory, with Python 3.10 or newer:

python3 analyze.py /path/to/extracted/csv --output data/recomputed.json
python3 verify_analysis.py /path/to/extracted/csv

The analysis uses Python's standard library and checks keys, joins and SQL agreement. data/analysis.json retains aggregates and input SHA-256 fingerprints.

For the notebook, install jupyter, nbclient, nbformat and matplotlib in a virtual environment. Put the CSVs in raw_data/, or set OLIST_DATA_DIR to their directory. Run analysis.ipynb from top to bottom. It checks the complete recomputation against the published snapshot before plotting. Saved outputs are included, and notebook.html is a readable export.

Scope and honesty

Historical orders from 2016–2018, not current operational performance. The main rate is delivered-only, so incomplete and non-delivered orders need separate monitoring. No inventory, stockout, causal, carrier-fault or realised-savings claims are made. Recommendations are proposals, not a deployed solution.

Attribution and licence

Source: Olist, Brazilian E-Commerce Public Dataset by Olist, distributed through Kaggle under CC BY-NC-SA 4.0. Retrieved 20 September 2026. This analytical adaptation and its derived aggregates retain that licence. Changes include cleaning rules, joins, aggregation, charts and interpretation. No endorsement is implied. Raw identifiers, review text and precise coordinates are not republished.