Trade / ProcurementML / Analytics
Task. A trading company chose suppliers based on buyers’ memory: evaluation criteria were not formalized, data on delivery failures and defects sat in scattered files, and comparing 5 data sources across legal entities and individuals was impossible.
Solution
We consolidated supplier data from 5 sources, formalized criteria (lead time, quality, price, documents) and built ML rating models. A buyer gets an objective supplier rating and a delivery-failure risk forecast before signing a contract.
Results
Technologies: Python, CatBoost, PostgreSQL, DataLens.
Project stages
- Criteria (1–2 weeks) — supplier-rating model design with procurement: price, lead time, quality, reliability.
- Data collection (2–4 weeks) — ERP, purchase history and external registries connection.
- Model build (4–6 weeks) — ML scoring, risk dashboards and supply-failure alerts.
- Process adoption — buyer training and tender-procedure integration.
Where this applies
Supplier analytics serves distributors in Kazakhstan and Uzbekistan, construction holdings in Bishkek and Almaty, and manufacturers in Tajikistan. See wholesale & distribution.