Read summarized version with
About project
Working time:
Since December 2025
Industry:
Consumer goods distribution
The service:
Data and AI, Generative AI
Overview
Pro Grup is the official importer and distributor in Poland of five foreign beauty and FMCG brands, with more than 140 active SKUs sold to Polish retail and drugstore chains, on Amazon in the EU and the UK, through wholesale and to hair salons. Every unit it sells is bought from producers in Italy, Spain, Ukraine and Turkey, held in its own stock and delivered against the chains’ orders. Pro Grup partnered with Dedicatted to replace a monthly spreadsheet forecast with a demand and supply planning platform on AWS that tells its purchasing team what to order, from whom and when.
Challenges
Pro Grup sits between foreign producers and Polish retail chains, so its business is a balancing act between supply and demand.
Long, rigid supply
- Producers ship with lead times of 4 to 10 weeks and only in minimum production batches, so every forecast error is locked in for a full replenishment cycle.
Demanding customers
- Retail chains order weekly through EDI and require a fill rate of 95 to 98 per cent of ordered quantities, with penalties for under-delivery and shelf listings at risk.
A forecast without signals
- The forecast and purchasing plan were produced once a month in Excel from sales history alone, with no seasonality and no view of chain promotions.
- In the six months before the new platform, about 11 per cent of SKUs were out of stock in any given month, the fill rate on chain orders was about 91 per cent, and stock on hand covered about 118 days of sales, tying up working capital in slow lines while fast movers ran out.
Want to get your copy of case study?
Download it here.
By submitting this form, you agree with our Terms & Conditions and Privacy Policy.
Our Approach
Dedicatted built a planning platform that a 25-person company with no IT team can run: every component is serverless or managed, and the ERP stays the system of record.
- One source of planning data. Sales, stock and open purchase orders from Comarch ERP Optima, chain orders and sell-out from Comarch EDI, and Amazon marketplace sales arrive every night in a data lake on Amazon S3, where retailer EAN codes are mapped to Pro Grup’s own SKUs.
- A weekly statistical forecast. Every week an ensemble of Chronos-Bolt and statistical models is retrained on Amazon SageMaker and produces a 26-week forecast for every SKU, channel and week, including promotion periods. Only a model version approved in the model registry drives purchasing.
- A supply plan, not just a forecast. The forecast is balanced against each supplier’s lead time and minimum order quantity, a safety stock for a 97 per cent service level, stock on hand and open orders, producing a proposed purchase order per supplier.
- An assistant for the purchasing team. A purchasing agent built on Amazon Bedrock AgentCore explains why a forecast changed, runs what-if scenarios and drafts orders, using the product catalogue, trade terms and chain promotion calendars as context.
- Orders back into the ERP. The purchaser approves orders in a web app, and approved orders return to Comarch ERP Optima as draft purchase orders.
AWS Services Deployed
- Data integration: AWS Transfer Family for SFTP exchange with the ERP and EDI provider, AWS Lambda for the Amazon Selling Partner API, Amazon S3, AWS Glue and Amazon Athena
- Forecasting: Amazon SageMaker Pipelines with AutoGluon-TimeSeries, SageMaker Model Registry, AWS Step Functions and Amazon EventBridge Scheduler
- Supply plan and app: AWS Lambda, Amazon DynamoDB, Amazon API Gateway, Amazon CloudFront and Amazon Cognito
- Generative AI: Amazon Bedrock AgentCore Runtime, Gateway and Memory, Claude on Amazon Bedrock with Bedrock Guardrails, Amazon OpenSearch Service
- Security and operations: AWS KMS, AWS Secrets Manager, AWS IAM Identity Center, Amazon CloudWatch, AWS CloudTrail and AWS Backup
Business Milestone: Every Purchase Order Planned on the Platform
Since April 2026, every purchase order Pro Grup places with its producers has been based on the platform’s recommendations, approved in the planner app and imported into the ERP.
Outcome & Impact
- Forecast error fell from 42 to 22 per cent (weighted absolute percentage error per SKU, April to September 2026, against the previous method backtested on October 2025 to March 2026).
- Fill rate on retail-chain orders rose from 91 to 97 per cent, meeting the chains’ requirement.
- SKUs out of stock in a month fell from 11 to 4 per cent.
- Stock on hand fell from 118 to 84 days of sales, releasing working capital.
- The monthly purchasing cycle fell from about 4 days to half a day.
Summary
Pro Grup moved from a monthly spreadsheet to a weekly, statistically grounded plan that balances what its customers will order against what its producers can supply. The purchasing team now reviews proposed orders instead of building them, the retail chains receive what they order, and the business holds less stock while running out less often.
