The client is a dynamically developing online sales platform operating on the Polish and British markets. The company struggled with an excess of seasonal stocks and the lack of availability of bestsellers, which resulted in the freezing of capital and the need to organize sales. Moreover, the company managed its inventory manually, which required a lot of time from the logistics department, and the management method itself was highly suboptimal and did not reflect the actual market situation.
Thanks to cooperation with Alterdata, it was possible to successfully implement advanced predictive models, automate warehouse processes and significantly increase operational efficiency.
Scope of work
Joint identification of business challenges
Together with the client, we discovered a number of key challenges that he faced before implementing the solution.
- Excess seasonal stock – capital was frozen in products that found no buyers, which led to costly sales. Moreover, the goods themselves took up a lot of warehouse space, which meant there was no room for products with high turnover and high demand.
- No bestsellers in stock – inaccurate forecasts meant that key products were often unavailable, leading to lost sales.
- Manual order management – manual decision-making processes did not allow for flexible response to market changes.
- Lack of precise demand analysis – previous planning methods did not take into account dynamic changes in trends and seasonality.
Data centralization
Before starting the implementation, the Alterdata team conducted an in-depth study data analysis historical client. The key challenge was to collect and standardize information from various sales channels: e-commerce, marketplaces and sales points.
Each channel generated different types of data, such as sales data, inventory data, returns and website traffic information. By integrating this data in BigQuery data warehouse, all information was collected in one place, which enabled better organization and easier access to key data.
Data analysis and implementation of predictive models
In the first step, together with the client's team, we defined key business indicators that influenced demand and inventory management. During the workshops, we also identified key indicators and standardized product identifiers, which allowed for effective forecasting.
Based on historical sales data and the analysis of demand patterns, taking into account seasonality and the impact of external factors such as promotions or price changes, we have developed a comprehensive analytical solution that enables optimal planning of inventory levels.
The solution explored various approaches to inventory modeling. Both time series analysis models (LSTM, ARIMA, ...) and regression models (e.g. XGBoost) were considered. Great emphasis was placed on ensuring that the model took into account seasonality and sales trends, but also paid attention to specific products, manufacturers, sizes, colors and their sales history.
The final solution was not based solely on Machine Learning models, but was integrated analytical system, which combined ML predictions with classic data analysis methods. This, in turn, allowed for optimal adjustment of order levels and minimization of the risk of outstanding inventories.
Automation of inventory management
Following use Machine Learning to implement predictive models, the key stage was their connection with the client's operational processes. We have introduced the following improvements:
- Implementation of the Reorder Point strategy – the system automatically initiates product orders when the ML model predicts that the stock level will not meet the future sales level in the near future
- Metoda Safety Stock – supported by ML algorithms allows you to dynamically maintain the appropriate level of safety stocks, eliminating the risk of sudden shortages.
- Order level optimization – algorithms adjust the size of orders in real time based on demand forecasts, limiting the freezing of capital in excess inventories.
Implementation results
1. Reduction of excess seasonal inventory
Thanks to the optimization of inventory levels, the value of products in stock decreased by 30%, which allowed us to avoid the need to organize costly sales for this part of the assortment.
2. Reducing warehouse losses
Warehouse losses were significantly reduced due to better availability of goods (which adequately met demand), and thanks to the implementation of the Safety Stock strategy, the client was able to sell 15% more bestsellers than before the implementation of the Alterdata solution.
3. Significant improvement in order management
Implementing automation in forecasting product demand allowed the logistics department to save time that was previously spent on optimizing the order process. Key calculations are now performed by ML models, which significantly support logistics processes.
4. More precise trend analysis
The use of models to forecast sales trends made it possible to more accurately predict future market expectations, i.e. identify product categories that will be in demand.
Summary
The project shows how modern technologies analytical and ML can revolutionize warehouse management in e-commerce. The implementation of predictive models, data centralization and process automation enabled the company to manage inventory more effectively, which translated into real savings and better customer service.
If your company is facing similar challenges, it's time for data-driven decisions - contact us!
