Accurate sales and demand prediction play a critical role in minimizing loss of revenue from stockouts, maintaining availability of products in the right quantity, and improving working capital efficiency by preventing inventory pileups.
Business Vertical:
e-Commerce Retailer
Region:
India
Size of the firm:
The client serves over 2 million people spread across 1000+ Indian cities annually, with the help of their 10+ warehouses.
Main Challenge:
Accurate Demand Prediction
Challenge
The client is a fast-growing e-Commerce company that sells health care products online. They have over 10 warehouses across the country dealing with over 20K SKUs each. In order for them to service their clients on time they have to plan the right stocking levels for each of their SKUs at their warehouses based on the expected demand in the cities around the warehouses. If they under predict and understock they will lose revenues and have dissatisfied customers. If they over predict and overstock, they are locking in too much capital on inventory that will expire and go to waste.
The client had been using traditional time-series forecasting methods like moving averages and regression for their demand prediction purposes. While these techniques are a good starting point, there are limitations on the level of accuracy that can be achieved using them. At the scale at which our client operates, even a small improvement in demand prediction accuracy means millions of dollars of incremental revenues and margin gains. Our client recognized that these additional gains in accuracy can only be achieved using more advanced AL/ML based prediction techniques and that’s where Qentelli comes into picture.
Solutions Proposed
- AI/ML based prediction models for every SKU, Warehouse combination
- Incorporate impact of demand influencing business and external factors
- Outlier detection to eliminate historical volatilities
- Automated model training and retraining
- Daily refresh of results to adopt to latest changes in trends
What Qentelli Did
After a careful assessment of the client’s status-quo and the target state of supply chain management, we have evaluated and selected a product that is a best fit for the client and also gives us the ability to fine tune the algorithm and continuously monitor the feedback. Picking an AI/ML solution is an ergonomic decision considering the demands of the project and ease of modifying the algorithms to suit the changing needs of the business. Here are some of the unique aspects of our approach:
Accurate bottom-up predictions - As demand patterns change, our algorithms continuously self-learn and adopt in-demand patterns for each SKU, warehouse combination.
Factor influence - Our predictions are enriched using impact of internal factors such as pricing and marketing spend as well as external factors such as holidays and weather.
Simulations - Our scenario planning functionality provides decision makers the capability to simulate the impact of different price points and marketing spend levels and pick the strategies that help maximize ROI.
Collaborative enhancements - Supply chain managers, category owners, marketing teams can all collaborate systematically, and override system generated projections to account for new strategies and business plans.
Scope of further improvement:
Incorporation of additional demand influencing signals like web traffic, views, promotions, etc.
Outcomes
15% - 33%
Improvement in Prediction Accuracy
~5%
Margin Gains
Our solution has been able to bring a prediction accuracy lift of 15% - 33% for different warehouse cities. This in turn has helped reduce Just in time (JIT) procurement and moved a significant share of purchasing to bulk procurement which provides margin gains of ~5% on what was previously procured in JIT.
%20(1).png){kind=small}
%20(1).jpg){kind=icon}