Best Practices in Grocery Forecasting & Replenishment Process
Grocery industry is facing unprecedented challenges on account of rising value chain costs, supply disruptions and a continued increase in share of online channels which have lower profitability. The competition continues to intensify, with traditional grocers facing strong competition from digitally native companies. Against this backdrop, it is crucial for grocers to deliver a reliable experience to consumers through high on-shelf availability at lowest cost. A robust Forecasting & Replenishment process can facilitate this.
Forecasting & Replenishment (F&R) in grocery is a very critical process as it affects stock outs, wastage, markdowns, freshness, working capital requirement and unplanned stock transfers. Companies which adopt leading practices in forecasting and replenishment have upto 40-50% fewer stock outs and 30-40% lower wastage. Given the intense pressure on profitability, these staggering numbers, which directly affect both the top and bottom lines, are even more relevant at this time.
However, F&R in grocery is quite challenging due to large number of Stock Keeping Units (SKUs), frequent promotions, price changes and other events which impact demand, short shelf life for many items, stock accuracy issues and so on. So, how are leading companies able to significantly perform better? It is a combination of adopting and relentlessly executing leading practices.
We will go over ten key practices that can drastically improve grocery planning and lead to significantly better business outcomes.
- Deploy advanced algorithms – With the advent of more advanced Machine Learning (ML) based algorithms that consider a wide range of demand influencing factors, retailers now have the opportunity to improve forecast accuracy in many scenarios, where traditional rule-based algorithms fall short. Algorithms should be deployed at SKU-location-day or even intra-day level (especially for items with a short shelf life), and tailored to the retail business. For new products or new situations, advanced machine learning models can automatically determine suitable reference data, eliminating the need for manual references.
- Event modelling and management – Events (promotions, price changes, externalities, weather etc.) have a consequential impact on demand. As a best practice, there should be a strong governance to timely and correctly update the event repository, and then deploying this information in forecasting to improve accuracy. Promotions in certain items may affect demand for other items (eg: offer on a certain brand of toothpaste may cannibalize the sales of other brands). Such complexities in event modeling should be managed by the forecasting system. It is seen that leveraging ML based algorithms can improve forecast accuracy by up to 30-40% for items whose demand is impacted by internal and external events.
- Continuous review of forecast error – A continuous review of forecast error, such as identifying a systemic pattern (eg: high error for event days or specific SKUs, or consistent bias) can uncover improvement opportunities (for example, there may be a need to revise event modelling approach). In addition, comparing SKU demand volatility to forecast error can provide a good assessment of forecasting effectiveness. It should be noted however, that forecasting can only be done for forecastable components of demand, and thus boiling the ocean to improve accuracy may not be prudent beyond a certain point.
- Cost optimal ordering – The traditional method of calculating replenishment order quantity works in a sequential manner, adding rule-based quantity step wise. For example, a safety stock is added to the forecasted demand, followed by presentation quantity adjustment, and then logistics rounding rules are applied. As a result, the final quantity may or may not be in line with the original goals such as desired service level or potential spoilage. More comprehensive approaches calculate cost-optimal order quantity by evaluating multiple scenarios in parallel on targeted KPIs such as desired service level, stock at the end of the period, potential lost sales and so on. The order quantity that yields the least overall cost is then chosen. Because KPIs are calculated based on future stock projections, the demand forecast and its distribution are critical inputs for this calculation.
- Differentiated replenishment strategies – Depending on the nature of the product, different replenishment strategies should be used. For example, for some products, the goal may be to reduce inventory while maintaining a certain level of service. For others, filling up to a defined shelf capacity might be the better option; and for fresh products, optimizing costs to achieve an economic balance between minimizing both lost sales and spoilage may be the best option. Simulation capabilities can help assess the impact of configuration changes, before applying them.
- Inventory accuracy – The accurate and real-time availability of usable stock on hand is a crucial input in determining the ordering quantity. Often, this input may not be accurate (actual and system stock mismatch) leading to wrong ordering. It is critical to have a streamlined process that captures the physical stock correctly. Furthermore, the stock data should exclude the quantity that is due to expire. This is especially important for items with a short shelf life.
- Incorporate business constraints in order quantity – The calculated replenishment quantity should take into account business-defined constraints such as shelf space, minimum order quantity, pack configurations, order schedules and so on. This ensures that the replenishment quantity thus generated and to be ordered to DCs, does not lead to issues in execution. Forward looking order plans that include these constraints can provide input for distribution center planning.
- Maintain active SKU master data – With thousands of SKUs, it is critical to have an efficient, seamless and accurate process for maintaining the active list of SKUs for which planning needs to be done. For example, if an SKU has not sold in a store for 30 or 60 days, there should be a prompt to review the status of the item. Furthermore, if an SKU needs to be temporarily activated or deactivated for a set period of time, the process should enable it easily. In the absence of correct and up-to-date master data, the orders for incorrect SKUs may get generated.
- Exception visibility – As the planning becomes more automated, planners should focus on exceptions and use their experience to resolve such situations. The business should clearly define these exceptions, which can include conditions such as more than targeted error for critical SKUs, a significant change in demand level, consistent bias, high stock out items and so on. Based on these exceptions, planners can decide the best course of action to further improve and supplement the capabilities of the planning system.
- Monitor business outcomes – While forecast accuracy is a good intermediate measure for planning effectiveness, the most important metrics are business oriented, such as stock outs, wastage, freshness and so on. Because many variables influence ordering accuracy, stock outs are possible even with high forecast accuracy. As a result, it is a good practice to compare the performance of business KPIs against the target on a regular basis and to take corrective action in the event of deviations.
Because of the numerous variables and complexities involved in grocery planning, a lack of depth in the process will result in poor performance on business-critical KPIs. However, by putting in place the right systems and practices, you can significantly improve business outcomes and give your company a much-needed competitive advantage.
Do share your feedback, thoughts or questions on this article in the comments section.