By: Gene Tyndall, President, MonarchFx
David Lengacher, Head of Data Science, MonarchFx

Introduction

This article extends the topic previously discussed in, The Challenges and Solutions for Sales Forecasting in eCommerce. We will address the question of how to best optimize inventories once the sales forecast is accepted.

Inventory optimization is complex, not only for eCommerce operations, but for all other channels. Over the years, statistical models and analytical methods have been developed for planning inventories and executing inventory controls.

Nevertheless, surveys often report that over 75 percent of all inventories are accurate. Inventory mistakes lead to billions in revenue lost due to markdowns or write-offs from excess inventories that cannot be sold. Physical audits often reveal up to 15 percent errors in inventory counts and issues such as shrink and fraud add up to even more. The rise in omnichannel has exacerbated this problem further, as has the proliferation of SKUs into the millions. Multi-echelon facilities (e.g., hub and spoke), shorter life cycles, and expanding assortments to compete in the world of fashion and new product releases, bring more complications.

The Challenges

Several challenges exist in traditional channels. Below we list the unique challenges for optimizing inventories in eCommerce.

  • The challenges begin with the demand forecasting process. As previously discussed, forecasting sales with its small and intermittent demand is notoriously difficult to achieve accurately. The less accurate the sales forecast the less accurate the stocking levels.
  • Traditional inventory management methods do not adapt well to eCommerce, when orders are picked in ‘eaches’. The tools have been developed for volume sales and bulk shipments.
  • The complexity of omnichannel fulfillment presents a greater number of variables influencing sales velocity – stores, online, and offline channels operate differently – with different KPI targets for each channel such as inventory turns.
  • With respect to retailers, over 80 percent have not digitized sufficiently to add analytical, predictive, and AI capabilities to support decisions on optimal stocking levels.
  • Distributed inventories or locating products closer to consumers, further complicates inventory optimization. The working capital required to increase inventory in these additional nodes can, if not optimized, expand greatly. 
  • Changing assortments also creates uncertainty, as they do in stores. These may differ for different channels, creating false stocking and replenishment points for different locations.
  • Different delivery methods also can create uncertainty. BOPIS, BORIS, and Buy in Store ship separately, can all affect optimized stocking levels.
  • The common goal of ‘Perfect Order Performance’ is increasingly expected by eCommerce customers, yet the vast majority of retailers and brands do not have reliable data or advanced analytics in place to capture it.

The Solutions

Five essential process improvement actions are needed to enable inventory optimization in the eCommerce world.

  1. The Process

    Our previous article addressed the need to transform the S&OP process in order to improve demand forecasting (the demand side). This is also relevant for inventory management (the supply side). The three-steps delineated in that article, together with smart technology, will upgrade the S&OP process by incorporating the following objectives in the supply process.

    • First, include the concept of stock ‘flows’ in planning for adequate inventories at points of fulfillment. While this objective may seem subtle, there is a special nuance between thinking of storage vs. thinking of flows. eCommerce implies fast-moving, rapid turns of ‘eaches’ not periodic pre-determined replenishments of pallets or cartons. This translates into smaller stock levels.
    • Second, re-evaluate the strategic sourcing strategies for eCommerce relevance. The least expensive production and/or distribution locations may work well for volumes and bulks, but their lead times may not work well for unpredictable eCommerce sales increases. Replenishments at forward stocking locations may need to be at least weekly, and perhaps, bi-weekly. 
    • Third, consider the overall supply chain network for goods ordering, production, sourcing, shipping, distribution, and fulfillment. While complex, the best companies are doing this in order to perform most profitably in the multi-channel world. A hybrid network approach that incorporates a multi-echelon model for eCommerce fulfillment and a high-volume approach for traditional customers, can maximize performance across all channels. The eFulfillment network can be a sub-set of the overall distribution strategy.
    • Fourth, understand the supply side overall. Channel inventories, supply chain risks, pipeline inventories, and sourcing strategies may need to change, as do product distribution strategies. The primary goal is to setup the new network model(s) so that all channels can be grown profitably.
    • Fifth, especially for retailers, recognize that consumers are demanding more to satisfy their shopping and searching needs for products. The growth of mobile devices presents the connection between customers, stores, and inventory information. Easy access to inventory is essential, regardless of the channel. There is no online or offline, there should be one brand, one channel, and one source of inventory. Technological silos are history.

    As earlier with sales forecasting, process improvements can be complex and are highly dependent on change management. Yet, solving these five process fixes is critical and urgent. Tompkins Consulting applies proven methods to re-design the S&OP process or to streamline it with executive leadership. Our consultants are experienced in helping retailers create a ‘unichannel’ approach.

  2. The Technology

    As with sales forecasting, technology for inventory optimization is constantly advancing. This objective differs from forecasting sales, as it is highly dependent on company strategies, distribution, multi-echelon networks, and inventory policies. Moreover, the expansion of eCommerce has complicated the locations for holding stocks and the volumes to be deployed.

    Inventory allocation is the industry term for the planning of assortments, stocking policies, and methods for determining what, when, and how much should be distributed where. Inventory optimization refers to the best possible allocations to maximize customer satisfaction and minimize costs. Again, the growth of eCommerce – with its demand volatility and speed – has made this objective even more complex.

    There are some excellent software applications in the market for inventory planning and several for inventory optimization. We were pleasantly surprised that the Vanguard Software Corporation model builders had included an inventory planning engine. Not only is it tied to demand forecasting, it also accommodates several inventory policies. We have worked with Vanguard to configure a platform that applies these policies in order to best approach inventory optimization for forward stocking locations. This is our Distributed Inventory Flow Forecasting (DIFF) model.

    DIFF Model

    DIFF Model

    Below are selected highlights of the functions and features of DIFF’s inventory optimization engine.

    • DIFF’s inventory optimization algorithms fall into three categories continuous, periodic, and duration-based. There are over a dozen base algorithms embedded in the model, customized models are also available.
    • The inventory planning engine is driven by the policies or business rules and the demand forecasts, which we described in the preceding article.
    • DIFF can accommodate inventory goals ranging from EOQ (for continuous replenishment), to fill rate-based (for periodic replenishment), to days-of-forecasted-supply (for homeostatic replenishment).
    • DIFF uses a stochastic approach to inventory optimization, which takes replenishment uncertainty into account and also accounts for demand volatility, one of the biggest challenges faced by the ‘supply side’. While high fill rates can be achieved with excessive inventories, improved operating margins can be achieved by understanding the sources of volatility and planning. The DIFF stochastic approach results in a better understanding of the inventory requirements.
    • The DIFF inventory engine pulls SKU-level forecasts from its forecasting engine and in-bound, on-hand, and shipment values from a WMS, or other inventory system of record.
    • DIFF also integrates events that may impact replenishment such as weather events and transportation disruptions. The ability to integrate these events into a replenishment schedule maximizes the probability of achieving desired fill rates.
    • Managing inventory ‘Flow’ is achieved by the deployment strategy of forward positioning and carrying primarily only the amount expected to turn most actively. By focusing on ‘Flow’ and high-turning SKUs, inventory is maintained at levels that balance demand servicing goals with minimizing working capital. To accomplish ‘Flow’, DIFF suggests more frequent replenishments versus episodic replenishments that result in large stocks.
    • Overall, DIFF optimizes stocks by executing timely and more frequent replenishments. Together with its sophisticated forecasting engine and inventory optimization engine, DIFF maximizes total inventory performance.

Conclusion

As with demand forecasting achieving inventory optimization for distributed logistics also presents several challenges that are both technological and process-based. The DIFF solution we have adopted for MonarchFx sellers supports both the volatility of demand and the complexities of inventory availability. The early results are proving the value of DIFF in minimizing working capital and inventory carrying costs, while providing a solution for volatile ordering.

This discussion on inventory optimization, together with our earlier discussion regarding sales forecasting, summarizes why we have configured DIFF to support MonarchFx sellers. Further information on the DIFF solution is available from our MonarchFx technology development team and on process improvement from our Tompkins Consulting team.