How endurance training pays off : Using demand forecasting to improve the Amazon Inventory Performance Index

Article Series 2 of 3

The second article in this series illustrates a few key concepts that will help to streamline the business planning processes and improve overall performance.

The Amazon Inventory Performance Index (IPI) and how Demand Forecasting and Portfolio Management can help.

Introduced in 2018 as a means to drive more efficiency into sales and inventory management, the Inventory Performance Index (IPI) is a score Amazon assesses to determine how well sellers are managing their inventory. You can score between 0 and 1000. An IPI score above 450 means your FBA inventory is performing well, and a score above 550 indicates your inventory is a top performer. A score under 350 could lead to limits being placed on your FBA storage and lead to overage fees, as outlined in the new storage limit policy 2.0k.

While Amazon is not fully transparent on how the score is calculated, it is commonly believed that the following factors play a key role in its calculation. To no surprise, all topics are heavily impacted by your planning performance. The factors are:

• Excess inventory, to reduce inventory carrying and storage costs
• In-stock inventory, to prevent stock outs and lost sales
• Stranded inventory, to ensure your inventory is available for purchase and delivery.

Now, where does endurance training come in, you ask?

Turns out, all the above factors (and more) are heavily influenced by the quality of your demand forecast and supporting processes. Its quality is what you can directly influence and control. Once the forecasts are available, it requires merely a few extra calculations to set the appropriate levels in your inventory management system. 

Now your system understands when to trigger notification for restocking. In combination with recurrent portfolio performance calculations (ABCxyz in combination with forecast performance) one can quickly decide which items to keep and which to discontinue. 

That entire process should be re-executed at least once a month, all outcomes stored for future review and potential corrective action. Let’s call it the monthly S&OP planning cycle. It entails the creation of planning levels, which provide the time series used in the portfolio management and forecasting calculations runs, that occur at the end of every month.

Let’s break down the components of this.

As an organization, you sell products via channels (market) to local customers (geography). To do this efficiently you utilize online marketplaces and their offerings e.g. Amazon’s Fulfillment via its own warehouses. Your business competes over price but also over its ability to deliver always and on time - high service level - while maintaining the lowest possible inventory position per Stock Keeping Unit (SKU).

A slightly more conceptual view of this scenario is the image below. The left hand side depicts the recurring nature of the planning cycle, the need to calculate, review, and action numbers monthly or weekly, while the three pyramids on the right are meant to convey the various levels of granularity that define your business and therefore require proper planning.

Let’s review an example. Assume company A sells cotton T-Shirts in colors red, green and white, via the Amazon and Shopify marketplace into the US. What planning would this entail and what business planning views could be extracted from this example?

Suppose we want to simply understand total sales over time. We would aggregate the red, green, and white T-Shirt into the ALL level for product (tip of the pyramid), we do the same for market, and since we only have one geographic location it's simple. 

The outcome, represented as a timeseries, is ALL-ALL-ALL over multiple months, with sales numbers in monthly aggregates. The level considered for our business planning activities (planning level) would be only one time series.

An SKU view of the same would have to take the different warehouse locations and channels into account, and consequently contain multiple time series with a much lower granularity e.g.

Each of these combinations should be created, forecasted, checked for performance, and reviewed to select those cases that allow for business planning and decision making. It might initially be counterintuitive to create planning views that don’t directly map with the data e.g. SKU view is available from data, all-all-all is not, but higher levels often make for much more stable history patterns, and therefore yield better performing forecast output. 

Without an automated approach, a system that handles the complexity is usually not feasible to gain these extra insights, one of the reasons why organizations only stick to SKU forecasting.

‍

About the author

Dirk is a successful serial entrepreneur with over 25 years of experience, driving digital transformation through strategic and technical leadership; including helping Fortune 50 and 100 companies across a broad array of industries deploy planning capabilities to significantly improve business performance.

He has held various senior leadership positions with companies that pioneered innovations in business planning technology, and delivered highly-respected portfolio, demand and supply chain, planning and management consulting services. He is widely respected as a thought leader around Digital Transformation, a frequent guest author, and book co-author and speaker. Dirk is also an avid ultra-marathoner, snowboarder, and art hobbyist.

‍