Updated: Jun 20
Shortages are now excesses. What happened?
Johnny Miller was one of the greatest golfers of all time. His career faded in the late 80s and 90s, but his performance in the 1970s represents one of the great multi-year streaks of all time. One point of advice that Johnny often gives is “when you get into trouble, the first thing you need to do is get out of trouble.” While he was talking about what you should do when you hook your drive into the deep rough, this advice can be applied to many things in life. And, it can be applied to the current inventory position of many retailers, particularly multi-category retailers. Brian Cornell, CEO of Target, is certainly one who is currently applying Mr. Miller’s advice with a vengeance.
It seems that the limited-output garden hose of our global supply chain has suddenly sprung leaks of inventory all over the place. What happened and what can we learn from it?
That Was Then, This is Now
Last year’s headline supply chain story was continuous disruptions and a general lack of inventory across various product categories, particularly items related to staying at home, including furniture, exercise equipment, televisions, computers, electronics equipment, toys, and even Thanksgiving turkeys. Media attention reached a feverish pitch with seemingly everyone from individual companies to high level government officials questioning the design of global supply chains. While the use of the term “supply chain” had been growing for years, it is now part of the daily language of ordinary citizens. For better or worse, it seems that this will be a lasting impact of the pandemic.
Part of the discussion in all forums from media to academic circles was that years of cost cutting by companies had made supply chains “too lean.” We debunk this theory using multi-year aggregate data in the research article “Are Supply Chains Too Lean?”
Fast forward to today, June 2022. The headline story today is excess inventory.
Sometimes it’s useful to think the opposite, as George Costanza once did. Some call this “analysis of extremes.” If all the inventory you need is currently short and late, the opposite is excess and early and this opposite may be arriving right around the corner. That is a typical result of the bullwhip effect; we suggested as much in “The Whack-a-Mole Global Supply Chain.”
The industry proxies for the excess inventory situation are Target and Walmart, two of the most sophisticated supply chain operators in the retail industry. (Their combined $683B in annual revenue is roughly half the revenue of the 45 department store and discount retailers we track and roughly14% of all 242 retailers we track). Let’s examine their inventory profiles to see if we can gain any insights. This is not intended to be a critique of these fine retailers, but a view into wider retail industry dynamics.
Walmart and Target
Under very challenging conditions, Walmart and Target have both been very effective in procuring and transporting inventory and keeping store shelves stocked. In the great 2021 scramble for product, they won. Now, they, like others, are faced with the boomerang effect, likely caused by a confluence of factors. Let’s look at the magnitude of their excess inventory.
Historically, Walmart and Target run very similar seasonal inventory patterns, centered on the critical Q4 holiday season (note: the holiday season ends in calendar year Q4; both Walmart and Target have fiscal years that end on 1/31). Inventories typically build up significantly ahead of the season and then draw down equally significantly throughout the holiday and post-holiday seasons. Inventory turnover typically peaks in the post-holiday and summer quarters (indicating the lowest inventory levels of the year).
This year was different. Inventories grew to levels comparable to years past, but the holiday and post-holiday drawdown did not happen. Inventory turns fell or stayed relatively flat from pre-season to post season. Figures 1 and 2 are pictures of the inventory turnover profiles of Walmart and Target for the years 2017 to 2021. In other words, Walmart and Target aggregate inventory positions on May 1 look as if both companies are ready for Christmas all over again. (Note: Summer = August 1, Pre-Holiday = November 1, Holiday = February 1 of the following year, and Post-Holiday = May 1 of the following year. Inventory turnover (turns) here is defined as COGS for the most recently reported four quarters divided by the inventory position of the most recently reported quarter).
Figure 1 – Walmart Seasonal Inventory Turnover Pattern
Figure 2 – Target Seasonal Inventory Turnover Pattern
How Much Excess?
If Walmart were operating at its historical inventory turns for the previous five years, its inventory position on May 1, 2022, would be $13.5 billion less, or 22% less. Likewise, if Target were operating at its historical inventory turns for the previous five years, its inventory position on May 1, 2022, would be $2.8 billion less, or 23% less. These two leaders are sitting on almost identical excess inventory on a percentage basis. Put another way, Walmart would have had to have realized roughly $18 billion in additional sales to arrive at its historical inventory position; Target would have had to have realized $3.8 billion in additional sales (this uses the average gross margin for both companies; it’s possible the inventory in question has higher gross margins typical of hard goods and electronics).
If these additional sales were achieved during the holiday season, both Walmart and Target would have had to realize more than 12 percentage points of additional sales. If these sales were achieved over the course of both the holiday and post-holiday quarters, both companies would have had to realize more than 6 percentage points in additional sales.
Both Walmart and Target had excellent Q4 results (which includes the holiday season). Walmart’s sales (net of divestitures) grew 7.6% and Target’s grew by 9%. It is unlikely that both companies were planning for an additional 6 percentage points of growth in either Q4 or Q1. It’s possible that some of the inventory that was supposed to arrive in November and December arrived in January or February, meaning lost holiday sales or substitute buying. Given the holiday results, it’s hard to believe that this is a significant part of the current inventory problem.
The fact that inventory turnover at the start of the holiday season in both the Walmart and Target cases was about what it has been historically indicates that: 1) inventory kept flowing in throughout the holidays and post-holiday; and 2) there were not sales to sop it up. This is a classic overshoot situation, that on the surface, is indicative of bullwhip effect behaviors (more on this in a minute).
Now, let’s come back to Johnny Miller and Brian Cornell. The Target CEO is not just looking at his own inventory, but also the inventory of everyone else in making the decision to move quickly to get the ball back in the fairway. He’s not trying to thread the ball through some trees in the hopes of somehow pulling a par or birdie out of his hat. He’s taking his medicine, chipping out and moving on. He’ll take a bogey to avoid a double-bogey. In other words, incentives and markdowns will make it look like Christmas in June and July.
So, What Happened?
Note: this involves a fair amount of speculation.
Figures 1 and 2 indicate that inventory grew ahead of sales up to the holiday season and then kept pace with sales throughout the season. Some of this is undoubtedly due to over-ordering but it is also a reflection of shrinking lead times as port and other constraints relaxed. In other words, a four-month lead time that suddenly normalizes to a two-month lead time means goods are arriving ahead of when plans indicated. In fact, retailers could be hit with additive negative effects – some late goods finally arriving, and some other goods arriving early (or earlier than the extended lead times caused by the pandemic) due to normalization of supply chain challenges.
The other thing to consider is that lead times embedded in replenishment and other supply chain software systems are typically static; they are changed infrequently. Over the past 24 months, actual lead times have been changing weekly, if not daily. Most replenishment systems are not designed for this level of change. It is unknown whether this is part of the equation here, but it highlights the need for supply chain policies and structures to be much more dynamic. If they are not, systems (particularly those that have been tuned for scale) will operate using incorrect information.
It's also certain that inventory did not arrive for which orders were not originally issued. So, both operators knew that the inventory was going to eventually arrive. Furthermore, it’s certain that both were not projecting holiday sales to increase an additional 6 percentage points.
Sophisticated supply chain practitioners have visibility to a time-phased view of their orders, along with a time-phased view of replenishment plans at each point in the supply chain from production through distribution through retail and ultimately to the end customer. Last year, these plans were oscillating all over the place as disruptions including factory shutdowns, trucker shortages, port backups, container and other equipment shortages, and lack of warehouse capacity caused continuous changes to projected receipt times.
Continuous, floating bottlenecks made it impossible to synchronize the various assets and people in the supply chain to create reliable plans. The name of the game of late 2020 and all of 2021 was to get inventory in the door in any way possible, including chartering private vessels to circumvent port bottlenecks. Precision was temporarily trumped by brute force.
Nonetheless, an aggregated time-phased view of planned receipts matched against a similar aggregated view of planned demand would recognize mismatches that result in excess inventory. It’s hard to believe that planned receipts would be a surprise since they were based on committed orders. Planned demand is undoubtedly part of the problem, as companies forecasted (or speculated) that pandemic trends would continue for at least another quarter or two.
All that being said, it is now appropriate to revisit the bullwhip effect.
Here we restate the primary causes of the bullwhip effect. These undoubtedly compounded the effect of forecast miscalculations associated with how long pandemic-induced demand trends would last. The discussion here repeats part of last year’s article “The Whack-a-Mole Global Supply Chain.”
Lee, Padmanabhan, and Whang showed in their paper of more than twenty years ago that the bullwhip effect is the result of four sources – demand signal processing, supply shortages, order batching, and price fluctuations. These four sources were previously discussed in Forrester’s work, about sixty years ago (Forrester discussed them in terms of order and inventory policy, limited factory capacity, advertising, and order handling). Forrester also discussed at length the impact of information processing delays. Below is a description of the sources of the bullwhip effect, along with an assessment of whether these contributed to the excess inventory situation (Note: this is a subjective assessment).
Demand signal processing – this means that end-of-the-supply chain real demand changes are distorted as they are translated into upstream orders (i.e., as the demand signals are processed). Demand signal processing is the process of taking a demand signal and translating it into an upstream order. Forecast updates based on recent demand signals tend to exaggerate the update in the direction of the recent demand signals. Furthermore, the amplitude of the distortion increases with the replenishment lead time. Assessment: it is likely that this behavior was present as retailers had forecast bias in the direction of pandemic trends, and thus overshot the mark as those trends reversed.
The rationing game – this behavior occurs when there are supply shortages. It involves the behavior of customers when an upstream supplier must put customers on allocation because of shortages (for whatever reason). The game goes something like this: If the downstream customer node anticipates that it will be put on allocation by an upstream node (e.g., a manufacturer), it will rationally increase its order to try to get a larger percentage of the pie relative to its competitors. In situations where demand for an item has increased, the rationing game adds to the distortion caused by the previously discussed demand signal processing. Assessment: there is little question that retailers participated in fierce competition for inventory, employing extraordinary measures to secure it.
Order batching – in the case of multiple downstream customers, said customers may have different ordering patterns, based on different economics. These orders are consolidated at upstream nodes. There may be cases when many or all orders arrive in the same time bucket, leading to upstream distortions. Some of this is driven by company order patterns that overlap (company A and company B order at the same time), and some is driven by dynamics specific to certain industries and products. An example of the latter is the hockey stick end-of-quarter orders prevalent in many B2B industries. Assessment: there is no reason to think that this phenomenon occurred any more or less than usual.
Fluctuating prices – Time-windowed pricing strategies requiring time-windowed ordering results in upstream orders differing significantly from downstream demand. This is particularly true when end-consumer pricing is not synchronized with upstream price changes. When prices are relatively low, downstream customers will order more during the period under which the low price is in effect. Conversely, when prices are relatively high, customers will order less during the period. This behavior is typically seen most notably in promotion-driven industries such as CPG. Assessment: in today’s price inflation environment this behavior might be widespread across all industries, as each upstream node announces price increase deadlines to their customers, causing customers to pull ahead orders.
Information delays – Feedback systems with information delays tend to distort responses
For example, a 10% uptick in real demand at a retailer that is seen by a distributor a week later will tend to result in a larger than 10% response. Companies without strong visibility capability (either technology or process-driven) will experience delays in understanding the actual status of their orders. This will cause them to react by changing, deleting, or adding to their orders. Assessment: it is likely that this behavior is widespread in today’s environment.
As we stated last year, the clear conclusion here is that bullwhip-driving behavior distorted supply chain operations and likely contributed to the excess inventory situation in which many retailers find themselves.
What About Semiconductors?
Now, let’s turn briefly to perhaps the broadest shortage of all – semiconductors. Because semiconductors are now part of all reasonably sophisticated products from vacuum cleaners to washing machines to refrigerators to cars, shortages have widespread impact. The challenge seems most acute in the auto industry, as demand for computers and related products and sophisticated high-margin chips siphons capacity away from the auto industry's historical capacity allocation. As for the case of home goods, semiconductors faced the double whammy of above normal demand and below normal supply, along with extreme supply volatility.
Is it helpful to apply George Costanza’s opposite thinking to semiconductors? Semiconductors are certainly different than furniture and other home goods (although many of them now have embedded semiconductors). After all, the long-term prospects for semiconductors seem limitless, as every thing and every person becomes connected. This would seem to indicate that semiconductors will not follow the shortage-excess pattern.
But the semiconductor business has historically been an industry characterized by peaks and troughs. This is because the industry is asset-intensive; increasing semiconductor fabrication capacity takes a long time – you start investing this year and out pops a new fab plant several years from now. When new capacity comes online, there is a step function in units of capacity that then gets absorbed over multiple years. This is the definition of long-cycle industry. There is a natural overshoot in capacity that does not get absorbed all at once.
The semiconductor industry spends about 16% of revenue on capital expenditures (aggregate CAPEX as a percentage of aggregate revenue). TSMC, the largest semiconductor fabrication company spent more than 50% of revenue on capital expenditures in the last four quarters and appears poised to continue this investment level for the next couple of years (its average for the previous ten years was about 41% of revenue).
Figure 3 shows the year-over-year growth of the semiconductor industry for the past ten years, along with nominal global and US GDP growth. Note the significant saw-toothed pattern to semiconductor industry growth. This pattern is likely to continue. Thus, while semiconductors will not follow the pattern of excess inventory, the current shortage will likely abate, and we will be faced with strong availability and reduced prices at some point in the future.
Figure 3 – Year-over-Year Semiconductor Industry Growth (aggregate revenue growth)
Where Do we Go from Here?
The excess inventory will undoubtedly be cleared. Inventory turns will increase through the summer, but probably not back to historical levels. That will likely happen in 2023, with the historical pattern probably returning. The name of the game for the next two quarters is to rotate inventory so that the right inventory is in place on Nov 1, in time for the critical holiday season.
But what about more strategically?
This, and other situations brought about by the pandemic, raise questions about the extent to which software and technology solutions can and should help.
The long lead-time supply chain we currently have is the result of globalization, open trade, deregulation, government incentives, labor arbitrage, and of course, the box, better known as the shipping container. It has performed marvelously well over the past thirty years. While variability and volatility have increased steadily over those thirty years, their levels are nothing like what we have seen in the past 24 months. Furthermore, the thirty years have been characterized by relative price stability (low inflation). Price inflation adds a significant hidden accelerant to the already existing demand and supply volatility.
Much emphasis has been placed on AI and AI-derivative enterprise solutions in the past five years. The pandemic provided many lessons on their limitations. The excess inventory discussion highlights the need for solutions that are agile and scalable. The problem for large retailers is to have both agility and scale. Many of the systems in place at large retailers are tuned for scalability but are brittle to change. They must be changed according to prescribed processes and procedures; these processes and procedures don’t always move at the pace of business.
It’s time for a different approach – one that can scale to hundreds of millions of SKU-location counts while also providing the agility to run in-line scenarios and to adapt solution policies at the pace required by the market environment.
The automotive industry struggled with this problem for years – in a different context. This context was the large automotive companies that produced multiple millions of cars per year struggled to simultaneously explode their bills of materials and provide rapid what-if scenarios in a scalable way. Typically, you had to do the BOM explosion through a batch process and then surgically do offline what-if analyses and then retrofit decisions back into the mainline BOM explosion process. Now, there are solutions that allow for this to be done simultaneously. Kinaxis, where I am a board member, is one of the first to deliver such a solution.
Humans Plus AI
We have now learned that AI is not the answer and it is certainly not historically “more important than fire.” AI and AI-derivative algorithms are very important ingredients in current and future solutions that also include humans, heuristics, optimization, and much improved data pipelines that reduce latency between what is happening in the physical world and what is represented by the digital world.
Furthermore, the pandemic has highlighted more than anything that humans and AI (or software in general) are individually mutually necessary but not sufficient by themselves. Only when combined with the right balance can they provide practical, usable solutions. This, and the retail solution discussion broached here, will be topics for another day.