A few weeks ago, Takeoff Technologies filed for bankruptcy (Douglas Moran 2024). Takeoff Technologies was one of the first, if not the first start-up to coin the term Microfulfillment Center. Perhaps they’ll be rescued, I’d certainly be sorry to see them go and I have lots of respect for them trying to shake up the industry. And the warehouse automation industry certainly does need to be shaken up; the amount of legacy constraints, lack of learning and improvement, slow planning, expensive maintenance, and (my favorite) awful IT at incomprehensible cost is begging for disruption. Alas, it seems Takeoff Technology will not be the disruptor.
In this article, I will share some thoughts about Takeoff Technology and MFCs. In particular, I will explain why their lack of success was predictable, in spite of several good ideas, because there were clear technical and conceptual reasons why their business could not work.
The Background
Microfulfillment Centers (MFCs) have been in the news for a couple of years. A lot of money has been poured into startups like Takeoff Technologies and Fabric and was spent on marketing campaigns. For many brick and mortar chains, MFCs have been the go-to solution amidst the surge in demand for online orders triggered by Covid-19. By now (2024), MFCs have not been in the news for two years or so. At LogiMAT 2022 (the first LogiMAT post Covid) I didn’t see a single company advertising MFCs.
I have been rather skeptical of MFCs from the beginning. When everybody seems to think something is a great idea, however, you better check your own assumptions if you’re the only one to disagree. After repeatedly checking my assumptions for some years, and after some dozen conversations on the topic, I still think for most food retailers MFCs are a really bad idea.
It was back in 2017 or 2018 when a gentleman from Canadian grocer Loblaws told one of my colleagues that “Witron has eaten your lunch for case picking, now Takeoff Is eating your lunch for online fulfillment”. At the time, I was working for SSI Schäfer and I had no idea who Takeoff was and what they were doing. From the little research I did then, I found that their offering seemed to be mostly centered around software while they were using existing shuttle technology from Austrian warehouse automation provider Knapp. One unconventional idea was that they would place the fast-moving products in the automated system while the slow-moving products would remain in the store for manual store-picking and later consolidation with the share of the order picked in the automated system. I wasn’t sure if this was because they didn’t know you’d normally do it the other way around or if there were deeper thoughts behind this decision. Then I was busy again with other things and didn’t pay too much attention to Takeoff.
A little later, I came across Fabric (then CommonSense Robotics) and I saw a fancy video that suggested order fulfillment with their MFC solution in places as uncommon as a parking garage in the city center. I thought their technology was cool, it was different, but I couldn’t really see how it was much better than what existed.
Shortly after, Dematic began to roll out their marketing campaign for their own MFC solution. At the time, I worked for a sister company of Dematic and received a detailed presentation of the solution during a team meeting. Back in my office, I made a quick back-of-the-envelope calculation and found that the solution they promoted did not seem to make any sense in terms of labor productivity. Effectively, it was a copy of the Takeoff concept and all public videos showed a two- or three-aisle shuttle system that was supposed to be placed in the backroom of a store. The videos suggested an 800 items/h pick rate, which I thought was ridiculous. I shared my doubts with some people at Dematic who told me something to the effect of “we ran a simulation and we know it works”. I also made some public comments about my view of MFCs in general, which wasn’t a great idea given that Dematic was just spending millions to promote their solution, so I retracted whatever I could retract and kept my mouth shut. In the meanwhile, Amazon got one for free (I think, but please correct me if you know better), which “analysts” at investment banks (or one at least) proclaimed “validates the MFC model”, while “journalists” did exactly what they are expected to do, which is mindlessly repeating whatever somebody uttered without understanding what it means, whether it is important, or has any veracity whatsoever (cf. Redman 2020).
Then Covid-19 hit and many investors swapped their brains for face masks. Zero (or negative) interest rates, plenty of stimulus money, strong demand for solutions for online fulfillment solutions, and stunningly incompetent “market analysts” created fertile soil for one of the rare full-fledged hypes in the world of warehouse logistics. Everybody talked about MFCs, and all automation companies head over toes tried to dress their existing tech as MFCs. One investment bank (Jefferies, the same that thought Amazon’s deal with Dematic “validates the MFC model”) called Kroger’s decision to go with Ocado’s large-scale fulfillment solutions a “multi-year mistake” and downgraded them from buy to hold (Wu 2019). They compared a large fulfillment center like Ocado’s with MFCs, assumed they would enable the same productivity in UPH, and concluded that because MFCs are cheaper and closer to customers and everything else is the same, MFCs are better. Unfortunately, this is not a joke, I read the report and sent the authors plenty of comments, but they were not very interested. Also, “market analysts” predicted that THOUSANDS of MFCs would be installed by 2030 (LogisticsIQ, n.d.).[1]
Death by Unit Economics
Food retailing is an industry characterized by high logistical requirements and low profit margins. The two main reasons why large food retail chains are so extraordinarily successful financially are that (a) the business is fairly predictable and (b) large volumes are moved. After all, everyone has to eat, and there are many of us.
For decades – ever since the concept of the modern supermarket has become the dominant design for purchasing grocery products – customers have happily done their own order picking, quality control, and distribution. They drove to stores and collected items individually: inefficient, time consuming and expensive for them, but efficient for retail chains which basically had to replenish shelves and count money. Beautiful business! In Germany, where I live, supermarkets and discounters still distribute their weekly junk mail on paper (!) with their latest discounted offers, and to my amazement large parts of the population even appreciate it.
So – that’s where we’re coming from. In fact, that’s where we still are for the most part, because even though Covid-19 did accelerate the growth of (already fast-growing) e-Grocery sales, the vast majority of people still goes to stores. Very 1980s. I do it, too, but I’m also born in the 1980s.
The great thing if your customers are willing to do their own order picking, distribution and quality control is that it removes some of the most expensive and complicated activities from your business. The moment customers order online and expect retailers to extend their scope of operations, the cost of doing business rises sharply. Customers don’t seem to care, however, and tend to be hesitant or even unwilling to pay higher prices, partly because people tend not to assign any monetary value to their leisure time, even if their limited time on earth arguably is the most valuable possession they have.
Anyway. The move towards online business for food retailers comes with two big buckets of cost: (1) order picking and (2) distribution. A third one, customer service and complaint management, comes on top, though it represents a much smaller challenge compared to the first two. And the only way to make money with online grocery is to be extremely efficient in operations and distribution, the very activities customers used to carry out themselves. In short, unit economics – the ratio of revenue versus operational cost spent on each unit sold – have to be absolutely fantastic, else you are going to lose money with every order you receive. In fact, that’s reality for almost every online grocery business, and probably every single one owned by traditional brick and mortar chains: their operations lose money. Some can recover part of the losses with delivery fees, but bottom line still does not look good.
Choose Your Battle
When Covid-19 hit, the sudden surge in online orders caught the established chains mostly off-guard. Some of them have been playing around with online business for years, but you could clearly see that most of them neither understood nor wanted that business. It was something they knew they had to deal with at some point, a dark cloud over the horizon which senior management hoped would not arrive before their retirement. Obviously they did not want it, because why would you want to turn profitable store business into loss-making online business? (Clearly, those online pure-player start-ups that decided they want to be in that industry had a different perspective). More surprisingly, however, I found that many did not even understand it. I had several conversations with senior managers of established chains, intelligent, hard-working men, who shared with me their perspective that the only way to make money with e-Grocery is picking orders from stores, the reason being that the infrastructure exists and that delivery cost would be comparatively low due to (relative) proximity to customers.
While I will confidently say that they were wrong, they still have a point: delivery cost, as one major cost bucket, can be lower if you are closer to your customers. And maybe customers would even pick up pre-picked orders from the store. The latter model, curbside pick-up, is more dominant in the US than in Europe; however, when you can make your customers drive to the store, it will certainly help your economics. I will not explain in detail why store picking is a really bad idea for any serious amount of online orders. The only thing worse than the high degree of operational inefficiency of in-store picking and replenishment is the lack of inventory control, leading to unacceptable amounts of stock-outs and cancelled orderlines. Still: stores tend to be close to customers. At least they will be closer than any sizable distribution center for order fulfillment would typically be.
And this very thinking has been one of the driving forces behind Microfulfillment Centers. The other driving force has been: let’s use automation to address the other cost bucket, order picking, and win on both fronts. This sounds appealing, and by and large this has been the main selling point of MFC providers: use the existing store network, put some fancy high-tech “robots”[2] in the backrooms of stores for “automated” (sic) order picking, and benefit from low distribution cost/curbside pick-up.
Is it true, though? Can you have both, low distribution cost due to customer proximity and high operational efficiency on the tiny footprint available to you in existing stores?
No, you cannot. There is a trade-off which you cannot walk away from. Operational efficiency in e-Grocery requires scale – scale which you will not achieve from a single store location. Ocado (UK) understands that. Picnic (NL) understands that. Oda (NO) understands that. As for the rest, I’m not so sure.
Conceptual and Technical Problems
The small size of MFCs creates an unhealthy amount of overhead, and it exacerbates a variety of principal technical problems with shuttle technology – as used by Takeoff, Dematic, and Savoye, to name a few. In particular, it is the segmentation of capacity in shuttle system which, in combination with the order structure and sequencing requirements we find in online grocery, leads to performance imbalance and overall much lower output per shuttle aisle than would be expected if you just looked at the capacity shuttle systems can achieve. I have described this problem in detail in another article on the limitations of shuttle systems, so I won’t repeat all of this here. But the short version is: Shuttles are the worst technology choice available for small scale GtP systems for online grocery order fulfillment. There is nothing else that fits less. In larger systems, you can compensate for the conceptual problems of shuttles with higher inventory and sequencing buffers, neither of which is foreseen in MFCs due to space constraints.
Apart from the conceptual misfit of shuttle technology, the situation is also that small projects create a relatively high share of project management cost. This is simple to explain: there is a base effort in project management; the larger the project becomes, the lower the cost impact of the base effort. The same is true for maintenance and service. A large system has higher maintenance and service cost than a small system, but the cost is not as much higher as the system is larger, again leaving the smaller (MFC) system with a cost disadvantage.
Due to their small size, MFCs come with a very limited number of storage locations. The archetypal Takeoff and Dematic systems were designed for about 3,000 locations per aisle. With two aisles in the ambient temperature zone, we would have about 6,000 locations. That means you either put such a small number of SKUs in the system that you have to consolidate most orders with items from manual picking in the store, where the majority of SKUs remains – or you try to increase SKU density in the system by means of compartmentalizing totes, in which case you will spend a whole lot of time compacting half-filled compartmentalized storage totes. An engineer told me that in one of the few Tesco MFCs (they once wanted to build many more…) one FTE is dedicated just for tote compacting. Either way, you’d have to accept a significant loss of productivity due to the lack of size. And you’d still have to consolidate a significant share of the orders. In the US, where MFCs were supposed to be a big deal, the number of SKUs in grocery stores is very high, perhaps the highest in the (western?) world, which exacerbates the problem.
The point about lack of storage locations is independent of the technology choice, by the way. Some technologies, such as AutoStore, achieve higher storage density, as AutoStore bros on LinkedIn will argue. But read on to find out why this argument does not apply here.
The hype around MFCs begs the questions what’s wrong with retailers that many of them adopted (or tried at least) a solution that’s so obviously not fitting.
What’s Wrong with Retailers?
Let’s try to understand why some U.S. and European chains have opted for MFCs even though technically and economically they don’t seem like a horse to bet on.
US retail chains are sitting on a trillion dollars’ worth of retail real estate, and the cumulative value of store locations in Europe won’t be much different. Naturally, they want to use this capital and not let it go to waste. Any concept, any idea that helps them keep that capital working for them will be welcomed with open arms. We can’t discuss MFCs and how useful they are (or aren’t) without considering this fact. What should retail chains do: sell their assets? A competing chain will capture those stores and poach the customers. Clearly, this isn’t a viable strategy as long as online orders represent only a small fraction of total sales. MFCs help established chains extend the useful life of their stores.
Retail chain executives needed to signal to shareholders that they are taking action. No one knew how long we’d have to live with Covid, and it was completely unclear how the online market would develop. Of course, they could have decided to build large automated plants. But planning and building these large automated warehouses takes a lot of time. It can easily take one to two years to contract for such a project and start construction. Then it takes another 1.5 years or so to complete the project, and then it takes another 6 months to 1 year to get the system up and fully operational. So, if you wanted to make a statement to your shareholders, this clearly couldn’t be your only answer to Covid. You’ve got to have some quick wins to show for it.
So, nothing wrong with retailers. The incentive structure of publicly traded companies combined with ignorance (on the part of retailer management and “journalists” and “market analysts” who don’t know the first thing about warehouse logistics) and exaggerated claims by MFC providers led to the predictable result that “everyone” wanted to try this concept.
Would MFCs with AutoStore Work Better?
Takeoff Technologies used shuttle systems, which are great for some applications, but the worst technology imaginable for small-scale systems that have to cope with fairly strict sequencing requirements and the typical online grocery order structure of about 25 OL/O. So pretty much any other technology in a microfulfillment center is better suited, including AutoStore. H-E-B (of Texas) has therefore installed more than a handful of AutoStore systems, initially as MFCs, later ones as stand-alone facilities, with Swisslog as their integrator. In a response to my article about the limitations of shuttle systems, H-E-B folks told me that I had described exactly the reasons why they chose AutoStore over a shuttle solution.
Problems solved? MFCs with shuttles bad, MFCs with AutoStore good?
No, it’s not that simple.
There are some technical arguments in favor of AutoStore: All other things being equal, AutoStore has virtually no problems with unbalanced capacity utilization, which makes its performance more predictable. Also, the first few thousand products create a fairly steep ABC distribution of orderlines, and as you know, that should have a positive effect on AutoStore’s performance.
There is another aspect of AutoStore performance that is almost never talked about, and that is the SKUs that are touched per time unit. The SKUs touched per time unit (e.g., per day or per hour) provide information about the active (= picked) SKUs in a certain period of time. This is of great importance for AutoStore as it determines the penetration of the storage block. If the share of SKUs touched per time unit in an AutoStore block are high, this means that a high proportion of products had to be retrieved from the storage block in this time unit. Now comes the tricky part: In online grocery, the SKUs touched per day are often very high; the percentage is higher for companies with a lower SKU count than for companies with a higher SKU count. For online grocery retailers with less than 10,000 SKUs, this KPI is often over 90%. This means that the AutoStore robots effectively touch ALMOST EVERY SINGLE TOTE in the storage block in a day, rendering AutoStore’s “We have a natural ABC sortation and therefore don’t need to retrieve totes from the depths as often” argument pointless. This is especially true if you store the faster rather than the slower products in the storage block, which is the case for MFCs. You can have a perfectly steep ABC distribution, but with a high percentage of SKUs being touched, you will still be very busy digging out the totes. Combined with the high number of orderlines per order, which can negatively affect AutoStore’s performance, too, the only way to avoid a performance collapse is to build a relatively wide and shallow system. This makes the storage density very low, lower than of some competing technologies, and reduces the number of storage locations available. In contrast, Exotec can cope much better with a high proportion of SKUs handled per unit time and is insensitive to the shape of the ABC distribution. Unfortunately, their pick station design is immature, however, and the idea that their AGVs have to deliver storage totes to pick stations ties up about a quarter of a million euros of capital per pick station, if you want to ensure that there is always a queue of totes and the pickers are never starved of storage totes, which should almost always bring down their business case. Otherwise, they would probably have eaten everyone’s lunch in the MFC domain early on.
It’s no surprise that the AutoStore system installed at Knuspr, the German subsidiary of Czech online retailer Rohlik, is only four totes deep. It’s not an MFC and it’s not much bigger than an MFC, but again, they had to consider the impact of deep storage and the high proportion of SKUs touched on performance and drew the right conclusions for the system design. Whether AutoStore is still the right technology if you only store four totes deep is another question.
What Takeoff Technologies Got Right
Takeoff Technologies has entered a fairly competitive industry with fairly high barriers to entry. Not only do they deserve respect for that alone, but they’ve also recognized some things they can improve. Some things that don’t work well. Shuttle systems have been around for 25 years and conveyor-based transportation for more than 70 years. However, planning and implementation still seem to be a big deal, and (too much) maintenance is often required (more so for shuttles than conveyors). How is it that AutoStore is virtually maintenance-free and very reliable, but shuttle systems are not? How is it that the commissioning of shuttle systems is still very time-consuming and there have been no improvements for years? How is it that every GtP system with shuttles requires a lot of IT effort (thousands of hours), as if it were the first time anyone had tried this new concept?
Takeoff recognized early on that great potential can be tapped through standardization. Most GtP systems are conceptually similar, which should allow for quick planning, efficient project management, swift implementation, predictable maintenance requirements and very low IT efforts. And so they wanted to build the same system many times, which should unlock significant savings. After all, the requirements in online grocery retail are roughly the same for most providers.
Takeoff has tried to accomplish the goal in partnership with an established warehouse technology company, and while there are many good things to be said about Knapp, I don’t think innovation, simplification and streamlining are among their core strengths.
Also, Takeoff has chosen a market segment that is known for… let’s call it tough love. Food retailers are known for squeezing their suppliers. Whether this applied to Takeoff, too, I don’t know. Perhaps the urgency of grocery retailers’ need for an e-com solution overruled their natural behavior.
And that’s why I’m a bit sad that Takeoff Technologies is leaving the stage. Because I had hoped they could teach the established companies something. The resistance to learning among incumbents is significant, which is great for AutoStore, Exotec, Instock and all the other start-ups that have emerged in recent years. But there are many large projects that require the stability and experience of established companies. The challenge is to leverage these strengths without the burden of corporate processes from the 1980s, the miserable IT, communication patterns that make meetings in public administration look like TED Talks, and the sluggish response times reminiscent of continental drift.
To everyone at Takeoff Technologies now seeking new opportunities: It was worth the effort! I wish you the best of luck in your new endeavors and hope you will share your valuable lessons learnt to advance the warehouse automation industry.
References
Douglas Moran, Catherine. 2024. “Takeoff Technologies Files for Chapter 11.” Grocery Dive. May 31, 2024. https://www.grocerydive.com/news/takeoff-technologies-chapter-11-grocery-ecommerce/717640/.
LogisticsIQ. n.d. “Micro Fulfillment Market.” LogisticsIQ. Accessed June 12, 2024. https://www.thelogisticsiq.com/research/micro-fulfillment-market/.
Redman, Russel. 2020. “Analyst: Reported Amazon-Dematic Partnership ‘Validates the MFC Model.’” Supermarket News. February 21, 2020. https://www.supermarketnews.com/online-retail/analyst-reported-amazon-dematic-partnership-validates-mfc-model.
Wu, Jasmine. 2019. “Kroger Made a Multiyear Mistake in How It Delivers Its Groceries, Jefferies Says.” CNBC. October 10, 2019. https://www.cnbc.com/2019/10/10/this-fulfillment-method-could-a-multiyear-mistake-for-grocery-chains.html.
[1] ”Micro Fulfillment Market is expected to have a cumulative opportunity worth $32B in next 7 years by 2030 with an installed base of 5600+ MFCs.” (https://www.thelogisticsiq.com/research/micro-fulfillment-market/).
[2] Until not so long ago, we called shuttles simply shuttles. Now they are often called robots, especially by “journalists”. Just like people suddenly started to call AGVs robots, too.
