Introduction
In most projects where a warehouse operator is investigating the use of automation, there are a number of different routes that can be taken. I’m not just talking about the different types of technology that can be installed and used, but also the general direction the project can take in terms of scale and scope: The company could go “all-in” and try to move all storage and picking to an automated system, or it could choose to pick only a portion of all orderlines in an automated system and continue to work with the existing manual system (or an improved version of it) for the rest. There are good arguments for both options. In this text, I would like to emphasize the role of productivity[1] gains – and their diminishing nature – as an important criterion for design decisions.
At the Crossroads
As a project planner, you will come to a point where any further analysis depends on some general decisions that must be made. Often these decisions may not be made by you (the project planner), but by someone higher up in the hierarchy (if you are working on internal projects) or by a decision maker at the client.
Imagine looking at the results of your data analysis. As it stands today, the average orderlines picked per hour is in the range of 650. However, there are significant fluctuations, and during peak hours, about twice the volume of orderlines is picked. In addition, the B2C nature of the business brings a significant spike in the run-up to Black Friday, with daily picking volumes three times the normal average. You can also see that the popularity of products is very skewed, with a fairly steep ABC distribution and 10% of products producing more than 70% of orderlines.
If you are in retail, these statistics may sound familiar, as they are representative of many businesses. From a logistics perspective, most retailers are very similar.
If you are in the situation that you want to automate your warehouse operations, you can go about it in (roughly speaking) two ways:
- you “automate everything” and install a big goods-to-person picking system, or
- you opt for partial automation and you maintain a fully functional manual system while moving only a subset of products into the automated system
There is also a third variant, where you install a goods-to-person system that includes all products but only covers base demand while a manual system, that also includes all products, is used to meet peak demand. For each orderline, you would decide whether it would be picked in the manual or in the automated system. This third option can be attractive if the cost for warehouse space is low and demand fluctuations are very high. In this case, the savings from not having to install excess automation capacity to meet peak demand more than offset the cost of the additional warehouse space required to maintain the full-size manual system. For the sake of argument, I will disregard the third option for now.
How do you decide which way to go? What are the criteria? The decision is not straightforward, and I claim most analysts will struggle to provide a clear recommendation.
Discussion of Variant 1: Going All-in
Let’s discuss the simple yet expensive variant first. This is the option that will likely be recommended by warehouse automation companies. For most companies that want to go for automation, this will probably be the default option. From an analytical point of view, the solution is simple: If the product assortment consists of small products, you throw everything into bins and place all the bins in an automated storage and retrieval system of your liking for good-to-person picking. Traditionally, these would have been Miniload cranes or shuttle systems; in recent years, AutoStore has developed into one of the most popular solutions for automated small-parts storage and picking. AutoStore’s success has inspired many other companies to develop similar cube storage systems to the market, but with a huge customer base, mature technology, and a growing list of distributors, AutoStore so far remains the uncontested leader in the field. Which system is the right one depends on several factors, but most importantly on the combination of static and dynamic capacity, i.e., storage locations and storage/retrieval movements.
Having a “fully automated” system[2] is typically associated with the advantage of most labor savings, translating into the highest labor productivity as measured in units per labor hour (UPH), a KPI that has gained prominence in warehousing thanks to online grocery[3]. UPH is a very simple productivity KPI that compares the number of flow units that are leaving the system for distribution with the number of labor hours required to enable this flow across all process steps (goods-in, storage, picking, packing, truck loading…). If you think about it, you will certainly see that it is one of the most honest KPIs in logistics and production. It also explains why advertising goods-to-person picking systems with proclaimed increases in picking rates is misleading, if not dishonest, since the advertisement conveniently forgets to mention that additional activities, such as decanting and (often) compacting now are required which cost labor hours and which were not necessary in your old manual system, so that the effective productivity improvement of the automated system over the manual system almost always will be significantly lower than expected. Quit besides the fact that in many goods-to-person systems, not even the expected picking rates are achieved…but that’s a discussion for another article.
Another benefit of having a “fully automated” system is that it is one system (provided all your products are suited for one system). Once you reach that island, you are burning all ships and you will do whatever it takes to make it work. Also, from a control perspective, having one system is much simpler than having multiple systems.
Discussion of Variant II: Partial Automation
Let’s talk about the alternative to “going all-in”: automating storage and picking for only a subset of products. There are some strong arguments for this option, but it’s also a bit more complicated. Having only some products in, and picking only some products from, an automated system requires that the warehouse management system (and its subordinate systems) can manage both, the manual part as well as the automated part. And will consolidation of picked orders be required? For standard B2C retail, this is unlikely because most orders consist of only one or two orderlines. But it can happen and you need a process for that, too. Process complexity becomes IT complexity, and since most IT systems are terrible, sometimes you are better off having a less intelligent, less sliced up and simpler system. Sometimes.
Obviously, investment cost is lower when you build a smaller system. Earlier I argued that with a highly (or “fully”) automated system, you can have most labor savings. Conversely, with only a small system, the labor savings you can realize will tend to be lower. So, while labor productivity measured in units per labor hour will improve, it will probably not improve as much as with a larger automated system.
So, large automation investment = big productivity improvement and low automation investment = low productivity improvement. Is it that simple? Not quite.
UPH is Great. But…
Units per labor hour is a great indicator. There are good reasons why companies like Ocado and Oda use UPH as a primary indicator to measure their productivity improvements. But UPH is also limited… to labor. It only looks at the output generated with a certain amount of labor hours. Labor hours, however, are but one input factor. The more the process cost is dominated by labor as primary input factor, the more important an indicator UPH is. When we include automation equipment in the equation, however, we substitute labor with capital. Of course, UPH will go up! But capital is an input factor to productivity, too, and must not be ignored. UPH of 140, enabled by a huge investment in automation equipment, is far less impressive than UPH of 120 with no automation whatsoever. Ocado’s proclaimed labor productivity of > 200 UPH in Andover and Purfleet[4] is a lot of smoke and mirrors considering the massive capital investments made to enable this labor productivity. And we’re not even talking about the R&D and product management cost they have had to develop their own AS/RS and IT solution. In the end, the only productivity that matters is multi-factor productivity (also referred to as total factor productivity)[5], a KPI that includes all input factors (converted to a common currency unit like EUR or USD) and compares it with the output they are generating (again, in EUR or USD). Because it doesn’t matter whether you waste your money on expensive labor or on expensive automation equipment or on electricity needed to power your automation equipment. You want to understand how many Euros you get out for each Euro you spend, and if you are spending money on automation equipment, you want to understand if this helps your productivity or not. Because sometimes it might not – and you knew that already, because that’s why buying expensive automation equipment in countries with very low labor cost doesn’t pay off. You would still improve UPH, but the “H” in “UPH” is so cheap that it doesn’t really matter (the same is true if your product contribution is very high; in this case, too, you wouldn’t care about the cost of labor so much.)[6]
Now, I presume that many readers are familiar with the concepts explained hitherto, and I know from experience that many others are not. Many warehouses aren’t managed systematically, and many warehouse managers either have no concept of productivity at all or they equate productivity with pick rate. And no wonder: many providers of automated solutions merely advertise increased pick rates and sometimes call them productivity, which is only a fraction of truth since there’s much more to warehouse productivity than picking, even if you only looked at labor productivity (for a start, how about replenishment?).
If you read this far, you just finished the introduction. I had to introduce some thoughts and concepts to be able to make the point I want to make next.
Will More Automation Bring Better Productivity? The Diminishing Marginal Product of Capital
Let’s talk about the marginal product of capital.
The marginal product of capital (MPK) is a fundamental concept in the field of economics. To comprehend MPK, it is crucial to understand its connection to productivity. Productivity, as we have discussed before, measures the efficiency of converting inputs (such as labor and capital) into outputs. It serves as a yardstick for gauging how efficiently resources are utilized to generate economic value.
Now, let’s delve into MPK. The marginal product of capital refers to the additional output gained by employing one more unit of capital while keeping other inputs, such as labor, constant. In simpler terms, it quantifies the increase in production resulting from an incremental investment in capital. The relationship between MPK and productivity becomes evident when we consider multi-factor productivity.
Multi-factor productivity (MFP) takes into account the combined influence of various inputs, including capital and labor, on overall output. When MPK is high, it signifies that an additional unit of capital contributes significantly to output, thus enhancing productivity. Conversely, a low MPK implies diminishing returns – additional capital yields proportionally smaller output gains. In other words, you get increasingly less bang for the buck.
In case you struggle with the concept, let’s look at a close cousin of the production function from which the marginal product of capital is derived: the utility function. Imagine a hot summer day and unlimited access to cold beer. With every glass of beer you consume, you get some utility. The first glass of beer feels amazing. As the effect of the alcohol slowly kicks, the second glass of beer will feel even better. So, already now there is a difference in the additional utility you gain from having additional glasses of beer. After the fourth glass of beer you feel invincible, but you also realize that the fifth glass makes you feel “less better” than the previous one. And after seven glasses of beer or so the utility becomes negative: every additional drop makes you feel worse, not better. You can think of the marginal utility as a reverse U-shaped function, with the difference being that it probably isn’t quite as symmetrical.
And the same is true for the marginal product of capital in the context of warehouse automation. The reason this is important to know is that there is a widespread believe that a higher automation degree in a warehouse or factory leads to better productivity. And that’s not necessarily true. Yes, labor productivity (UPH) will typically increase with higher investments in automation (provided the system is well designed). The marginal product of capital will initially increase before you will experience positive but decreasing marginal returns on every Euro invested. As of a certain point, each additional increment in UPH will become so expensive that overall productivity will be declining. Once you notice during the planning process[7] that every additional improvement in UPH would come at disproportionately high cost for automation equipment, you should pause and carefully revisit the scale and scope of your system to make sure you’re not blasting more money than is helpful.
We could also explain the phenomenon quite simply: The 80/20 principle also applies to investments in automation. You can achieve 80% of the desired effect with 20% of the investment costs. But if you aim for “full automation” and want to maximize labor productivity, it will cost you the other 80% of your budget just to achieve an additional 20% of the benefit. Is it worth it? Sometimes it is. Sometimes it is not. It depends on your objectives. What has been the driver behind the decision to automate? Productivity improvements/cost savings? Availability of labor (or the lack thereof)? Space savings? If productivity improvements are your primary objective, you should leverage the 80/20 principle. If you can’t find warehouse workers, you will probably want to maximize UPH. And the same applies if space savings are your primary concern.
Conclusion
In this article we have looked at labor productivity, total factor productivity and the diminishing marginal product of capital. We explored the question of whether higher investment in warehouse automation technology leads to higher productivity, and we have found that this is not necessarily the case. There is a sweet spot beyond which every additional Euro invested will barely move the needle in terms of labor productivity. This means that beyond a certain point, it’s not worth investing more money in automation if productivity is the goal, because total factor productivity is likely to fall: you will have higher CAPEX than is justified by the OPEX savings. And while this isn’t rocket science and is obvious to economists, my impression that it isn’t all that clear to everyone I speak to in the logistics industry, either on the supplier side or the customer side. As we navigate the evolving landscape of warehouse automation, informed decisions based on a more nuanced understanding of productivity are critical to the commercial success of automation projects.
[1] Productivity is a core concept in Operations Management. It appears in Operations Management textbooks, but frequently does not receive the attention it deserves.
[2] “Fully automated” is quite a misnomer, however common it is, because the only thing automated usually is the storage and retrieval of bins and parts of transportation inside the warehouse, but nothing else.
[3] … but for whatever reason is largely neglected by many other warehouse operators, which is a big mistake.
[4] Ocado Group: H1 FY22 Results. Available online: https://ocadogroup.com/media/meakmqqc/fy22-h1-results-presentation-july-22.pdf, last access: 2023-09-10
[5] I recommend you ignore the Wikipedia page on Total Factor Productivity (https://en.wikipedia.org/wiki/Total_factor_productivity) as well as the Cobb-Douglas equation. The Wiki Just follow the argument and you will see
[6] So, UPH is most interesting where (a) process cost is mostly determined by labor cost, (b) labor cost is high, and (c) product contribution is low. It is most irrelevant where (a) labor cost represents only a small share of process cost, (b) labor cost is low, and (c) product contribution is high.
[7] You should during the planning process very consciously think about productivity improvement. While it will be difficult to consider all input factors for MFP, the labor productivity you can expect from the new system should be analyzed in detail. Without it, you cannot know if the investment you are about to make will ever pay off.
