To Automate, or Not to Automate? That Is the Question—Part 1
By Alberto Moel, Vice President Strategy and Partnerships, Veo Robotics
Welcome back, loyal reader, to yet another musing on the economics of automation. Our next couple of blog posts in this series will focus on the value of flexibility in manufacturing and how that determines when it does and doesn’t make sense to automate. In previous work, we have pointed out the trends toward mass customization and shorter product cycles and argued that remaining flexible is the best way to respond to these trends.
We have also made quite a concerted effort to show you that flexibility in manufacturing has value and that humans are best suited to provide that flexibility. We’ve even provided you with a case study showing how humans and machines provide the best economic solution to many manufacturing challenges.
But we haven’t really given you a (quantitative) sense of how much that flexibility is “worth,” when exercising that flexibility has value, and what kind of economic gains can be achieved by remaining flexible. Manufacturing engineers can likely intuit that flexibility has value and guess that while a short production run isn’t going to be amenable to a lot of automation, a stable and long-lived production process would be. But we can go much further and be quite rigorous in quantifying that tradeoff.
In today’s post we’ll wade into some options theory and discuss how the value of optionality should influence whether you decide to fully automate or use only human workers. In the next post, we’ll take this discussion to the next level, where we’ll confirm that the optimal approach is a mix of human flexibility and machine automation.
The value of flexibility
Let’s revisit the manufacturing step we described in the Human-Robot Collaboration Case Study: inserting rubber bushings into a machined suspension knuckle. And let’s say the production plan calls for the manufacture of 50,000 units of this knuckle-bushing subassembly. If you are pretty certain that you will only need to produce 50,000 units, building flexibility into your production system is an unnecessary cost. Instead, you can design a workflow with the best unit economics to make 50,000 parts, no more and no less.
But what would happen if, instead of needing 50,000 units, you end up having to produce, say, 100,000 units, or maybe only 10,000? Does your manufacturing process have the flexibility to respond to that unexpected event, to that uncertainty? And does your response materially alter your production economics?
If your manufacturing process is fully flexible, it will be able to easily and quickly ramp production up or down depending on demand, and that adjustment can be done costlessly so that unit economics are not affected. If, on the other hand, your process is inflexible, adjusting production volumes up or down will entail costly rigamarole (e.g., additional fixturing and programming, maybe a change of process) and your unit economics will be negatively affected.
The Value of Flexibility
The takeaway from all of this is that production flexibility has value, and its value is highest when you’re uncertain about your process requirements and lowest when you are certain.
This result doesn’t only apply to demand variations; it covers any system or process variable for which you have an uncertain outcome—cost structure, technology evolution, production prices, or anything else. Incorporating flexibility can be costly, so it only makes sense to build it into your system if its value is high. If it is not needed and the value is low, it is much more economical to leave it out and optimize for a narrower set of outcomes.
Optionality
The ideal approach to a situation in which you face uncertainty is to have the option (i.e., the right, but not the obligation) to respond to that uncertainty by paying some upfront cost. If the expected value of that option is higher than the premium, then you could build the option into your system in case you need it.1
Let’s focus again on the specific example of the insertion of three rubber bushings into a knuckle assembly. We explored three approaches for designing a workcell and production process for this task in our previous post: employing humans only, building full automation, or using a mixture of human + robot labor. And we looked at the unit economics of the different options to determine the optimal approach.
At one extreme, a fully-human process is certainly flexible—the human doing the work can adapt to changes and small discrepancies on the fly. But, in the end, it’s clear that a human-only process is inefficient—too much human effort is wasted, and the worker could suffer from repetitive stress injuries, accidents, or just plain boredom.
Unit Costs as a Function of Production Volume for Our Knuckle Assembly Example
At the other extreme is the zero-flexibility model of total automation. It requires a much more complex workcell and additional fixturing, programming, and design, but there are some clear benefits to full automation: increased speeds, labor cost savings, higher yields, and lower scrap rates. The issue is that the costs of building a fully-automated workcell may outweigh the benefits.
In the case study, we concluded that your approach should be determined by production volumes. For low production volumes, the fully-human process is a lower cost per unit. For high production volumes, the fully-automated process wins with greater efficiency and lower labor costs. But what if you are uncertain about the exact volume you need to produce?
Where is this value of flexibility you speak of?
Now, down to brass tacks. Let’s say that you are expecting a production volume of around 50,000 units. In this case, the unit economics of the fully-human (high-flexibility) and fully-automated (low-flexibility) processes are roughly similar, so you can choose whichever tickles your fancy.
But let’s add a wrinkle: although you expect a production volume of 50,000 bushing-knuckle assemblies, you’re uncertain as to how many you’ll actually need to manufacture. If you decide to go full automation and the actual number is below 50,000, you’ll be “overpaying” per unit. Similarly, if you stick to humans and it turns out that you need to make 100,000 units, the labor costs will shrink your profit margins into non-existence. What to do? How do we incorporate this uncertainty?
A simple but illustrative approach is to assume that the number of units to be produced is a random variable, which we model as a normal distribution with a mean of 50,000 units and a given standard deviation that captures the uncertainty around our mean guess.2 A higher standard deviation (in units of production) means we are more uncertain about how many units we’ll eventually end up making. Similarly, a lower standard deviation means we’re pretty sure of how many units we’ll need to manufacture.
So, if the uncertainty (as proxied by the standard deviation) is low, then the value of flexibility is low (or even negative), but if we really have low confidence on how many units we’ll need, this flexibility is quite valuable.
Expected Unit Costs as a Function of Standard Deviation (Uncertainty) for Our Knuckle Assembly Example
Some simple statistical manipulation allows us to estimate the expected cost per unit given this model production profile.
For low standard deviations (low uncertainty), we know the production profile will be very close to the expected 50,000 units. This means it’s best to take away flexibility and optionality and fully automate the process, as the value of that flexibility is actually negative!
However, if we only have a vague idea of how many units we’ll be making (high standard deviation and uncertainty), we should retain the flexibility of the fully-human workcell, even if it “looks” more expensive. In the extreme case where the standard deviation is 2x the mean of 50,000 units (in other words, we have no clue how many units will be needed), the value of retaining the flexibility of the fully-human cell is about 30% of the cost. A big number, to be sure.
Clearly, flexibility (and having the option to be flexible) is valuable in many cases but simply not worth it in others. In the next blog post, we’ll continue to build on this case study to show how being judicious about what kind of flexibility we keep and what kind we take away (i.e., automate) leads to the best financial outcome.
1 The branch of financial economics concerned with the valuation of such investment decisions is called Real Options Analysis, and it uses analogies to financial options such as call options (e.g., the right, but not the obligation to purchase a stock at a specific price at some specified time in the future). The field had its heyday in the late 1990s, but it all went haywire with pointless academic squabbles about the theory, which has limited its widespread application. A nice example of Sayre’s Law, as formulated by Henry Kissinger: The intensity of academic politics and the bitterness of it is in inverse proportion to the importance of the subject they're discussing. But the idea that flexibility has value is fundamental and a core tenet of valuation under uncertainty.
2 As a reminder, the actual number of units needed will land between +/- 1 standard deviation 68% of the time, and +/- 2 standard deviations 95% of the time. Read up on normal distribution if you need a quick refresher.