As Originally Published in Supply Chain Brain
The replacement of the horse and buggy with the automobile was, all things considered, a good thing for humanity. Replacing retail workers with robots sounds identical in theory: replace raw muscle with clean, efficient machines.
Walmart Inc. used to think so too. They partnered with Bossa Nova, a robotics company that wanted to replace retail employees’ need to ever have to check the inventory levels by having six-foot-tall robots trawl the aisles. But at the end of 2020 Walmart announced that they would be ending their contract with Bossa Nova. Five years, over 500 robots in more than 500 Walmart locations — all scrapped. And it wasn’t because the machines didn't do what they promised. The problem was far more fundamental.
Thiel’s law states: A startup messed up at its foundation cannot be fixed. In this case, the “messed up” part of inventory-counting robotics ventures has to do with the fact that they’re trying to solve a dynamic problem with a static solution.
The Challenge of Change
Horse-and-buggies and automobiles are static problems: how to get from point A to point B. Safety, speed, comfort, obviously all factor in — but the underlying goal of each method is simple transportation. Inventory counts, on the other hand, are not static problems, but rather dynamic. The goal of counting inventory completely depends, with each use case modifying the action taken after the count.
In some cases, inventory counts are static. Regulatory compliance, insurance audits, customers asking an employee “do you have any toothpaste” — in these cases a correct number is all that’s needed. But in some crucial cases, something must be done to the inventory after it's counted. And that entirely depends on what the retailer wants to accomplish. It turns out, humans adapt much more quickly to these changing purposes than robots can (so far, at least). Fundamentally, this is why Bossa Nova failed.
Walmart found that online order fulfillment exploded during the pandemic, causing a need for human workers to grab products for online fulfillment after counting. There’s also the need to rearrange store layout on a whim, based on expected demand (a practice Japan’s Seven Eleven has long been employing, to great success). Then there’s the need to remove spoilage. But current technology simply isn’t at the point where it can reliably discern expiration dates due to the huge variance in print styles and locations. The best way is still the good old-fashioned way — either a customer alerts an employee to a spoiled item, or employees comb their shelves and read them manually. Finally, inventory never stays put; customers constantly move products to where it doesn’t belong. The Sisyphean task of re-placing inventory customers have moved would need a fleet of robots dedicated solely to solving this one issue.
These are only some of the issues. As retailers scale, and technology changes, the need to physically interact with inventory changes in tandem. And so far, robots simply can’t keep up with the need to constantly adapt to these dynamic changes.
Until AI is advanced enough to both understand and manage these dynamic tasks, and robotics become nimble enough to perform them, we will not see much change. A human’s ability to adapt is simply too useful a resource to retailers.
That does not mean robotics can’t help. Walmart is still toying with the idea of using robots as floor scrubbers for static tasks like basic hygiene, and other companies like Sam’s Club are looking into simpler versions of inventory counting robotics. According to a report from Deloitte, 74% of supply chain managers plan to incorporate robotics in their logistics systems in the next decade. The real question, though, is whether these robotic implementations will be adaptive.
In this sense, robotics will be useful only in as much as they support adaptive behavior. For example, an employee’s ability to log which areas of a shelf are disorganized, and have a robot sweep over for rearrangement. The key point is that technology is only effective when it can successfully replace an entire process, or when it fits with and is adapted to the current process.
Replacing entire dynamic processes is, so far, out of the question. Robotics must therefore meet two crucial criteria to fit existing workflows:
Adaptability. Seasonality or cyclical consumer demand can necessitate a business’ throughput to increase by multiple times. Static automation systems simply can’t cope, and even sometimes detract productivity, in this scenario. Owners need to be able to both quickly and easily modify robotic workflow to account for fluctuating demand, without needing to constantly contact the robotics company for costly and time-consuming upgrades. This is certainly true for larger retailers, where dynamic changes in demand fluctuate across hundreds of thousands of unique products. Smaller retailers are hurt by this too, but usually in a different regard. The sunk-cost of static automation begins to determine owners’ workflow instead of the other way around.
Scalability. Investing in robotics ventures requires the need for retailers to see years into the future and predict the application of static automation processes. Smaller retailers simply can’t compete in this regard. The uncertainty of these ventures makes the decision almost always too costly to even consider adopting, whereas larger retailers like Walmart can swallow the risk of testing these ventures at any of their locations. The more static the automation, the more risk calculations compound for retailers: Will this robot be relevant in five years’ time? Will this problem even be present in five years’ time?
If robots are adaptive to scale, smaller retailers can match automation to fluctuations in business needs at any point along the growth process. Costs decrease as adaptive robotics ramp up productivity and decrease the need for static automation systems at every level. Risk calculations decrease in scale, as adaptivity reduces the risk of future redundancy.