In Part I, we discussed how the exponential growth in eCommerce leads to an increase in cost of fulfillment and shipping as a percentage of net sales, driven by rising wage and consumer expectation. Worker wages in fulfilment centers sit at $17 an hour, almost two and a half times more than minimum wage. Consumers expect shipping cost to become cheaper and delivery to arrive faster at no additional cost. These two forces are driving the need for automation, particularly in the warehousing space where the environment is more constrained and highly labor intensive.
Similar to the way we analyze construction automation opportunities (See: The $400 Billion Market Part II: Opportunities in Construction Automation), we will deconstruct the warehouse workflow to analyze the challenges, automation feasibility, and competitive landscape.
Warehouse workflow: how goods flow in and out of a fulfilment center
In 2019, worker wage made up 47.6% as a share of the warehouse industry revenue, according to IBISWorld. Transportation and material moving worker wage makes $15-17 an hour and composes 72.7% of total employment in the warehousing industry. Material moving workers alone make up 62.0% of total employment in the industry, making the manual workforce the backbone of the industry both in terms of cost and employment. Figure 1 below depicts a typical work flow in a fulfillment center.
Unload: Unload palletized goods from truck onto dock.
QA/QC: Double check on the quality and quantity of goods.
Sort: Register and sort goods into the defined category and and allocate to the target storage
Place: Move goods to the target storage.
Pick: Upon purchase order, pick up the identified goods from the storage.
Pack: Grouped items together by each order and put into the same parcel.
Palletize: Group several parcels together on a palette and wrap with plastic.
Load: Load pallet into the carrier or fleet for delivery.
Figure 1. Typical workflow in a fulfillment center.
Mobility vs. Gripping, at Speed and with Safety
Before we dive into examining the opportunities created by automating this workflow, we must first understand the limitations of state of the art robotics technologies.
As a human being, we take our hands, legs, and eyes for granted. Building moving robots is difficult. Building a robot that can grab objects with high precision and the right amount of force is extremely difficult, and doing both while maintaining considerable speed, consistently, and safely remain a holy grail.
With the advancement of computer vision and object recognition technology, mobility is the easiest problem to solve among these. Because of this, we have seen a proliferation of “picking robots” that move from aisle to aisle and pick out or store a box and pallet full of goods.
A much harder problem to solve, however, is the mimicry of our hands. The process of picking up a mug seems like an ostensibly simple task but, in fact, is only accomplished by a complex network of circuitry inside our brain. We have to determine how many fingers to use to grab the mug, how best to fit the finger around a mug, while factoring in the physics of friction and gravity, all in a split second.
Products come in a variety of shapes and sizes, adding another layer of complexity. Perhaps the most challenging, however, is the balance between speed and safety. If a robot moves too fast and is unable to stop in time, it risks injuring coworkers; move too slow and it would not fulfill the orders during the Christmas rush.
Process by Process and Opportunity by Opportunity
Based on our understanding of the workflow, limitation and capability of robotic technologies, we have analyzed each of the warehouse workflows and examined automation opportunities within each of the processes. This is presented in Table 1 below.
While most VC investments have aimed at robotics companies solving the picking and storing process, we believe that the most lucrative market opportunities for logistic automation lie in breakthrough technologies that mimic the working of our hands to automate the picking and packing process, which are the most labor-intensive part of the fulfillment process.
Bloodbath Among the Pickers of the Fulfillment World
The effort to add autonomy to warehouses is not new. One of the most common advanced warehouse automation solutions is an Automated Storage and Retrieval System (AS/RS), which uses a computer-controlled system to automatically place and retrieve loads from a defined storage. The shelves are very densely placed together and can be as tall as ten storage buildings.
Another solution involves the use of Automated Guided Vehicles (AGV), which follows fixed routes, usually along wires and magnets embedded in the ground. AGV offers more flexibility than AS/RS but generally comes at a higher cost.
A new generation robot that tackles the “pick and place” workflow is called Autonomous Mobile Robot (AMR). Rather than being restricted to fixed routes, an AMR navigates the floor dynamically using a map. AMR can change its route on the fly, adjusting to changes in order and reacting to people, cars, forklifts and more.
IAM Robotics is one such company that offers a mobile picking robot called Swift, which operates with an extended arm to perform the picking operations with the help of a vacuum based end-effector. There are at least several dozen AMR companies globally tackling the picking and placing operations in the warehouse.
AMR requires that the fulfilment center invest in new hardware. To circumvent the need to deploy capital into new CAPEX, companies such as Vecna Robotics retrofit existing forklift and turn them autonomous. Other competitors in the space include Sierra.ai, Third Wave Automation, and Baylo, an Amazon funded company.
The pick and place operations is full of incumbent AS/RS and AGV players, but it is being challenged by AMR and autonomous forklift startups. The technology is no longer new, and while there is bound to be a big winner, the bigger question is whether you are prepared to jump into a food delivery space in 2015 when there were over 20 companies doing the (almost) exact same thing and able to correctly pick a Doordash out of them. While the opportunity is very interesting, the competitive landscape has softened our investment appetite.
Hand Mimicry is the Golden Opportunity
As mentioned above, hand mimicry is a very hard problem to solve in robotics. But the reason that this makes for a great investment opportunity is not only because the technology is difficult to duplicate – leading to fewer competitors – but also because it is critical to automating some of the most laborious processes in the warehouse workflow. Combined, we believe the technological defensibility and size of opportunity would lead to higher exit values of the winning companies.
One such process is sortation. In a high volume environment, the sorting process is distinct from packing to ensure quality control of order correctness. Once products are picked from shelves, they are brought to the “put wall” station. Here a tote of batch-picked, mixed SKU merchandise is sorted into individual, direct-to-consumer or ecommerce orders.
When the merchandise is brought to the put wall, the operator scans an item and immediately puts it into the bin illuminated by the put-to-light module. He or she then pushes the button/light to turn off the light and confirm the put.
Sorting requires an operator to do three things:
- Pick individual products from a tote
- Scan the product
- Place the product into the correct bin (and confirm the put by turning off the lights).
This requires a robot with object recognition technology to identify the correct product, the gripping technology that can handle a different shape and size products, and ability to recognize the “environment” in the bin to correct place the object into the correct bin (i.e. the bin may already have several products and it needs to be placed in a specific way).
Again, human beings do this very easily, but robots do not. There are multiple problems the company needs to solve to automate this specific process. One such company that attempts such a feat is called Ambidextrous, a Berkeley-based company that is attempting to solve the put wall sortation problem.
Still, solving a sorting problem in fulfilment center requires one extra layer many griping-based companies may not think about: mobility.
A sorter typically does not stay in one location. A put wall is, as its name suggests, a wall. The robot must be able to move across the wall to different locations with ease because products do not necessarily converge to the exact same spot.
Once the product is sorted on one side of the put wall, packers (or a line of packers) stand on the other side of the put wall then pack the products into a box.
The gripping technology for packing is much harder than sorting. It needs to be able to manipulate and reorient objects according to other products in the same box; the latter of which requires advanced AI algorithms to calculate different configurations that saves the most space.
Because of such technical complexity, another packing operation that is also labor intensive may be a better starting point. Southie Autonomy tackles co-packing operations for FMCG, which typically package the same type of products in final finished packaging. Contract packagers can be tasked with something as simple as adding a barcode sticker to a product or as complex as planning, designing, producing and fulfilling the entire package. However, they do so by building a no-code software that allows engineers to program exiting packing robots within one hour.
Next Opportunity: Between Fulfillment Center and Consumer
Automating ecommerce operations is difficult because of the complexity of unstructured environments coupled with a variety of product SKUs that come in different shapes and sizes.
At Creative, we pay particular attention to not only the features the technologies need to offer, but the specific workflow and environment the technology needs to operate in, as well as the type of problem it needs to solve.
To demonstrate this, another process that requires a similar level of technology is the loading and unloading processes, by which the boxes are loaded into the truck. Unloading process, for example, while laborious, is not a good use case because the utilization of the robots will be extremely low. The truck driver spends most of their time driving the truck, not unloading the package. Such use cases are unlikely to generate high ROI and short payback period.
In Part III of the logistic automation article, we will turn our attention to the process after the package leaves the warehouse and examine whether the billions of dollars poured into automating the last mile delivery make any sense, and more importantly, in which case does it make and not make sense.