True story: last year, on a field trip to Beijing with some of our American portfolio founders, we were sitting down to eat lunch when one of our founders spoke admiringly of a panda-shaped toothpick holder on the restaurant table — we ordered the same item on JD.com for her a few seconds later and she got it by dinner-time.
Experiences like this are delightful for the customer and fascinating for the investor. What allows such a supercharged supply chain to exist? What hurdles have to be overcome? And, more importantly, what are potential future opportunities and what would it take to get there?
In the USA, over the last decade, we’ve all become spoiled by Amazon Prime’s two-day delivery option. Nowadays, we expect to order a hairband online and to have it delivered to our doorstep in two days’ time. Walmart, Ebay, and every other platform is trying to catch up.
In China, as the above story illustrates, the expectations are even more demanding — JD.com (a huge Chinese online retailer) makes 90% of its deliveries within 24 hours, with 57% arriving within 12 hours.
We’ve been digging deep into warehousing, supply chain and logistics over the last few months and are eager to share some of our initial findings and hypotheses. We’re eager to get feedback from others who know far more than us about this topic, whether it’s startup founders, warehouse managers, other investors, etc. Feel free to drop us a line at firstname.lastname@example.org!
For now, since supply chain is a huge topic area, we’ll just focus on warehouses and only those in the USA. First, we’ll start with the basics, then paint what we think the future might look like, and finally show some possible paths to getting there.
Lastly, we also went out to experts in warehousing to get their take on what are the most value add areas that a startup should be working on in this space. We’re huge fans of Guy Raz’s podcast “How I Built This”, where he interviews entrepreneurs about how they built their businesses. In a similar vein, we’re doing a “What To Build” series, starting with warehousing, to help entrepreneurs find the most critical problems to be solved within each industry vertical. The full series will be released by mid-May 2019.
We all know that a warehouse is a commercial building for storage of goods. However, there are actually many different types of warehouses, from distribution centers to fulfillment centers to cross-docking to cold-storage, etc. For now, we’ll focus on fulfillment centers — warehouses specially equipped to quickly ship smaller items to end customers such as you or I.
In broad strokes, we believe in a future where there will be:
- Mass customization: customers can get products tailored specifically to their needs (custom sizing, design, functionality, even pricing) produced (or assembled) as soon as the order is placed and delivered via their channel of choice.
- Rapid delivery: rather than large distribution centers located remotely that stock hundreds of thousands of SKUs, there will be a distributed network of smaller fulfillment centers producing and supplying goods in real-time (often leveraging existing unused infrastructure such as gas-stations, empty space in grocery stores, etc.) that allow precisely timed deliveries within as little as 20 minutes of order placement.
- Drastically improved speed & accuracy: using automation to supplement or replace human labor, with the end state of reaching “lights-out” levels where machines fill orders without any human intervention via technologies such as autonomous mobile robots (AGVs, forklifts, etc.), automated storage and retrieval systems (AS/RS), automated asset tracking technologies, high-speed conveyors, etc. in their daily operations.
We know we can’t predict the future and would love your input and challenges to our future vision. We’d love to learn from you!
How do we get there?
Let’s start with an understanding of where we are today, see who’s currently already innovating, and then talk about some potential opportunities that will lead us to the future state.
A. Why now?
If the earlier example didn’t catch your imagination, here’s some additional figures that might pique your interest in this space:
- In the past five years, warehouse construction spending increased 29% annually vs. in the previous 19 years, when it rose 1% annually.
- The average warehouse size since 2007 has increased ~140% to ~185,000 square feet and average height has increased almost four feet higher to 32.3 feet.
- Across all warehouses in the U.S., processes are getting increasingly more complex: moving many more smaller items at a much faster pace than before.
- Labor costs comprise ~65% of most warehouse facilities’ operating budgets and, from 2006 to 2016, the average hourly wage of all warehouse/logistics employees rose by 16%.
So what’s driving all of these changes?
Primarily it’s the rise of e-commerce, growing at ~15% annually in the U.S.
And secondarily, as mentioned earlier, it’s the spoiling of consumer expectations by Amazon Prime. This is forcing all e-commerce vendors to demand more of all the other players along the supply chain — from faster turnaround in manufacturing to faster deliveries to fulfillment centers and finally to faster last mile delivery to the end customer. Warehouses, and especially fulfillment centers, are trying to keep up with both demand, speed, and quality expectations — and the only way to keep up is to invest in better technology.
Moreover, online commerce is still only ~10% of all retail, suggesting that there’s much more room for future growth.
B. Who’s innovating in what areas?
There’s many different types of startups playing in the warehousing space so we’ll briefly go through an overview of five broad categories.
A big portion of startup activity in the warehousing space right now is focused on automating order picking, the process through which customers’ orders are gathered from warehouse storage locations, sorted into the right orders, and shipped out.
The focus on the order picking makes sense in that this process makes up more than half of a warehouse’s operating costs; breaking down order picking even further, traveling comprises ~55% of order picking time. Moreover, warehouse labor costs have risen ~15% over the last decade and it’s been harder for warehouse managers to staff quality employees.
Various startups are focused on reducing the travel time for order picking in different ways:
- Autonomous mobile robots guiding human pickers to the right locations within warehouses and providing info on what and how many to pick. Humans can be stationed within specific pick-zones. Examples of startups: Fetch, Six River, Locus.
- More automated goods to picker solutions, with variations on bringing whole shelves to humans to pick off of such as Kiva to only bringing the right container to pick from such as Exotec and Ocado. There are also versions specifically for micro-fulfillment centers, as in the case of Commonsense Robotics or Takeoff Technologies.
- Optimizing inventory storage — for example, using AI to place more frequently sold items in the shortest-distance locations (and/or grouped closer together) vs. previous rudimentary versions of this, there are now methods for dynamically adjusting the placement of these goods at levels of complexity previously impossible.
There’s another segment of startups that are focused on automating the actual order picking itself — whether it’s via robots assisting human pickers, AR solutions or even simple LED lighting systems that guide the picker to the right items to pick at the right times. This is separate from many other methods of sorting items including via conveyor belts, etc. One of the most difficult problems within bin picking right now is in the case of randomized bin picking, which is picking from a bin where random items are stacked against and on top of each other (ex. think about the returns process, where multiple random items need to be re-sorted and stored in a warehouse).
Randomized bin picking remains a difficult problem because an universal solution would need to hit upon, at a minimum, all of the below criteria:
- Fast speed: preferably 3–4 seconds per pick or less
- Ability to handle a wide range of products / items
- Ability to pick vertical to place horizontal or vice versa
- Accuracy: at least 3 sigma accuracy
- Robotic arm agnostic
One of our investments, Covariant.ai, is leveraging reinforcement and imitation learning to develop cognitive abilities for robots to operate in extremely complex, real world warehouse and logistics facilities. Their approach enables robotic automation for a wide range of processes that have not otherwise been possible to date.
Another major bucket of startups in the warehousing space are focused on creating better warehouse software tools whether it’s in warehouse management software (WMS), warehouse control system (WCS), warehouse execution system (WES), etc. The diagram below lays out some of their differences, which are increasing getting blurry as both WES and WCS are adopting features that historically have only been part of WMS software.
Source: Envista Corp.
These software providers are all generally riding the wave of fulfillment centers needing to upgrade their software to keep up with rapid consumer demand. For example, helping customers better handle omni-channel orders or even better processing of single orders vs. historical bulk orders. Larger startups who have been around for a number of years include TradeGecko, ShipHero, FishBowl, Logiwa, Deposco, etc.
There’s also been an increase in on-demand warehousing startups. The oldest and largest of such is Flexe, which caters to larger enterprise customers and fast growing e-commerce startups. Others include Flowspace, Stord, and Darkstore; there’s gradations in how much involvement each startup has with the warehouse operations.
Lastly, there are startups who are providing fulfillment services, in effect acting like 3PLs (third party logistics providers) for customers, albeit with variations in business model. Some of these startups are asset heavy such as ShipBob and ShipMonk, who own and run their own fulfillment centers across the US that are located close to major metropolitan centers for their e-commerce customers so that they can make faster last mile deliveries at lower rates. In contrast, Deliverr owns no assets but instead partners with under-utilized warehouses across the country to do fulfillment services for e-commerce customers who wish to have 2-day delivery without using Amazon Fulfillment Services.
C. Where are potentially unexplored opportunities or gaps? Aka What To build?
Now that we’ve mapped out a bit of what’s happening in the warehousing space, let’s dive deeper into where we think there’s still unexplored territory for startups.
Taking a first principles approach, we can play the mental game of: What if this assumption doesn’t hold true anymore / in the near future? What would that mean for this process and its resultant efficiency? How can AI and robotics be applied to increase efficiency, quality, and throughput?
For example, we believe that certain historical trade-offs within warehousing no longer hold in an era of real-time sensing, faster and more powerful compute, and smarter automation. Some trade-offs that we think are becoming less valid are:
- increased throughput vs. reduction in labor costs
- increased pick rates versus accuracy
- speed versus safety
- storage density versus quicker pallet, carton, piece unit extraction
- inventory holding costs versus cost of stock outs
Looking at the first listed trade-off of “increased throughput vs. reduction in labor costs”, we believe that there will be various ways to achieve both increased throughput while reducing or maintaining labor costs. Some of this might come in the form of truly intelligent WES, which are taking on more and more of the functionalities of both WMS and WCS; examples of what such software might be able to accomplish include better path optimizations for human pickers, more efficient storage for inventory, and more efficient traffic control for syncing across all types of mobile robots from different vendors.
While some of these ideas sound simple, they’re quite difficult to carry out. For example, in order to find the best true pick-path, the optimization algorithm must know the warehouse’s geometric layout, including distances between all pairs of storage locations and most current WMS do not have this level of information. However, this is all now technically feasible to implement. Moreover, better tracking of human picker performance can also lead to designing better pick paths based on individual human behavior. Similar lines of thought can be applied to all of the other historical trade-off assumptions, and more.
We also went out to experts in warehousing to get their take on what are the most value add areas that a startup should be working on in this space. Our interviews with warehousing experts are below (the full series can be found here):
- Gina Chung: Head of Innovation Americas @ DHL
- Peter Puchwein: Vice President Innovation of R&D @ KNAPP
- Claude Dinsmoor: General Manager General Industry and Automotive Segment Robotics Division @ FANUC America
- Prof. Benoit Montreuil, the Coca-Cola Material Handling & Distribution Chair and Professor and Co-Director Supply Chain & Logistics Institute at Georgia Tech
- Roger Counihan: Vice President of Sales @ Fortna
- Sune Stilling: Head of Growth and VC @ Maersk Group
- Rimas Kapeskas: Partner @ Cambridge Capital, ex-UPS Ventures
- David Schwebel: Senior Director of Business Development @ Swisslog