The one metric that highlights the challenge
Amazon is an e-commerce giant, an incredible success story and therefore a role model and a feared competitor. In this article, we take a deeper look into logistics and analyze the hidden and expensive problem behind the glittering fulfillment curtain.
There is one metric which made we wonder about the Amazon KIVA solution for warehouse automation. The Amazon Financial Reports show that Amazon’s fulfillment costs are growing much faster than sales. A meaningful indicator is the share of fulfillment costs in net sales. This was already at a very high rate of 10.2% in 2012, the year Amazon bought KIVA systems, and rose to a massive 16.4% 10 years later.
Share of fulfillment costs in net sales!
Given Amazon’s growth over these 10 years and the economies of scale that come with it, the exact opposite would have been expected. Instead, the costs of fulfillment are eating away at an ever-increasing portion of revenue. One might argue that the Amazon Prime delivery privileges play into this costs development. Firstly, the proportion of Prime customers cannot explain this increase, and it is not always the case that faster order processing is more expensive. So let’s look a little deeper.
The year 2012 marked an important turning point in warehouse automation at Amazon. In March 2012, the purchase of the robot start-up KIVA was announced. Earlier than many others, Amazon opted for a modern Automated Mobile Robots solution and the goods-to-person principle.
Deep Dive
Small transport machines that look like an over-sized vacuum cleaner robot drive under racks with an edge length of approx. 1 x 1 m and a height of approx. 2 m, lift them up and transport them to the order picker. Depending on which of the four sides of the rack the required item is on, the KIVA robot rotates on its own axis, presenting the required item to the picker. This ingenious idea seemed ideal for a company that is not only growing dynamically, but is also constantly expanding its range of items. This is why Amazon decided to buy the entire company behind it.
The KIVA system only requires an empty hall, a barcode grid stuck to the floor for orientation and a few charging stations for the KIVA robots. These set up the rows of racks themselves, move the individual racks to fill/replenish them and are then ready for operational picking operations.
This means that this solution is not only quick to implement, but also easy to scale and adapt to different items. Furthermore its features also apply to returns, a topic which is very important for most e-com operations. Nevertheless, it is precisely this automation concept that seems to be responsible for a large part of the high fulfillment costs at Amazon.
The KIVA principle – which is now offered in a similar form by various manufacturers and integrators worldwide – has four serious deficiencies. These are inherent in the solution itself and therefore cannot be solved without changing the design or concept.
Deficiency #1: it is not a universal solution
A basic principle of warehouse optimization is the application of the ABC rule. Articles are ranked according to their popularity – i.e. frequency of access or sale – from fast-moving items (A) down to rarely needed items (C).
KIVA is a goods-to-person (GtP) solution. This principle is particularly advantageous for so-called medium-moving items, i.e. items that extend over the middle area of the ABC curve. For items that are rarely needed (C), automation is usually not worthwhile and for fast-moving items (A), the GtP principle is inefficient because these items would be in almost constant motion within the GtP system. The use of for example flow channels or direct picking from a pallet is simpler and, above all, much more efficient for fast moving items.
The attempt to apply a GtP solution such as KIVA to the entire range of items therefore leads to inefficiency and high costs for both, fast-moving and slow-moving items. Now one could argue that a higher base efficiency of the KIVA solution could compensate for these disadvantages at the edges of the ABC curve.
In fact, however, KIVA is less efficient than other GtP concepts and thus the disadvantages at the edges of the ABC curve are even more serious. Let us take a closer look at why this is the case. Which brings us to
Deficiency #2: bad ergonomics
Mike Moutz, one of the founders of KIVA, promoted the new solution with a picker who simply extends his arm and a KIVA robot places a shelf with the desired item right in front of him. Well in approximately 1/3 of the cases it works very much like promoted by the KIVA inventors. Conversely, this means that in 2/3, hence the majority of cases it is not that easy.
In fact, KIVA is the GtP solution with probably the worst ergonomics of all GtP concepts. A KIVA rack is around 2m high. Minus the space for the KIVA robot to drive underneath, the items are distributed vertically from around 30-40 cm to around 2 m. For the average picker, the reach space between 80 cm and 120 cm height is considered ergonomically optimal. More than 2/3 of the items in the KIVA rack are therefore in places that should actually be avoided!
Anyone who has been to an automated Amazon warehouse has seen the small stepladders at the workstations to access items higher up in the KIVA rack and also unhealthy bending of the workers to reach the items at the lower levels. Every experienced warehouse operator knows that good ergonomics are an important factor for high efficiency and quality and therefore for low costs.
Deficiency #3: very low replenishment efficiency
A weak point of practically all GtP solutions is replenishment, i.e. the transfer to the buffer storage of the GtP system. With the predominant shuttle based storage and retrieval for GtP solutions, this means, for example, the transfer to the plastic totes of the shuttle storage system, which in turn supplies the GtP workstations.
This relocation is an additional touch and therefore reduces efficiency and cost-effectiveness. When filling a plastic tote, typically around 5-15 items of an article are relocated at once per hand movement. This means that typically the storage tote of the GtP system is refilled after 2-3 replenishment moves.
The KIVA solution is completely different. In order to achieve a high level of accessibility of the items and at the same time a low storage volume or required space for the KIVA system, the items are distributed in small quantities (1-5 items) per KIVA rack without additional load carriers.
Not only does this mean significantly more touches required for replenishment than other GtP solutions, but the disadvantage of poor ergonomics described above additionally comes into play. Refilling a container at an ergonomic workstation with high quantities is much more efficient than sorting 1-5 items into the KIVA shelves or compartments spread over as low as 30-40 cm up to heights of about 2m.
Deficiency #4: low storage density
At first glance, the KIVA solution looks good here: racks can be placed close together because the KIVA robots drive underneath the racks, lift the required rack and transport it to its destination. However, in order for the required rack to actually be transported, the KIVA system requires aisles and thus empty space – just like in any other warehouse. Otherwise a time consuming shuffling of racks would be required to “free” the desired rack for transport.
Much more important, however, is the low height of the KIVA racks. The height is limited at around 2 m, so that the order picker can still reach the items. But a typical warehouse is 6 to 10 m high. KIVA therefore either wastes a large building volume or intermediate levels / mezzanines are necessary. These are not only expensive, but also undermine the simple basic principle of KIVA. In addition to a vertical connection, a consolidation level or order container logistics now becomes also necessary.
Conclusion
Comparing these 4 deficiencies with the advantages of the KIVA principle mentioned at the beginning, it quickly becomes clear why this idea is not so brilliant after all and represents a significant contribution to the high fulfillment costs at Amazon.
If you follow the logistics news, you’ve probably already heard about test applications Amazon does with plastic totes and respective GtP processes. If I was in Sales or Solutions Development of a major warehouse automation provider I would reach out to Amazon to discuss what comes next after KIVA…