In this article, we want to explore the combination of goods-to-person (GtP) picking systems with zone picking systems. This particular solution design will be discussed in the context of high-performance picking requirements – systems that will have to cope with strong customer demand, i.e., several thousand orderlines per hour. Combining GtP picking with zone picking counts as advanced solution design; it is nothing a warehouse operator should aim for with a warehouse automation provider with no prior experience with that solution design.
First, we will look into the reasons as to why you combine these two picking systems (as opposed to just using one). Then, we will discuss control aspects of this solution. Finally, we will explore if there are restrictions with respect to the choice of the specific automated storage and retrieval system (AS/RS) one would use to power the GtP picking system.
Why Combining Zone Picking and Goods-to-Person Picking can Make Sense
Why would you even consider combining a zone picking system with a goods-to-person system? Why not just having either a zone picking systems or a goods-to-person system?
Each system has some unique advantages and disadvantages. Zone picking systems (sometimes referred to as zone routing systems) are great for a lower number of SKUs whereas in GtP systems you can store and pick a practically unlimited number of different SKUs. Also, zone picking systems are pretty low cost whereas GtP systems tend to be very expensive. To give you an idea of the cost difference: for the same number of SKUs and the same throughput requirements, the cost of GtP systems can easily be 10x the cost of zone picking systems, which is just one of several reasons why the SKU count you want to offer is a strategic choice and must be considered carefully, as I have explained in this article.
Now, sometimes your throughput requirements and your SKU count are so high that there is no way around GtP systems. GtP systems are powered by automated storage and retrieval systems (AS/RS), and there are various technical solutions from many different vendors on the market that will do the job, such as miniload cranes, shuttle systems, carousels, AutoStore, Exotec‘s Skypod, Attabotics, …. as well as AMR-based systems from Swisslog, Grenzebach, GreyOrange, Geek+, or Quicktron, to name just a few. Note that for some order structures, some AS/RS will do better than others, and shuttle systems in particular have some challenges dealing with the order structure typically found in eGrocery, as I have explained in detail in this article.
As so many other things in life, demand for products of most companies in most industries follow a Pareto-type distribution (which in this article I will refer to as ABC distribution): a small share of the total SKU count generates the largest share of sales (these are the A-products or fast-movers). Accordingly, in the warehouse a small number of products generates the largest amount of picks. For most people in the logistics industry, this is common knowledge. And many manually operated warehouses are organized accordingly: fast-moving products are easily accessible, slow-moving products are stored somewhere in the back of the warehouse.
When automating warehouses, planning engineers (should) pay attention to the ABC distribution, too. Given space or capacity constraints, the fastest-moving products are often picked outside the automated system. Another reason to pick fast-movers outside a GtP system is that the overall productivity of picking them manually straight of the pallet often times is higher than unpacking them, decanting them into storage bins, and picking them in the GtP system.
However, finding a solution for fast-mover picking outside the GtP system has quite significant impact on the overall investment in the automated system, too. At this point, it is important to understand that the design and size of AS/RS is driven either by the need for storage locations (static capacity requirements) or by the need for storage and retrieval capacity (dynamic capacity requirements). In every well-dimensioned AS/RS, static and dynamic capacity requirements need to be balanced.1
Neither do you want to have many more storage locations than you need just because you need more miniload cranes, more shuttles, or more AutoStore bots to be able to retrieve all the storage bins for GtP picking, nor do you want to have more miniload cranes, shuttles, or AutoStore bots than you need just because you need more rack aisles to store all the products you want have in your automated storage system. Such imbalance is expensive and diminished the business case of the automated system.
Let’s assume a classical 80-20 ABC distribution, meaning that 20% of products generate 80% of the picks. And, for the sake of clarity, let’s assume your hourly picking volume amounts to 10.000 orderlines (OL) and your SKU count is 10.000. Picking 10.000 OL/h in a GtP system would require between 12 and 16 shuttle aisles (depending on number of shuttles, number of lifts, order structure, number of SKUs, pace at GtP stations…). If you take the 20% of fastest-moving products out of the GtP, you reduce the dynamic capacity requirements by 80%. That means, you can build an 80% smaller – and cheaper! – GtP system. Only 2.000 OL/h remain in the GtP system, which can be done with four shuttle aisles. Note that the average cost for a shuttle aisle with periphery is about €1M.
Now, you will have to take care of the 2.000 SKUs/8.000 OL/h outside the GtP system. And a zone picking system will do the job – at a small share of the cost of the shuttle aisles you have saved. Figure 1 illustrates the split.
That’s why it makes sense to combine zone picking systems with GtP systems in high-performance systems. You can reduce the investment cost of the overall system by quite a bit.
Control of Picking Orders
From a hardware point of view, the issue seems straightforward, and the CAPEX benefits are undeniable. Depending on the order structure, however, making such a combined system work smoothly is not simple. And while it makes lots of sense for most AS/RS, there are exceptions (see last paragraph of the article).
In order to be able to pick in a GtP pick station, the arrival of storage bins (pick-from) and order bins (pick-to) has to be synchronized. This requires some computations and optimization on part of the warehouse control system (WCS), taking into account (for instance) the queue of orders to be processed, the utilization of the AS/RS, and the travel time from the AS/RS to the GtP picking stations.
Similarly, a zone picking system can only operate at high performance levels when the release of orders into the system is well adjusted with (among other things) the utilization of the pick station the order bins need to visit.2
In the simplest case, orders are small and order bins have to visit only one system to be completed. In that case, there are no difficulties related to the control of order release and flow of order bins.
Things are different, however, if orders are large and frequently require picking in both systems to be completed, as it is the case for eGrocery orders with > 25 OL/order on average and a high probability that some slow-mover will be included in most orders. Since both picking systems require optimized order release to reach full performance, it is even more difficult to release orders such that both systems operate at peak performance.
If you release orders to the zone picking system first, the zone picking system will be fine. Order bins are virgin and the order release algorithm of the zone picking system can send bins to any picking station, taking into account station utilization. When these order bins continue to the GtP system for completion, however, they are not virgin any longer; they have been assigned a specific order and they require specific storage bins to arrive at the GtP stations for the picker to be able to finish the order. This implies that the AS/RS has to react, not least because arrival times from the zone picking system are somewhat unpredictable. The AS/RS will have to send storage bins to picking stations without the WCS first going through its careful and deliberate order release logic.
If, on the other hand, you release orders to the GtP system first, the GtP system will be fine. Order bins at the GtP stations are virgin and any order for which storage totes are being retrieved from the AS/RS can be picked immediately. When these order bins continue to the zone picking system, again, they are not virgin any longer. They belong to a specific order and they need to visit specific stations in the zone picking system. The order release algorithm of the zone picking system has very limited possibilities to take into account any of the parameters it would normally look at to enable even utilization of the various pick zones.
So, what’s better? Starting orders in the GtP system and sacrificing efficiency in the zone picking system? Or starting orders in the zone picking system and making life difficult for the GtP system?
Well, there is no clear answer to this questions. You can make it work either way. And either way requires that you include extensive (1) buffer locations for order bins travelling from one system to the other, as well as (2) sortation capabilities that allow you to alter the sequence of order bins that are being forwarded to the second picking system. Moreover, if you decide to start orders in the zone picking system and forward bins to the GtP, there is one important implication for GtP pick station design: pickers should be able to pick to pick several open orders at the same time (1:x design). Do not build 1:1 pick stations (1 source bin, 1 target bin): such stations will impose strict sequencing requirements on the system and thus will make synchronization between non-virgin order bins arriving from the zone picking system and storage bins arriving from the shuttle system even more difficult, exacerbating the reactive behavior of the AS/RS and diminishing its performance. In a 1:4 pick station, on the other hand, the WCS and the AS/RS can do some optimization and retrieve suitable storage bins for any of the four order bins at the station.
The Caveat: The Type of AS/RS Matters
In spite of the complexity of order release control and order bin flow between picking systems, the combination of zone picking and GtP system can make overwhelming sense from a budget as well as from a productivity perspective. It is obvious that you can save significant CAPEX if you don’t have to build the GtP system multiple times larger because of dynamic performance requirements to accommodate the fast-moving articles. And if you run the numbers and compute overall productivity, you will find that storing fast-movers in the GtP system may diminish your productivity and not improve it.
Once you remove fast-movers from the GtP system, the ABC distribution of the articles stored in the AS/RS becomes flatter. It is easy to see why: the first, steep part of the ABC distribution is cut off, and what remains is the almost flat part of the curve – the long tail. Now, for most AS/RS, this does not matter so much. Miniloads, shuttle systems, Exotec’s Skypod, Attabotics’ 3D hive, AMR-based systems… they are largely insensitive to the ABC distribution and will perform well even if only working with the long-tail of products. The reason is that these systems typically store single-deep or double-deep, and thus the amount of unproductive bin retrievals in order to get access to the bins the system actually needs will be limited. This is not true for AutoStore and other multi-deep storage systems. The performance of these systems is very sensitive to the ABC distribution: with a steep ABC distribution, fast-movers will “naturally” float to the top of the storage cube as they are accessed more frequently, while slow-movers sink to the ground and only infrequently have to be dug up. With a flat ABC distribution – or if only the long-tail of the distribution is stored in the system – there are no clear A-movers any longer, and thus probability that any particular bin is needed is almost evenly spread across the SKU range stored in the AS/RS. This implies that frequent reshuffling of storage bins will be needed: the flatter the ABC distribution, the more AutoStore and other multi-deep AS/RS will need to dig to reach the right storage bin. Therefore, multi-deep storage systems should only be used with a steep ABC distribution of picking demand; the flatter the ABC becomes, the more important it is to have direct and immediate access to all storage bins. Therefore, multi-deep AS/RS are not suitable for combination with zone picking.
In this article, we investigated why it can make sense to combine GtP systems with zone picking systems in high-performance settings. We found that by leveraging the ABC distribution and moving the fast-moving products from the GtP system to the zone picking system, we can reduce system size and CAPEX requirements quit significantly.
Also, we discussed aspects of order release and control logic. For order structures with large orders, combining two picking systems will make system control more complicated and it will require buffering and sortation capabilities between the two systems to make it work. Also, pick station setup is a crucial parameter to ease sequencing requirements in the GtP system if orders are started in the zone picking system.
The important caveat is that the AS/RS of choice needs to be insensitive to a flat ABC distribution of product demand, because that’s what’s remaining for the AS/RS once you bring the fast-movers into the zone picking system. AS/RS are less sensitive to ABC distribution to the degree they have direct access to all storage bins. Accordingly, multi-deep storage systems like AutoStore are highly sensitive, whereas single-deep or double-steep AS/RS will be able to cope with the requirements well.
So, when the AS/RS of choice in your GtP system is a miniload or a shuttle system, you can benefit from combining the GtP system with a zone with a zone picking system. You can save significant CAPEX and improve productivity. With AutoStore, however, you should maintain the full product assortment in the storage cube and not take out fast-moves, since otherwise the performance of the AS/RS will drastically decrease, possibly to the point where the GtP system becomes useless. This means multi-deep systems like AutoStore are not suitable for combination with zone picking.
1This is one of our eleven design principles for planning of automated logistics systems.
2 A good order release algorithm for zone picking systems requires consideration of a bunch of parameters and is quite complicated. Few companies have really figured it out, and none of the dominant warehouse automation companies is among them.