Finding the Right eGrocery Solution

Introduction

Finding the right eGrocery solution is a struggle for many retailers. eGrocery is a comparatively new industry in the logistics automation world. At the same time, it is the industry with the very highest requirements on the design of logistics systems. No dominant solution design has emerged, so different solutions – and not only solutions, but entire concepts, including selection of the right location and distribution concept (e.g., hub and spoke vs. direct delivery, transportation in frames vs. transportation on dolleys or pallets vs. floor loading) – can be found in practice. Whenever experts gather and discuss the matter, they will find themselves favoring different concepts. 

It turns out there is good reason to disagree. Different solutions fit to different circumstances better than others. What has been missing so far, however, is an overview when which solutions fits best. This article aims to shed a light on this question by relating different solutions to different order volumes the eGrocer wants to fulfill from one location. There can be other aspects, too, that must be taken into consideration for the most economic system design. However, relating system design to order volume is likely to approximate best the design choice with the best unit economics. After all, we try to improve unit economics by reducing variable cost without letting fixed cost escalate. Obviously, when venturing into automation, we accept a certain, sometimes fairly high, capital expenditure (CAPEX) in order to enable low operational expenditure (OPEX). As a rule of thumb, and if you’re doing it right, the more CAPEX you are willing to spend upfront, the lower your resulting OPEX – within boundaries, obviously, but if your higher CAPEX results in higher OPEX, all other things equal, there is no discussion you are doing it wrong. Since Full Time Equivalents (FTE) are a main factor in variable cost and thus in unit economics, operationally you need to make sure that all FTE are utilized as much as possible with value-adding (!) activities. In all successful automation projects with eGrocery companies, FTE productivity must be a leading consideration. 

There is a lot of (simple, yet comprehensive) calculations behind the arguments put forth. To maintain readability, detailed calculations are omitted here. The technically interested reader will be able to connect the dots very easily. The boundaries between the concepts are not clear cut: they can differ depending on order structure, location of the fulfillment center, SKU range, and more. The overview will give you a good indication, however, and will help you find the right eGrocery solution for your business.

Step 1: In-store Fulfillment 

If you start off with online orders and you have got a store network available, at order volumes of less than 20 orders per hour the only sensible option seems to be to pick straight from stores. You can add additional stores to increase the range of orders you can serve; serving more than 20 orders per hour from one single store however, likely begins to impact on your regular store customers’ shopping experience.  

In conversations with senior management of incumbent grocery chains I have heard them voice the opinion that they believe only in-store picking will allow profitable eGrocery service. Advantages certainly are that you can start off with practically zero overhead and that you can scale easily (or so it seems) by adding online fulfillment to additional stores. Services like Instacart are based on the supposition that in-store eGrocery fulfillment can be profitable business. There is good reason to doubt this: picking performance is poor due to store layouts being optimized for anything but logistics, professional “shoppers” (i.e., pickers) collide with regular customers, and there is no effective inventory management, so you will experience stock-outs for certain products more often than is appropriate. In fact, lack of effective inventory management is sometimes cited as the main reason why store picking can be nothing but an intermediate solution to get things going. So, counter to the wide-spread notion, it does not scale well.  

In-store fulfillment may be the most viable option to start with, provided stores exist, yet it is fair to assume that operations are not going to be profitable, ever, and that customers will not be happy due to frequent unavailability of items they ordered. If your sales volumes allow, you should think bigger and move ahead quickly. 

Step 2: Manual Darkstores 

Once you exceed roughly 20 orders per hour in a store (or in each of your multiple stores offering the service), the next stage for eGrocery fulfillment is a darkstore, a small warehouse set up like a supermarket but dedicated to professional shoppers who pick for online orders. The huge advantage of a darkstore compared to regular stores is that you can optimize the darkstore for logistics operations. SKU slotting – the allocation of products to shelves – can follow logistical considerations, such as the ABC distribution of sales, rather than the prerogative of marketing. Aisles can be setup such that routes through the darkstore are less time-consuming and short-cuts can be taken. Also, no customer will mess with your stock management – one of the most obvious problems with in-store picking, since you never know if someone just bought the very last avocado just 1 minute ago when you receive an online order that includes avocados. Pickers can pick without consideration of regular customers whose shopping experience could otherwise suffer. 

At 80 orders per hour, you have roughly 6 to 10 pickers rushing through the aisles of the darkstore with their pick carts, each equipped with multiple (between 8 and 14) orders totes which they pick to simultaneously (multi-order picking). This is manageable, even when replenishment takes place simultaneously in the same aisles. You may want to mirror the fastest-moving SKUs in order to avoid congestion in picking aisles. Chaotic storage reduces the chance of pickers blocking each other and facilitates replenishment, but reduces the pick rate; it is really only a good option if space and thus congestion are a big issue, but it certainly is no productivity hack.

Step 3: Darkstore with Partial Automation (Stage I) 

At 150 orders per hour in a manually run darkstore, you have got about 15 people pushing carts which now may be a little larger (e.g. 12 – 16 totes), assuming a pick rate of 250 picks/h, which experience shows is doable for darkstores not exceeding 8.000 SKUs. Things have begun to become messy, however. 150 orders per hour means something like 300 to 500 order totes per hour which need to leave the warehouse in an organized way. 

The logical next step is partial automation of the darkstore. This is a huge topic in and by itself and I will only scratch the surface here. Several degrees of automation are conceivable for the darkstore, pertaining to the different functions in the warehouse, such as transportation, replenishment, picking, sortation, buffering, and stacking. There is a fine line between partial automation of a darkstore and building a full-blown highly automated DC; in fact, there is no clear demarcation and it is mostly a question of scale. 

Due to the nature of requirements on eGrocery systems, in particular the volatility of demand (each day of the week shows different demand patterns, so does each hour of the day) and the distribution of goods across different temperature zones, both of which induces a need for buffering, sortation, and consolidation, shipping buffer solutions have long been favored as first step into automation with an otherwise manually run darkstore. There is a number of eGrocery companies employing this model. The shipping buffer allows all other warehouse functions to run smoother and more efficiently. While it is an enabler for better, more efficient operations, when you try to assess it from an ROI perspective, it is not obvious how it pays off since the numbers supporting the business case are not easily measurable. From a certain level of throughput onwards, however, it simply becomes a necessity as manual sortation and consolidation simply requires too much floor space and manpower, and picking aisles congest when order fulfillment is not spread out more evenly across working hours. 

Getting back to the example above, when you manage 120 – 150 orders per hour, this seems like a good point in time to think about a shipping buffer. Sorting and consolidating 300 to 500 totes per hour in outbound manually is a hassle. And don’t forget that eGrocery systems are highly volatile in terms of demand, so you are likely to experience peaks which are much higher. These peaks tend to come right before important days (national holidays, etc.), and this is when new customers try you out and you get the chance to satisfy or disappoint them once and for all. 

Picking and replenishment, too, become an issue. Until so far, pickers and replenishers could share the aisles and operate simultaneously. You could work with regular shelves, such as those used in supermarkets. At 120 – 150 orders per hour, you should physically separate these two operations. When an average grocery colli holds roughly 10 to 12 individual grocery items, your replenishment flow will be roughly 1/10 to 1/12 of your picking flow, if you replenish in collies. If you replenish in individual pieces (which allows faster picking, yet reduces replenishment performance significantly, possibly leaving you with a net loss in productivity), replenishment traffic is even higher. Colli flow racks become your go-to rack solution (yet force you to move away from piece-wise replenishment). Due to their inherent buffer functionality, they also allow better decoupling of picking and replenishment, hence allowing better scheduling of replenishment and, accordingly, better productivity. The drawback clearly is that you need more footprint, though this may simply be the price you need to pay in order to be able to serve more customers. If you replenished piece-wise before, note that the task of disposing of trash, such as cartons and plastic wrap, now mostly (yet not entirely) will have to be performed by the picker, so you have to have a trash concept in place on both sides of the racks. 

Step 4: Darkstore with Partial Automation (Stage II) 

Picking could need some support, too. Until so far, if you had good pickers (picking at a rate 250 pieces per hour), you could get by. Having even more pickers and maintaining picking efficiency in manual operations is hardly feasible, however. Semi-automated (i.e., automation supported) picking will be the next thing to implement.  

Rather than going for CAPEX-heavy goods-to-person systems, which are unlikely to pay off at this throughput and can only be justified by the need for a much higher SKU range, a well-designed zone picking system appears to be the best solution. Zone picking systems have been around forever, yet their potential has remained unseen and their implementation in almost every single instance I have seen was mediocre, if at all. In most existing zone picking systems, pickers do between 150 and 200 picks per hour. That’s not a lot – not more than could be done in a manual person-to-goods system with multi-order picking. With a well-designed zone picking system (let’s call it Zone Picking 2.0, or ZP 2.0, for clear distinction), picking rates of more than 350 picks per hour are easily achievable, and much more is possible. Also, maintenance is simple and cheap and literally everybody who can build conveyors can supply such a system.  

There are only two drawbacks of ZP 2.0 systems: (1) the range of SKUs they can hold is limited, so you would still have to have a manual area, and (2) – well, try to find somebody who knows how to design a zone picking system well.  (Other than us, that is).

As for the first point, with a combination of zone picking and a manual area, you will still get a highly efficient system as long as you stay below roughly 12.000 – 15.000 SKUs and you find a good way to integrate the two areas (there are good ways).  

As for the second point, eGrocery companies who run such systems protect it from visitors since they consider it an important competitive advantage – and they are right, in fact. Related to the second problem is the inability of literally all the incumbent logistics automation companies to provide a good piece of software that can run a zone picking system properly (see below for some more comments on software). 

Assuming overall picking productivity of 350 picks per hour, you will need roughly 10 people picking in your ZP 2.0 system in the beginning, each taking care of two picking stations (small, easily accessible stations with reasonably sized buffers allow one picker to operate two stations).1 As you increase your throughput, you will staff more and more stations individually whereas you can still run two (or more) stations per person in times of low demand.  

The maximum throughput in a ZP 2.0 systems is limited by the capacity of the conveyors. Due to the many stops and detours, capacity is limited at roughly 1.000 order totes throughput per hour in one temperature zone, as demonstrated by both, simulation and practice. Assuming two order totes per order in the Ambient temperature range, you will reach this limit at between 300 and 500 orders per hour (since many order totes will loop through the system at least once before they are completed, 300 orders per hour are more realistic, unless you tweak the system with some smart add-ons which are not discussed in this article). With 300 orders per hour, we can reach almost 5.000 orders in 16 hours of operations. Due to demand volatility, we are unlikely to process 300 orders every hour, but you get a sense of the capacity. Realistically, daily capacity is limited somewhere around 10.000 orders with such a system. 

If you want to reduce FTE in the ZP 2.0 system, picking robots are currently becoming an increasingly attractive option. Robots for stacking/palletizing and smart concepts for replenishment drive up CAPEX but can make a huge difference in terms of productivity. When using picking robots, you will have to allow for manual back up stations, so footprint is likely to increase. Since pickers are highly productive in a ZP 2.0 system, there are more important problems to solve before you replace pickers with robots, however. 

A very attractive add-on to a zone picking system is automation of replenishment. Replenishment flow is significant. Since most order totes hold roughly 10 – 15 items, as a rule of thumb you can remember you have to replenish about as many grocery collies as you produce order totes. If you can find a way to save FTE in this process, the investment may provide an attractive ROI. One way of doing this is by using a light-weight shuttle system – not to feed goods-to-person pick stations, but solely for replenishment of flow channels which are directly attached to the shuttle rack, and maybe as a shipping buffer. You can further tweak the system by filling certain channels dynamically with the SKUs needed for the next two hours or so, thereby enabling your system to manage many more SKUs than would be possible with permanent SKU slotting. 

Once you have shuttle aisles in place, you can add even more functionality by using a small share of their dynamic capacity for goods-to-person picking of the long-tail, thereby increasing the viable number of SKUs in the system even more. Things begin to get complex and expensive here, however. 

Note how we are slowly leaving the world of low CAPEX lightweight automated darkstores. This is the time to discuss large-scale highly automated FCs. 

Step 5: Large-scale Highly Automated Central FCs 

If you want to process more than 10.000 orders per day, large-scale goods-to-person (GtP) systems will enter the stage. While a high-performance ZP 2.0 system will cost you around €5M, large-scale goods-to-person systems for more than 10.000 orders per day will cost you at least €35-50M. Smart combinations of zone picking and GtP can reduce the cost somewhat, but the high throughput still makes the system very complex and expensive. There is reason to believe that you would be better off replicating the ZP 2.0 system rather than going for large-scale goods-to-person. 

Anything above 20.000 orders per day requires so much technical effort that it is likely to become technologically and economically unfeasible. You will hit several technical constraints, and the overall complexity of the solution will be a challenge to manage in daily business. 

A Note on Software 

When we talk about eGrocery fulfillment, whether manual or (partially) automated, we should not forget that software really makes a huge difference. Steel racks and conveyors are mostly simple, but software is not. You won’t possibly get anywhere close to profitability if your WMS software is not top-notch and flexible enough to grow with your business and to adjust to the ever-changing requirements in your warehouse. It is no coincident that some of the best-run eGrocery systems belong to software companies disguised as online supermarkets. They build their own software and they improve and change it all the time. This is nothing you can do with a piece of software from a third party that makes you dependent on that third party’s resource availability. Also, it would be going to cost a fortune and quality of outcome is unclear – most software vendors have little to no experience with well running food e-com systems, and this includes pretty much all the incumbent logistics automation providers. If there is a single most important technical capability for an eGrocery company to possess, it is software engineering. 

What about Microfulfillment Centers? 

In the above discussion, you may miss microfulfillment centers. There is much noise about them right now. In early 2020, even Amazon announced they would partner up with Dematic for microfulfillment centers. Fabric, another provider of microfulfillment solutions, raised $110M for further expansion. TakeOff (with their technology partner Knapp) signed a number of deals in the past year.  

Microfulfillment centers often aim at the range of 30 – 60 orders per hour with small-scale GtP concepts based on shuttle systems (typically 3 aisles) in a backroom of a larger regular store. The price tag is at about €3.5M; some companies (like TakeOff) charge a recurrent fee for software and service. The overall productivity of employees (as measured in UPE) of such a solution is at about the level of a well-run manual solution – while the cost is almost at about the level of a zone picking system which can do 10x the throughput, however.  

In economic terms, they are strictly dominated by the other solutions. I yet have to find strong arguments for microfulfillment solutions; the only benefit really seems to be their small footprint which allows placement closer to customers, hence lowering distribution cost. The low distribution cost is likely to be offset by higher fulfillment cost, however. Cost of such solutions must go down drastically before they cease to be strictly dominated by the other solutions discussed above, which seems unlikely to happen anytime soon. The problems of microfulfillment centers are systemic and mostly independent of the vendor. A detailed discussion of microfulfillment centers and why they are (clearly) unfit for profitable online fulfillment will follow in another article. Some of the challenges MFCs based on conventional shuttle systems face are discussed in this article.  

Do you run eGrocery operations and want to discuss solution design? Reach out, we are most likely able to help you make your operations more productive.

 

This Post Has 2 Comments

Leave a Reply