Understanding the impact of Amazon Go innovations

I have held back from adding to the noise around Amazon Go for several reasons.

• First, the zeitgeist around anything Amazon does is such that most analysis appears (to me anyway) as fawning at best and a guaranteed spreading of FUD in the minds of other retailers. That's justified on occasion but this time I say a guarded 'no.'
• Second, the introduction of cashierless stories is nothing radical at all but a logical development from what we've seen in the past.
• Third, some arguments believe Amazon is using the cashierless element as a way of sucking out marginal cost in the retail environment. I disagree with that argument.

I'll take each in turn.

The Amazon Go customer experience

There is no question that the Amazon store experience is far removed from that of 99.999% of other retail environments. I experienced that first hand at the Amazon Books store in San Diego in 2016. The (then) checkout experience was much smoother than any I had seen in the past. The store itself had the feel of a Waterstones with its well laid out and clearly marked sections and, dare I say, serene atmosphere. Amazon Go is a logical next step. Stores would do well to learn those lessons, and especially the convenience of just walking out. But equally, I get that is not something that's universally possible, or, even desirable.

Cashierless means more profit

Much attention has been devoted to the cashierless checkout experience. I particularly enjoyed the description given by one person who tried to fool Amazon's systems. But then CNBC tech reporter Deirdre Bosa got in on the act, claiming to have accidentally shoplifted some yogurt. Amazon seems to have taken it in good heart but then I'm sure there will be tweaks to the system to prevent a repeat performance. Why?

According to the NRF, inventory 'shrinkage' amounts to 1.44% or close to $50 billion annually. That number is comprised of:

• Shoplifting/external (including ORC) = 36.5 percent
• Employee theft/internal = 30 percent
• Administrative and paperwork error = 21.3 percent
• Vendor fraud or error = 5.4 percent
• Unknown loss = 6.8 percent

Assuming Amazon systems are capable of largely eliminating shoplifting and employee theft, then that represents around 0.96%. A small number but huge when weighed against industry average net profit for convenience stores that come in at 1.8%.

Quite rightly, most have commented that the in-store experience is much improved from the elimination of lines and the presence of staff whose job is to help customers.

But is this a radical move? Not really. In the UK, it is routine to find self-checkout options. While these promise convenience and hold the promise of a cost-effective experience, there are too many occasions where the systems fail or don't understand what's presented to them. Jon Reed showcased one possible answer in his early report from NRF, but Amazon has gone one step further. Will the two co-exist? Absolutely.

The use case Reed discovered is far better suited to those retail environments where the customer checks out a variable amount of product as in one or two apples. From what I've seen of the Amazon Go environment, that is geared towards selection from shelves where the products can be easily lined up.

Will we see more Amazon Go examples beyond the flagship store in Seattle? That is unknown. Amazon has not made any announcements to suggest a nationwide rollout but it would surprise me if it didn't use the Seattle experience as a jump-off point for more stores. If you accept that working theory then will Amazon apply the 'Go' system to Whole Foods?

The cost question

My final issue comes in the argument Ben Thompson pushed about the elimination of cashiers as a marginal cost. From Wikipedia:

In economics, marginal cost is the change in the opportunity cost that arises when the quantity produced is incremented by one unit, that is, it is the cost of producing one more unit of a good.^[1] Intuitively, marginal cost at each level of production includes the cost of any additional inputs required to produce the next unit. At each level of production and time period being considered, marginal costs include all costs that vary with the level of production, whereas other costs that do not vary with production are considered fixed. For example, the marginal cost of producing an automobile will generally include the costs of labor and parts needed for the additional automobile and not the fixed costs of the factory that have already been incurred. In practice, marginal analysis is segregated into short and long-run cases, so that, over the long run, all costs (including fixed costs) become marginal.

Thompson says:

A cashier — and forgive the bloodless language for what is flesh and blood — is a marginal cost. That is, for a convenience store to sell one more item requires some amount of time on the part of a cashier, and that time costs the convenience store operator money. To sell 100 more items requires even more time — costs increase in line with revenue.

Fixed costs, on the other hand, have no relation to revenue. In the case of convenience stores, rent is a fixed cost; 7-11 has to pay its lease whether it serves 100 customers or serves 1,000 in any given month. Certainly the more it serves the better: that means the store is achieving more “leverage” on its fixed costs.

That argument is only partially correct and is something of a stretch.

I am a huge believer in marginal cost accounting because to me, it more accurately reflects the way things happen in the real world rather than the standard cost method.

I could get into a long diatribe around this but in essence, the lower the marginal cost of producing a good combined with fixed cost utilization can (but not always) predictably lead to higher profits. This is something Thompson understands up to a point in his exposition of the way Amazon invests and leverages data center technology (as one example.)

Here he says that by going cashierless, Amazon Go has used its understanding of technology to eliminate a marginal cost in much the same way robots take out labor costs elsewhere in the world of Amazon.

But in retail, it's much harder to apply the same principle with any degree of precision. And marignal costing depends on precision to be a useful measurement and decision making tool.

The problem of marginal cost in retail

This example - albeit old by Internet years - explains the problem very well. The store in question had long lines regardless of the time of day or day of the week. Here is the explanation which, in my mind, perfectly explains why cashier labor is not a marginal cost.

(The store manager) said that every store in the chain (there are more than 1,000) has a “payroll bucket” from which to pay cashiers and, if my memory serves me correctly, each store’s payroll bucket is based on the store’s square footage.

The problem with this store, she said, is that it takes in a lot of money per square foot because this is Manhattan, where square feet are scarce but where customers aren’t. In fact, she said, this new store had immediately become one of the most profitable stores in the chain. And yet she couldn’t shake free any more payroll money (yet, at least — she said she’d been trying hard) because of the existing square-footage/payroll formula. She said she wanted more money to not only hire more cashiers but also to build more registers.

(My emphasis added.)

In this case, the company considers cashier payroll to be a fixed cost, regardless of volume and, like every other store, has a fixed number of cash desks.

If you consider this case more fully then you'll see that as a fixed cost, they are operating at 100% efficiency (in the most simplified of cases) and, therefore, under economic theory, labor becomes a measurable variable/marginal cost. This is especially true when someone leaves the line because in that case, the marginal cost of serving them cannot be met.

The only way that any store can flex its cash desk payroll is by assuming footfall patterns and then adjusting the rotas accordingly. Even that is the best guess because daily and hourly traffic is affected by a multitude of factors. Retail stores err on the side of caution, hence the lines. On the other hand, they have to continue paying cashiers, even if they're sat around doing nothing.

That latter argument is less true these days as cashiers often fulfill multiple functions. The point remains the same. They are not a marginal cost of delivery but a fixed cost that can be moved up or down, but only in steps, rather than at the individual output level.

You see that all the time when cashdesks are unmanned and then supervisors wait until lines are long enough to staff more cash desks - assuming the staff are available to do that.

Curiously, I think that alongside Amazon endeavoring to improve the customer experience they have also eliminated or transferred the cashier cost. In that sense, they have pulled off a remarkable feat by reducing the (near) fixed cost of payroll associated with cashiers.

It is unclear the extent to which cashier costs have been transferred to other activities like helping customers. Commenters noted that there is adult supervision over the liquor and beer/wine counters. Another fixed, but in this case necessary cost it might be possible to eliminate as and when Amazon can enrich our personal data to the point of including the verifiable birthdate.

Those same commenters noted that there are folk engaged in making sandwiches and the like. That is a quasi-fixed cost too, although it is possible to model the associated labor cost as a marginal cost.

Thompson thinks that longer term, Amazon will leverage its various investments to entirely roboticize the retail experience such that shelf stackers and food preparation will be machine driven. That is eminently possible and would certainly allow Amazon to reduce operational costs to near zero. But it won't do that if it degrades the customer experience so don't expect to see those floor walkers disappear any time soon.

But before it gets that far, Amazon might want to check out the way changes in display stacking mandates at its Whole Foods stores is making quite the mess...all in the name of supply chain optimization via its Order to Shelf system. Hmm...

My take

Substituting cashiers for check-in/out gates is clearly an innovation but it is sustaining rather than especially disrupting while improving the customer experience. That's unusual but not without precedent.

The labor cost impact taken with the likely inventory shrinkage improvements though are important and will encourage other retailers to consider how they too can use technology to achieve a similar outcome.

The only question is whether we, as consumers, will welcome such changes. After all, there is a certain pleasure to be drawn from idly chatting to a cashier. Even if it is at the lcoal convenience store. No robot can replace that - yet.