Growth Potential of Amazon Go stores is estimated at only $48M annually should it open downtown grocery stores in the most populated U.S. cities. A drop in the bucket. Yet it could gain $850M+ in annual recurring revenue by selling its no checkout and in store analytics tech to just 5% of U.S. grocery stores and pharmacies. Success is not guaranteed. Amazon needs to convince both shoppers and competing retailers to trust it with their data, as well as entice top employees to stay amid high turnover rates. Or, Amazon may not care about revenue at all and simply see Go stores as a retention program to keep us thinking “Amazon” at lunch.
Threatens self-checkout kiosks and smartphone checkout apps, while putting to shame existing Wifi-based in store analytics solutions.
Opportunity for Google to leverage their expertise in computer vision and artificial intelligence to extend Google Analytics into the offline world. Existing in-store analytics startups need heavy investment in AI and develop integrations with order management systems to catch-up.
Amazon Go opened earlier this year to the public. It promises a frustration free shopping experience with no lines, no cashiers and no checkout. Just walk right out with your bananas.
We investigate in detail: 1) problems of long wait times at checkout and incomplete omnichannel data for retailers, 2) competing solutions such as self-checkout kiosks and wifi-based in store analytics, 3) potential impact on revenue by opening more Go stores or by selling its tech, and 4) important risks facing the success of Amazon Go. Enjoy.
Waiting sucks, especially at lunch downtown
Much research has been done on the subject of waiting. Psychologists observed that waiting causes us much added stress and anxiety as we feel uncertain about the future . That’s why we’re always on the lookout for the next bus even though it won’t come any quicker. Having clear expectations calms us.
When it comes to shopping, we especially hate waiting when :
- It’s lunch time, as opposed to slack/break time, evenings or weekends
- We’re at work downtown, as opposed to being in the suburbs
To save time (and appear busy to our bosses), many of us have migrated to online shopping during work. Yet buying grocery online remains aloof. We still prefer to see what we’re buying and take possession of our groceries right away. So only 1% to 3% of groceries are bought online.
That’s why Amazon Go’s no checkout could be so disruptive. It brings the convenience and sense of control over our time we’re used to when shopping online to physical stores.
Let’s explore existing checkout solutions and see how they stack up.
Please wait for a store associate
Self-checkout machines have been around for years. Fujitsu, NCR, Pan-Oston, Toshiba TEC, and Wincor Nixdorf International are top vendors, each charging $30,000+ per machine. Relatively inexpensive compared to a human cashier who needs salary and benefits year after year.
Interestingly, stores don’t expect to save on labor costs with self-checkout kiosks. Replaced cashiers are given other duties around the store. Machines are actually installed for the same reason Amazon is building its No Checkout technology: To provide a faster checkout alternative.
Does it work? Depends what you’re buying. Researchers found that customers will use self-checkout kiosks if the queue seems shorter compared to regular checkout . Customers simply want the fastest way out. And that’s not always the self-checkout lane. According to the same researchers, most customers who use self-checkout lanes only have a handful of items to buy. More than 60% of shoppers who self-checkout don’t use a shopping cart or basket. They also don’t buy fruits and vegetables. This suggests we still perceive cashiers as more efficient when it comes to buying a lot of groceries or when we have items that are complicated to identify and weigh.
“Unrecognized item in the bagging area.”
“Please remove item from the bagging area.”
“Please wait for a store associate.”
Self-checkout machines are some of the least functional machines. Between ATM machines, gas station pumps and airport check-in kiosks, self-checkout kiosks were overwhelmingly rated as the most troublesome to use in an academic survey . Over 43% of respondents disliked their frustrating design. It really makes us wonder whether the engineers behind those machines ever used one.
Users’ unwieldy experience with self-checkout kiosks is limiting its adoption. Researchers have observed less tech-savvy customers shy away from them . That tends to be older shoppers, lower-income populations and female customers. These are important customer segments for grocers. Boomers also prefer to interact with people rather than machines. They perceive self-checkout kiosks as entirely benefiting corporations rather than customers. Boomers have a point. Marketing researchers have recommended stores to train customers to “familiarize” them with the technology. Effectively pushing the problem onto customers and making us cashiers.
Ikea, Albertsons and CVS have actually removed self-checkout kiosks in a bid to improve customer experience. Other retailers should take note. It’s been observed that customers blame the retailer when they face problems, not the machine’s manufacturer . Especially when we’re waiting for help from an associate that’s nowhere to be seen. The supposed quicker experience suddenly becomes the most frustrating as customers feel cheated out of their expectation. It’s not fast and the service is incompetent. Not much different than going to the post office! But unlike the post office, we have choices on where to buy bananas.
There’s an app for that
In line with making checkout the customer’s problem, Wal-mart and Kroger offer apps to let customers scan products and checkout on their phones. An employee then audits purchases before customers leave the store.
Wal-mart touts its app’s ability to help us track spending. Interestingly, that may be the last thing retailers want us to do. We typically spend more in store than online as we don’t know our order total until checkout. We also make more impulse buys as we get distracted by fancy layouts and product discounts. Plus, they make us walk through so many aisles just to get a jug of milk. An app reminding us of our bill may actually reduce impulse buying and lower average order value.
The checkout apps are relatively OK rated in both Android and Apple app stores, scoring around 4.2/5, yet are far from perfect. Users report having issues scanning items, paying with gift cards, incorrect pricing, and failed login. More worrisome are security issues. Some report having their credit card info stolen after using the app. Success with the app has also backfired. Some users report wait lines at the exit door as associates struggle to quickly audit people’s purchases. Replacing one line with another one.
Here’s a sample negative review:
The customer above wasted precious time for forgetting to scan a $.98 can of soup. By comparison, Amazon Go couldn’t care less what you accidentally shoplifted. Who do you think creates more happy return customers?
These technical issues can surely be solved over time, but a quick search on LinkedIn shows there’s currently only 8 people working on Wal-mart’s Scan & Go app, compared with Amazon Go’s 50+ staff. There may lie the difference between Wal-mart and Amazon: One’s a retailer who uses technology, and the other a technology company who happens to sell stuff.
Throw people at the problem.
Researchers have found that customers tend to focus on the length of a line, rather than speed, when choosing checkout lanes . Perhaps because it’s quicker to assess length than speed.
Retailers could therefore open more lanes and employ more cashiers. This would certainly make some boomers happy as they seek human interactions. Retailers don’t seem to be opposed to the idea. There’s actually more grocery store cashiers today than a decade ago, 867,920 in 2016 compared to 824,050 in 2004.
Comparing solutions to cashiers
Closing the loop on omnichannel analytics
While retailers can track and analyze our online shopping behavior, they know very little about our in store purchases. That’s because we don’t login to shop in store. It’s in turn difficult for retailers to match our purchases to our profiles. In the end, they can’t account for our offline lifetime value, repeat purchase rate, nor predict what items we’re likely to buy in the future.
There lies the beauty of Amazon Go. It forces each customer to login before shopping. Effectively closing the loop on a user’s online and offline purchase data. Complete omnichannel analytics.
Amazon is however not alone. It joins a rare club of retailers that force customers to identify themselves, including Costco and Sam’s Club. All tout some sort of value-added memberships in return.
It’s not just what we buy. It’s how we buy.
Amazon Go will however know much more about our in store behavior than Costco ever has. The store uses a fleet of cameras and sensors to track our in store movements and purchases. These same cameras will allow Amazon to see and study how we make purchase decisions.
A group of researchers recently built software to differentiate between a search task and a decision making task, by analyzing video of people’s eye movements . You can bet Amazon is thinking of commercializing this type of research to maximize sales at stores.
Other marketing researchers have explored how to analyze in store video footage. Analyses include where customers go, percentage of store visited, paths followed, interactions with shelves, walking speed, and time between product selections . Armed with these insights, retailers will be able to tweak store layouts, product placements, signage, pricing and other stimuli to increase sales: Facilitate our decision making and prompt impulse buys.
It’s been demonstrated that a grocery chain’s profitability could increase almost 40% if each customer purchased an additional item on impulse . And what store element has the most impact on impulse buying? Layout! Compared to employee friendliness, music and even lighting, store layout has the greater impact on impulse buys. Amazon Go’s video insights will undoubtedly help.
Retailers will ideally be able to spot behavior patterns that can help tailor shopping experiences to different types of crowds. A store’s design could change at different times of the day depending on who’s visiting. Similar to how we get unique product suggestions on Amazon.com. Lunch crowds could get more ready-made foods, while afternoon shoppers could see more coffee and snack options.
If successful, this type of customer tailored shopping could disrupt grocery economics. Amazon Go can offer exactly what customers want in a smaller space, rather than stock everything we could possibly want in a gigantic warehouse.
Grocery stores are in desperate need of Amazon Go’s technology. They have little customer data available compared to e-commerce stores, and are one of the only retail sectors growing their brick and mortar presence. With Amazon Go, they will finally be able to track individual purchasing behavior at scale.
Not the only Big Brother
Matching in store behavior data to customer profiles remains a challenge for most retailers.
Apps such as Wal-mart’s Scan & Go can effectively track a customer’s in store purchases, yet doesn’t offer any insights as to how the purchase decisions were made, or what alternate products were considered. It doesn’t visually track the customer throughout the store. It’s also optional.
Loyalty programs requiring customers to swipe a card at checkout or asking for email/phone numbers face similar limitations.
The most promising in store analytics solutions involve movement tracking via smartphones. Startups such as Euclid, retailNext and Locately track customer movements in store using Wifi and GPS technologies, sometimes in combination with cameras. A major limitation of these systems compared to Amazon Go lie in the fact they cannot identify specific customers and track their purchases. This leaves a critical gap in omnichannel sales data at the customer level.
Solving the identification problem may however only be a matter of time. Researchers have already made big advances in leveraging wifi signals to identify individuals . Others have used cameras for facial recognition. Even Amazon has patented a facial recognition check-in process.
Caught off guard are the current manufacturers of self-checkout kiosks. They may very well be disrupted by the current wave of in store analytics startups. Especially if the latter group find a way to reproduce Amazon Go’s no checkout technology.
Build stores or sell the tech?
Amazon hasn’t announced how it plans to grow Amazon Go, but that doesn’t stop us from exploring some scenarios.
They’ll need thousands of Go stores to impact revenue
Amazon Go is likely to open more stores. Its career page is seeking a Senior Real Estate Manager and Construction Manager among other positions at the time of writing. Part of their responsibilities are to select new sites and kickoff new store projects.
Where will the stores be? Amazon Go’s no checkout will appeal most to those who can’t afford to wait: Busy professionals downtown. Perhaps to pick up lunch as well as their latest packages from Amazon Lockers. Its current test store in Seattle seems ideally sized for the task, with 1,800 square feet of ready-to-go food and some popular grocery items.
To work out how many Amazon Go stores will be built, let’s assume Amazon Go stores will open in the downtown area of all U.S. cities with at least 1M+ people, breaking ground wherever they find a CVS pharmacy. Pharmacies are just as strategically located in downtown areas, providing convenience to many of the same customers Amazon Go targets. I’ve therefore counted the number of CVS pharmacies in downtown areas from Austin to New York:
|Population||No. CVS Pharmacies Downtown|
Source: Wikipedia and Google Maps
In total, we find 53 downtown CVS pharmacies in cities with population over 1M people.
To estimate revenue, let’s use the grocery store average sales of $507 per sq. ft. annually. It works out to around $900K per store annually. That’s a conservative estimate by all accounts considering it will likely attract much more foot traffic than the average grocery store. Big brother Whole Foods makes $937 per sq. ft. and Trader Joe’s leads the pack with $1,723 per sq. ft.
Sticking to our estimates, Amazon Go could generate annual revenue of at least $48M by opening 53 stores in the downtown core of our most populous cities. A drop in the bucket relative to Amazon’s total revenue. They’ll need to open thousands more stores to impact the bottom line.
Our math suggests that Amazon’s goal may not be to boost revenue through Go stores, but perhaps keep you within their sphere of influence wherever you may be. Never escaping the Amazon. Lunch is afterall one of the last moments we don’t think of Amazon first. But if we can pick up our packages and grab a quick bite before that 1pm meeting, why not?
Amazon Go in a grocery store near you
Amazon should rather sell its No Checkout technology to other retailers to increase sales. Should it manage to convince competitors to share their data.
There are currently about 38.441 grocery stores in the U.S (this number is growing). Of those, let’s discount the 5,412 Wal-mart stores and 2,790 Kroger stores that are unlikely to ever partner with Amazon. That still leaves us with more than 30,000 grocery stores Amazon can sell its tech to.
Pharmacies may also be great customers. They sell convenience and quick-bites. Between CVS, Walgreens and Rite-Aid, we count more than 22,500 stores in the U.S.
Let’s now assume only 5% of U.S. grocery stores and pharmacies will implement Amazon’s Go’s technology in an initial wave. That gives us about 2,600 stores.
Compared to self-checkout solutions, Amazon Go offers way more in terms of convenience to customer and in store analytics. It can therefore charge much more than their $125,000 installation fee per store. However, let’s be conservative and assume it offers its technology at the same price to compete aggressively. Amazon could then reap in sales of $325M by installing its technology in only 5% of grocery stores and pharmacies in the U.S.
What also make more sense is to charge a transaction fee on top of device installation. Amazon Go’s checkout technology works like a payment processor after all. Existing payment processors such as Cayan, Square or Stripe charge between 2% to 2.9% as transaction fee. Let’s again be conservative and assume Amazon Go will only charge 1.5% as fees to compete aggressively.
How much sales do the 5% of grocery stores and pharmacies do? The grocery store industry is valued at about $600B annually. Wal-mart and Kroger claim 17.3% and 8.9% of that market respectively. That leaves $442.8B for less tech-savvy grocers that are more likely to partner with Amazon. On the pharmacy side, Rite-aid, Walgreens and CVS made $26.8B, $87.3B and $81.1B in retail sales respectively. That’s over $195B. Amazon could process over $638B in total. Charging a fee of 1.5% to 5% of that sum translates into revenue of $478M.
Let’s then not forget Amazon Go’s unique value add to retailers: customer level in store analytics. Startups such as RetailNext charge $2,000 a year for access to heat maps and other in store insights. Amazon could charge much more for effectively closing the loop on omnichannel analytics. A better comparison may be Google Analytics 360 (Premium) offered at $150,000 per year not including support plans. Amazon may even convince retailers to fork that sum for each of their stores, accounting each store as a separate business or storefront. Retailers may have no choice but to accept as there are no alternatives to their very painful omnichannel analytics problem. Amazon Go’s insights are invaluable to help chains tailor each store to their unique customer base. Should that pricing model be adopted, Amazon stands to reap in another $390M in annual recurring revenue for implementing the service at 2,600 stores (5% of total U.S. groceries and pharmacies).
Amazon Go could therefore generate an annual recurring revenue of over $868M and a one-time installation fee of $325M if offered as a service to just 5% of U.S. groceries and pharmacies. Not a bad start. It would help increase Amazon’s B2B AWS revenue ($17.46B in 2017) by over 6.8%.
No lines, no checkout. What’s in the way?
Uh oh, I just stole a yogurt.
Amazon Go started building its no checkout system four years ago. The technology is however still in its infancy. Unpredictable children and people with similar body types are known to confuse its artificial intelligence. Other researchers working on retail video analysis have had trouble with large crowds and differentiate customers from employees . No wonder Amazon Go was only open to employees for more than a year.
Unreliable technology results in lost revenue and free yogurts for customers who escape the radar. A serious problem as grocers operate at profit margins of 1% to 2%.
Existing self-checkout machines are already known to cause loss rate of 4% as customers unintentionally steal produce or input wrong information. More than double the average loss rate of grocery stores with cashiers. Amazon Go will have to do better to be cash positive.
Amazon is investing much resources to advance its technology. A quick search on LinkedIn reveals at least 52 people actively working on Amazon Go: 23 on software, 7 on computer vision research, 6 on hardware, and 10 on operations. And they’re looking to grow that team by more than 50%. Amazon’s career page has 36 openings on their Amazon Go team at the time of writing: 11 software engineers, 4 computer vision researchers, 4 hardware engineers, and 12 operational roles. Some of these openings may also be permanent as it seeks to attract as many experts in computer vision and artificial intelligence as possible.
It’s only a matter of time before Amazon addresses all major technical issues that no checkout faces. It has both great minds at work and the power of AWS to help process the gigantic video datasets.
It may also not care about profits or getting six sigma precision to its no checkout process. Amazon Go may be accounted for as another retention tool to keep their brand top of mind at lunch. Just like Prime Video.
Store closing! Everything on Sale (including the lease)
To find downtown real estate, Amazon go will benefit from the ongoing retail apocalypse. Over 6,700 stores closed in the first three quarters of 2017 alone. It’s a great opportunity to take over leases of closing stores at discount.
Amazon may also choose to partner with another retailer and establish Go stores within their stores, as it did with Kohls. Macy’s may fit the profile. Amazon Go could access top notch downtown real estate, which is worth more than Macy’s market value itself, while Macy’s gets more foot traffic.
My milkshake tastes better than yours
Mr. Bezos has had a string of successes in disrupting industries. It first crushed Barnes & Noble. It now dominates the smart speaker market after beating Apple and Google to the punch. It’s now threatening Facebook and Google’s golden goose: advertising revenue.
Breaking into groceries has however been challenging. Before buying Whole Foods, it commanded only 0.19% of the market (Whole Foods adds another 1.21%). Wal-mart and Kroger dominate the market as customers still prefer to visit physical stores to get bananas and milk.
In trying to disrupt the industry, Amazon offers a confusing array of grocery options. A search for Milano cookies on Amazon.com yields:
- 36 packs of 1oz bags for $36 with Amazon Prime
- 7.5oz bag of cookies for $3.36 with Amazon Pantry
- 7.5oz bag of cookies for $3.29 with Amazon Fresh
These initiatives are increasingly sources of internal tension at the firm. They compete to sell the same thing. There will come a point where some initiatives will be shut-down, just as it scaled back Amazon Fresh.
Amazon Go’s growth could therefore be limited by competing investments in other grocery projects. Politics and data storytelling may decide its fate.
Help wanted: Behavioral Analysts
Beyond engineering challenges, Amazon and retailers will need a small army of consumer behavioral analysts to sift through customer browsing footage and help design store layouts. They’ll play a critical role in facilitating customer decision making and maximizing impulse buying. It will be challenging.
As described by behavioral analysts themselves, the work is less data-crunching and focuses rather on observing how customers interact with products and make decisions . This means retailers can’t simply change responsibilities on their army of data scientists. Human skills are required, and PHD graduates tend to lack them.
Behavioral analysts with backgrounds in consumer marketing will be hard to find. Behavioral analysis education programs tend to focus on developmental disabilities . Over 60% of articles in Journal of Applied Behavior Analysis focus on this area . This means most behavioral analysts are trained to help people with autism, not retailers. The situation may slowly change as retailers offer better compensation and schools look to please employers.
Behavioral analysts are also rare regardless of specialty. A search on LInkedIn reveals around 6,200 people with a “behavior analyst” title, compared with over 62,000 “data scientists”. None of them were employed by Amazon or Wal-mart, a sign retailers haven’t yet recognized their value. Perhaps too focused on getting the technology right. The top industries employing behavioral analysts according to LinkedIn are related to healthcare (see below screenshot). Again confirming the fact finding behavioral analysts with marketing backgrounds will be difficult.
Please agree to our terms of service and be watched by 101 cameras
Americans value their privacy. 90% believe it to be important to control what information is collected.
Having hidden cameras monitor our every move may therefore be unsettling. Amazon customer service reps already have access to our names, home address and purchase history. Imagine if they could also access footage of us browsing their store. Slightly creepy. It’s real enough of an issue the FTC has been policing existing in store analytics providers.
Retailers may argue the technology is no different than how their websites track user clicks, but I personally feel there is something different when it comes to being physically watched. Especially without active consent. And without knowing who might be watching me.
Privacy issues could certainly bring a backlash against retailers using in store tracking systems. Uber can attest to that. It would be very hard to win people’s trust back.
No, I am not likely to recommend a friend to work at Amazon
Sifting through detailed job reviews reveals that research scientists (a critical role for Amazon Go) love the impact their projects have, the potential to disrupt industries, and the sheer amount of data available to play with. Yet frustrations abound. Employees complain about underpar research facilities and a frugal culture where they have to find their own ways to access research journals. In-fighting and politics often distract employees. Researchers also complain of being treated like software engineers and asked to deliver results quickly. Perhaps shorting their ability to innovate or to pinpoint root cause of problems. More interestingly, one has to wait hours to run simple SQL queries on their database.
Always having fresh blood may actually work to its benefit. Scientists have observed that newcomers tend to be more optimistic and innovative than old-timers . Constant challenge from new employees may also keep old-timers on their toes. Always challenging the status quo. Just as Bezos likes it.
That hiring system may work when it comes to software engineers, who graduate by the tens of thousands, but not so much for top research scientists in computer vision and artificial intelligence. Such talent is in great demand by self-driving car companies such as Tesla and Waymo.
Amazon should hold on dearly to its experts in computer vision. Google has dozens of openings for such talent. While these scientists may currently be hired to work on Pixel phone’s camera or Waymo, it’s only a matter of time before Sundar Pichai realizes an opportunity to extend Google Analytics into the physical space and merge it with Google Wallet to compete head on with Amazon Go.
Should that be the case, retailers would almost certainly prefer to purchase from Google. It’s a partner they trust. Amazon is a ruthless competitor.
 Edgar Elías Osuna, The psychological cost of waiting, Journal of Mathematical Psychology, Volume 29, Issue 1, 1985, Pages 82-105, ISSN 0022-2496,
 Davis, M. M., & Voilmann, T. E. (1990). A framework for relating waiting time and customer satisfaction in a service operation. Journal of Services Marketing, 4(1), 61-69. 10.1108/EUM0000000002506
 Wang, C., Harris, J., Patterson, P.G. Customer choice of self-service technology: The roles of situational influences and past experience (2012) Journal of Service Management, 23 (1), pp. 54-78.
 Dean, D.H. Shopper age and the use of self-service technologies (2008) Managing Service Quality, 18 (3), pp. 225-238.
 Lee, H.-J., Cho, H.J., Xu, W., Fairhurst, A. The influence of consumer traits and demographics on intention to use retail self-service checkouts (2010) Marketing Intelligence and Planning, 28 (1), pp. 46-58.
 Lee, H.-J., Fairhurst, A.E., Lee, M.-Y. The importance of self-service kiosks in developing consumers’ retail patronage intentions (2009) Managing Service Quality, 19 (6), pp. 687-701.
 Yina Lu, Andrés Musalem, Marcelo Olivares, Ariel Schilkrut, (2013) Measuring the Effect of Queues on Customer Purchases. Management Science 59(8):1743-1763.
 Gidlöf, K., Wallin, A., Dewhurst, R., Holmqvist, K. Using eye tracking to trace a cognitive process: Gaze behaviour during decision making in a natural environment (2013) Journal of Eye Movement Research, 6 (1)
 Larsen, N.M., Sigurdsson, V., Breivik, J. The Use of Observational Technology to Study In-Store Behavior: Consumer Choice, Video Surveillance, and Retail Analytics (2017) Behavior Analyst, 40 (2), pp. 343-371.
 Mohan, G., Sivakumaran, B., Sharma, P. Impact of store environment on impulse buying behavior (2013) European Journal of Marketing, 47 (10), pp. 1711-1732.
 Zhang, J., Wei, B., Hu, W., Kanhere, S.S. WiFi-ID: Human identification using WiFi signal (2016) Proceedings – 12th Annual International Conference on Distributed Computing in Sensor Systems, DCOSS 2016, art. no. 7536315, pp. 75-82.
 Hassan, E., Maurya, A.K. Real-time video analysis for retail stores (2015) Proceedings of SPIE – The International Society for Optical Engineering, 9443, art. no. 94432B
 Connell, J., Fan, Q., Gabbur, P., Haas, N., Pankanti, S., Trinh, H. Retail video analytics: An overview and survey (2013) Proceedings of SPIE – The International Society for Optical Engineering, 8663, art. no. 86630X
 Larsen, N.M., Sigurdsson, V., Breivik, J. The Use of Observational Technology to Study In-Store Behavior: Consumer Choice, Video Surveillance, and Retail Analytics (2017) Behavior Analyst, 40 (2), pp. 343-371.
 Woods, D. W., Miltenberger, R. G., & Carr, J. E. (2006). Introduction to the special section on clinical behavior analysis. Journal of Applied Behavior Analysis, 39(4), 407–411.
 Surprise and sense making: what newcomers experience in entering unfamiliar organizational settings. (1980) Administrative Science Quarterly, 25 (2), pp. 226-251. Cited 1186 times.