Feature
The Gist
- Technology reversal. Amazon is removing its heralded Just Walk Out Technology, an in-store automated checkout feature that allows customers to purchase items without a cashier or checkout lines.
- Human intervention. Amazon removed it after reports emerged that it used remote workers rather than AI to confirm scanned goods.
- Ethics examination. The incident is an AI usage case study for managing AIOps and ethics beyond an AI system.
Brands often run into snags when implementing innovation technology, especially at launch. The snag that forced a decision to pull the Amazon Just Walk Out technology came two years after the program began. This move offers numerous lessons for planning AIOps for customer experience.
Let’s take a deep dive into what happened with the Amazon Just Walk out technology.
A Look at the Amazon Just Walk Out AI and Amazon Go
Amazon announced Amazon Go, a cashierless corner store concept, featuring the Amazon Just Walk Out tech in 2018. Amazon Go represented the future of retail — cashierless stores. Cashierless stores are retail that are partially automated, allowing customers to purchase products without being checked out by a human cashier. Twenty-two Amazon Go stores operate in neighborhoods of large U.S. and U.K. cities.
The Amazon Just Walk Out system is an AI system employing a network of cameras and sensors that pick up images of the items customers are grabbing. Sensors record imagery of the items customers select and place the images automatically in a data list so that the underlying model can acknowledge the purchase when the customer leaves the store. Payment is registered through an in-store app code the customer activates upon entry. A credit card reader at the gate is also a payment option. The result is that the customer avoids a long checkout line and gains easy in-and-out access to groceries in a neighborhood store.
Amazon’s initial strategy for retaining Amazon Go customers was to wow them with integrations that leverage Amazon Just Walk Out capability, such as the convenience of setting a grocery list at home and having it appear in the smart cart.
Amazon also planned to sell the AI operational model to other stores. Several stores have adopted the Amazon Just Walk Out technology but have not yet disclosed whether they will discontinue it following Amazon's lead.
Related Article: How Amazon's Just Walk Out Tech Will Change Shopping
Peeling Back the Issue of AIOps for Customer Experience
Ars Technica and several sites reported that Amazon relied on 1,000 workers based in India to manually review transactions and label images from store videos to train Amazon Just Walk Out’s machine learning model. This was done to minimize identification errors in the model.
Two significant red flags exist. First, the use of machine learning deviated from its advertised purpose to some extent. Models are typically retrained on new data subsets so they can recognize items when processing video and audio data. While incorporating human oversight is common in machine learning as part of a checks-and-balances system, the model should still train on sufficient data to refine its parameters. The Ars Technica report suggests that the model may never reach a production-ready state of improvement.
The second issue is an ethical concern that jeopardizes the customer experience. Amazon promoted the machine learning capability behind the launch of Amazon Go. However, employing continuous human oversight contradicts the AI automation Amazon promised and the branding Amazon Go aimed for. This contradiction is at the heart of the FTC’s concerns about genuine AI usage. The use of AI in Amazon Go is not merely a product or service for the end user; the human intervention contradicts the automated experience that the Go store chain is branded with.
Related Article: Navigating the AI Customer Experience Retail Landscape
Amazon Shows Essential Role of AIOps in Customer Experience
Amazon is likely to overcome the criticisms surrounding the program. The company has established a solid reputation for reasonable machine-learning experimentation within its operations and new market ventures like Amazon Go. However, the Amazon Go experiment raises a crucial question that all marketers considering AI must address: How should a customer experience featuring AI be delivered?
Artificial intelligence operations (AIOps), a term coined by Gartner in 2016, aim to address key aspects of that question. AIOps involves using machine learning (ML) and operational data analytics to detect and respond to operational issues in real time. Its purpose is to enable automated tasks to produce an output or an element of an output for a product, service or operational process.
AIOps is a process designed to manage data generated from operational activities, derive continuous insights from that data and automatically complete downstream tasks. Its applications typically include monitoring cloud infrastructure such as storage systems, issuing alerts based on its understanding of normal versus abnormal conditions that require a response, and initiating other processes integral to the product or service delivered to the end user. AIOps is increasingly becoming a part of delivering customer experience, aiding in customer-facing processes such as service quality monitoring. Therefore, familiarity with AIOps for customer experience features is valuable domain knowledge for marketing teams.
As AIOps is implemented, it reveals data relationships across a given tech stack. This revelation uncovers how the technology functions, identifying which operations run on which data and tracing the systems that continuously interact with each other.
Learning Opportunities
Webinar
May
9
Driving Customer-Centricity in the Financial Services Industry
Discover how financial organizations are leveraging generative AI technologies.
Register
Conference
May
9
An Irish Roundtable in the Valley: The Economic Impact of AI on the World of Work Stanford 2024
Exploring the Impact of AI On the World of Work with Irish diaspora leaders and tech business champions from the North of Ireland
Register
Related Article: AI’s Role in Digital and Retail Personalization, Part 1: The Big Picture
Implementing Cashierless Technology
Amazon plans to emphasize the use of smart carts, called Dash Carts, at the Amazon Go locations. The smart carts have been available in several California and Illinois stores since Amazon Go launched. Amazon also plans to make self-checkout stands available.
These steps will remedy the measurement needs as Amazon revises its AI model and support. Their implementation comes with lessons that any business implementing AI should consider.
Related Article: 10 AI Customer Experience Statistics You Should Know About
Familiar Technology Loses Its Luster
Self-checkout addresses a need for Amazon by maintaining a touchpoint for customer activity data for advanced analytics. However, self-checkout is familiar to customers, as leading retailers such as Target and Walmart have offered it for years. Thus, self-checkout does not provide a competitive advantage or unique customer benefit for Amazon in the grocery marketplace.
Innovation does have an expiration date, as customers grow accustomed to a technology and begin to expect it as standard. Monitor the lifecycle of innovations to ensure that opportunities are timely and enhance your offering, while avoiding the promotion of features that customers no longer perceive as beneficial.
Related Article: Customer-Centric AI Strategies and Why You Need One
Identifying Signs of Stagnation in AIOps Training Progress
Training is a fundamental aspect of any AI project. Developing a functional AI model involves training data to craft the model, followed by adjusting the model parameters as more real-time data becomes available. However, Amazon Go had been operational for six years, ample time to acquire better data for refining the underlying model. Yet, workers were still being employed as recently as 2022, suggesting that no plan for slow progress had been considered.
Marketers should consider model performance inspections as a short-term analysis rather than setting an open-ended, long-term timeline. Allowing for a prolonged assessment period can burden development and marketing teams with significant technical debt, whether it's a low-performing process that hampers customer experiences or a process that makes certain customer interactions expensive to deliver. When implementing AIOps for customer experience, teams must establish a clear consensus on the advantages and disadvantages of current tools before pursuing a new solution.
Understand the Role of Human Oversight in AI Strategies
The Amazon Go incident represents an AI failure, as it relied on human assistance to confirm statistical relationships that an AIOps for customer experience model should normally handle. Proper training, even on a subset of data, should have eliminated the need for this step.
Additionally, Amazon's decision to outsource labor abroad rather than hire more workers at Go stores suggests a disregard for supporting local employment. This approach is at odds with community efforts to protect local jobs, especially amid ongoing debates over the elimination of cashier roles and broader concerns about AI replacing jobs.
Marketing teams must be savvy about the potential consequences of deviating from best data model practices. Many of the worst errors involve how people are treated in a process.
Monitor Customer Awareness of Data Usage Consent
The issue of consent was not central to the decision to discontinue the Amazon Just Walk Out technology. However, as retailers increasingly automate the checkout process, consent is becoming a more prominent concern. For instance, recently a Target customer in Illinois initiated a class action lawsuit alleging that the retailer used her biometric data without consent, violating the Illinois Biometric Information Privacy Act. Amazon faced a similar lawsuit from customers in New York City in 2023
Consumers require consent reminders that are prominently displayed during their regular interaction with a product or service. Many tech companies simply place a notice on a website page, but consumers typically consider consent when activating an app or device. Effective consent reminders should be positioned where customers most frequently use a product or service, not just in areas that require consumers to seek out information on their permission rights.
The Future of Retail at Amazon
Amazon’s shift occurs at a pivotal moment in the grocery marketplace. According to eMarketer’s US Digital Grocery 2024 report, digital grocery is projected to become the largest ecommerce category by 2026. Capitalizing on sales growth opportunities will hinge on brands' understanding of what drives consumers to purchase groceries online and how the digital grocery experience can be enhanced.
As organizations increasingly adopt AI-assisted processes, managers are becoming aware of the curves and potholes on the road to AI automation. Innovation is crucial for enhancing customer experiences, but achieving these improvements digitally is often non-linear. Amazon's decision may appear to be a setback, but the company had a sound AI automation strategy and executed it well as a brand. Moreover, Amazon’s ability to learn quickly is key to smoothing out innovation snags.
The good news is that organizations have opportunities, similar to Amazon, to glean lessons at each step.
