Retail 2021: AI Ushers in New Decade

Making store employees’ life easier is key, especially at a time when retailers must do more with less. As an example, NTT DATA’s Smart Retail Operations data analytics platform is intended to enable proactive communication of exception events — such as out-of-stock situations or compliance issues related to planograms, pricing, or promotions. It processes data from almost any data source, including IoT sensors, and transforms it into actionable intelligence. This helps to accelerate Smart Store pilot projects and subsequently scales across many stores.

Customers want their loyalty rewarded with more flexible programs and differentiated experiences. To strengthen customer loyalty, retailers must engage customers proactively through hyper-personalized benefits tailored to their interests. An AI solution can generate product recommendations and early churn alerts based on past behavior patterns. For example, using the data from customer’s purchases, you could predict which fresh and pre-cooked dishes customers will buy.

Customers expect a quick, easy, and safe shopping experience. NTT DATA is working with stores to provide unmanned shopping experiences. The approach combines AI, machine learning, IoT sensors, and computer-vision-based algorithms to enhance the consumer experience. To use it, customers download a dedicated app, select their preferred payment method, and grab the product in-store. The customer can exit without scanning products.

RIS: How should retailers address the ethical and privacy concerns that continue to be raised by AI?

Kushner: My colleague, Lisa Woodley, reveals in an interview on “Designing Ethical Customer Experiences,” that at times the question of “what should we do” is lost to the question of “what can we do” as technologies like AI become more advanced. And what we can do often raises ethical concerns. When misused, AI provides a real opportunity for companies to manipulate and abuse customers, especially in the retail space. We can drive positive ethical change if we bring a human perspective to technological innovation.

Ethical considerations center around bias. As human beings, we are naturally biased, but with AI, those biases can surface faster and with greater harm than ever before. Facial recognition may be important for retailers that want to prevent theft and loss of inventory. However, implementing AI programs associated with this data necessitates being cognizant of subtle implications. For instance, a recent article in Wired online opens with this sentence: “Men often judge women by their appearance. Turns out, computers do too.” It reports that researchers had sent images of congressmembers through Google’s image recognition service, which applied annotations to the individuals’ physical appearances, then labeled the male images as “official” or “businessperson” and the female images as “smile” or “chin.” This is a good example of perpetuating a long-time gender bias. 

AI represents challenges to business because privacy concerns as well. Privacy has been a topic for as long as the Internet has existed. The General Data Protection Act (GDPA) in Europe, the California Privacy Rights Act (CPRA), and the Brazilian Lei Geral de Proteção de Dados (LGPD) all lay out how an individual’s privacy must be protected and stipulate some heavy fines for non-compliant companies. These new laws also apply to the use of data in algorithms, including algorithms that harness AI, making it impossible to apply data that could identify an individual in any way. So, AI data scientists must be very aware of privacy laws in the countries where they develop AI programs, as well as in the countries where these programs are deployed.

To avoid bias and violating customer privacy, AI applications need to be created by diverse teams. That means diversity in thought as much as diversity in ethnicity, gender, etc. The first task of this team should be to evaluate the data that will be used with the algorithm to ensure that it was collected, managed, and curated without bias and in line with privacy considerations.

RIS: It has been said that people are the real key to digital transformation. How does this play out in a digital transformation that involves embracing AI, and what role does change management play in this scenario?

Krishnanji: My colleague, Kim Curley, insightfully noted in a recent article that having a growth mindset is key at the organizational and individual levels alike. While being good at implementing and using a tool is important, success is dependent on how well you can handle change. A human-centric approach and change management are critical to AI success, particularly at scale.

The culture of the organization will influence AI adoption. A culture that is data-driven, analytical, collaborative, vulnerable, curious, and — most importantly — nimble enough to take a ‘test and learn’ approach is vital. AI can generate constant, significant change within an organization, and only those who can go with that kind of flow will be successful.

AI projects must involve business stakeholders from the beginning to frame the right problem to solve, to build trust and credibility, and to set clear intentions as to how the organization will digest the kinds of change created by AI. Listening to and gathering feedback from users is key. Algorithms — unlike humans — don’t consider what happens when the result is applied. So, upskilling your workforce to become AI coaches enables more rapid improvements to AI solutions and broader application.

RIS: What do retailers need to keep in mind to successfully build, deploy and scale AI capabilities

Kushner: As retailers prepare for successful adoption of AI into their operational environments, there are a few important things to keep in mind:

Find a sponsor. AI requires support from the most senior executive levels to truly gain a position within a retail operation. The sponsor should be able to provide the necessary funding, along with much-needed organizational support.

Identify a good problem. A good, first AI problem to solve impacts the business, is somewhat easy to explain, and addresses a customer or employee need. Business impact means increased revenue, decreased costs, or increase in customer loyalty. Problems in these areas can require different types of AI applications, so zeroing in on one that can be easily explained to your sellers and marketing teams is imperative. Most operations begin with a straightforward machine learning algorithm that "learns" patterns from the data it gathers as it is applied. For example, if your goal is to increase revenue, you might create a cross-sell, up-sell algorithm that deduces from previous customer transactions what the “next best” purchase may be.

Ensure your data is solid. Most AI applications fail because of poor data. Either the data is incomplete, inaccurate, biased, or is totally missing the variables that are most predictive or required for solving the problem. Data is the foundation of any good AI solution, so break down data silos to understand where data will be useful and how to make it available to your algorithms in timely fashion.

Think scale at the start. A good many data scientists have been hired across corporate America to create AI algorithms. Unfortunately, not all of them understand that the “science” is the first part but deploying the science is the important part. This is why many AI proofs-of-concept or pilots never make a difference in the business. The key to success with AI is to apply it to the most appropriate stage of the business process. AI operations becomes very important here. Why? AI operations simply takes the data scientist-created algorithms and embeds them into application systems. But these applications must be monitored and governed as they are deployed because they learn with each new piece of data that is processed. The AI algorithms will need adjustment to ensure success over time — AI is not a one-time project.

Before launching any AI project, remember that this is a journey, not a destination. AI is designed to keep learning within the environment. Consequently, business processes may have to be redesigned so that the algorithm can be applied in a beneficial way. Or, data may need to be collected in a way that makes it easier to use. These situations may require some additional work to ensure the success of the AI project.

LEARN MORE

• Drive digital to rapidly adapt to changing consumer requirements with our Smart Retail and Consumer Packaged Goods industry services
• AI Study by NTT DATA and Oxford Economics – The Great Shift to AI and Automation
• Point of View paper on Accelerate to an Intelligent Enterprise
• Turn Data into Decisions with NTT Smart Platform
• Focus on the people who matter with our Customer Experience services