Artificial intelligence uses machines — generally, computers — to mimic human intelligence.
Originally printed in the October 2023 issue of Produce Business.
For all the excitement about ChatGPT, it requires significant effort for companies to harness the power of artificial intelligence and related technologies, such as machine learning. However, doing so isn’t just advantageous, it’s crucial — because any company that doesn’t keep pace with developments is going to watch as its competitors master it.
Walmart is looking at the opportunity comprehensively.
In the company’s latest quarterly conference call, Walmart President and Chief Executive Doug McMillon says the company sees AI and technology applicable to retail as central to its operational advancement.
“As it relates to technology, our approach to new tools like generative AI is to focus on making shopping easier and more convenient for our customers and members, and helping our associates enjoy more satisfying and productive work,” says McMillon. “Ultimately, the power of generative AI or any technology is only as good as the data that powers it. Our data assets are unique, and we’re excited about the potential to leverage them in new and impactful ways.”
“The power of generative AI or any technology is only as good as the data that powers it.”— Doug McMillon, Walmart chief executive
McMillon says Walmart has been preparing itself technologically to incorporate AI into its processes for some time.
“We’re taking large language models (LLMs) developed by our partners and by the broader tech community and adding retail context to create models that are uniquely suited to the needs of our customers, our associates and our supply chain,” he says.
Walmart is exploring how such technology can help it tailor operations and communications in the everyday low price environment, and, so, create a more personalized shopper experience. Walmart also is working on ways of using technology to make employee jobs more satisfying, in part by cutting back on tedious repetitive tasks it can relegate to AI.
In addition, via AI, it is making supply chain improvements based on greater efficiency and the availability of more concise and precise information about what’s moving through it.
The key to successful AI implementation is defining the business purpose or goal before tapping the technology.
In terms of potential, Waqqas Mahmood, director of strategic consulting for business advisory Marcum LLP, New York, NY, says the firm is advising its clients to look at AI as a force they can apply across departments — in finance, budgeting and forecasting, as well as operations. He says companies can identify those operations that saddle employees with simple repetitive tasks and jump on the AI-enabled opportunity to automate them.
Mahmood says a thorough beginning assessment and reassessment of IT practices should be a precursor to AI initiatives, including data management and enterprise resource planning software to ensure that ERP, accounting and other systems can talk to one another. Then, businesses should create a road map that puts AI to work in meeting immediate needs and then in addressing longer term goals. The effort should include solid security measures.
Ensure systems provide clean data so that a business has “one source of truth,” or, to put it another way, to ensure the data is accumulated in a single pool that’s reliable and available across the entire operation.
Mahmood says this process is necessary because companies often use different codes to identify the same thing, whether a product or person or process, and even mix in vendor codes. In that circumstance, AI is going to generate errors, just as a person would if they sorted through a file not realizing three different codes identified the same person.
A company needs to ensure its stored data is consistent and recognizable by AI. The cleaner the data, the better performing the AI model, and the less humans have to get involved trying to edit and rework dubious results.
In some quarters, AI is long established and past applications illustrate where it can take various functions. Pricing optimization isn’t new, as anyone purchasing constantly repriced airline tickets today understands — AI has been part of its operation for years. However, as it has progressed, AI has made pricing optimization more capable of dealing with complex cases, such as produce pricing, giving retailers the ability to make evidence-based decisions on what to charge when.
The ability to rapidly change prices to suit demand and promotional strategies is ubiquitous online, but has been slower to take hold in stores. Where retailers use electronic shelf labels linked to a communications system such as Wi-Fi, they can automatically update prices, but it’s rare for store runners to use them throughout a store.
Produce is a product category that is subject to a lot of change in stores, and Matt Pavich, senior director of strategy and innovation at Revionics, Alpharetta, GA, says, pricing optimization for perishables was more complicated to develop, but some companies have been mastering it.
“One of the bigger differentiators in how sophisticated an AI solution is, is whether or not it can handle fresh categories.”— Matt Pavich, Revionics, Alpharetta, GA
“One of the bigger differentiators in how sophisticated an AI solution is, is whether or not it can handle fresh categories,” he says. “We’re able to price fresh categories.”
Given that appearance, size, shelf-life, expiration dates, seasonality and other variable factors affect produce, price optimization has to cope with a lot of data and learns over time, particularly coupled with machine learning that can take the mass or material and refine it to satisfy business needs.
Retailers that have good measurements and data capabilities can work with AI, machine learning and other relevant technology to generate more precise information that can enhance the effectiveness of the entire produce supply chain, by improving purchasing flow at the store level and providing data that extends back through the system to ensure sufficient supply.
AI can provide efficiencies because retailers can price product in the pipeline based not only on established everyday and promotional schedules, but also on actual condition and demand at the moment.
Retailers can use pricing optimization to meet various business goals, such as getting maximum results from the introduction of seasonal fruits and vegetables. It can adjust prices down to local levels so products move in good time and curtail food waste and boost profitably. Across a large or small geography, AI-based systems can take local demand, price tolerances and favored product mix into account, as well as cost, condition and other relevant factors, in setting prices that balance movement and profitability based on forecasting, ongoing purchasing and adjustments to actual results.
With recent inflation, retailers that used sophisticated pricing optimization capabilities enjoyed advantages that led to market share and margin gains simultaneously, says Pavich.
AI PRODUCE RECOGNITION
AI has been critical in the development of self-checkout technology, and plays a role in security and detecting scanning evasion. Produce recognition — a critical issue given the diversity of products and varieties moving through stores, and their many attendant price points — is another way AI is helping retailers make the most of their produce operations.
Toshiba’s ELERA Commerce Platform provides self-checkout produce recognition, which identifies produce and eliminates the need to input produce codes at checkout manually. Retailers are already benefiting from the produce recognition innovation through improved inventory accuracy and throughput in the self-service experience.
“Toshiba’s ELERA Produce Recognition solution uses AI and computer vision technology to increase scanning accuracy and reduce the need to manually input codes and intervention from store associates.”— Yevgeni Tsirulnik, senior vice president, Incubation and Innovation, at Toshiba Global Commerce Solutions, Durham, NC
“Toshiba’s ELERA Produce Recognition solution uses AI and computer vision technology to increase scanning accuracy and reduce the need to manually input codes and intervention from store associates,” says Yevgeni Tsirulnik, senior vice president, Incubation and Innovation, at Toshiba Global Commerce Solutions, Durham, NC.
“When an item is incorrectly entered, the ELERA Produce Recognition system automatically intervenes with prompts for the shopper to correct the error and continue the checkout process on their own,” explains Tsirulnik. “If a shopper is unable to resolve the issue on their own or is ignoring the prompts, then a store associate can be notified to assist with the checkout. This gives shoppers a self-checkout experience that is faster, friendlier and smarter.”
Toshiba recognizes the shopper experience should be the primary driver of retail technology.
“To simplify the process for consumers, instead of relying on PLU numbers, ELERA Produce Recognition provides the shopper with a curated set of produce items to select from based upon what was scanned by the camera and the application’s AI,” says Tsirulnik. “The benefit is twofold: supporting ease of use for shoppers and reducing the likelihood of mistake-driven shrink.”
Tsirulnik says AI-based solutions enable the produce recognition capability to evolve into a customized solution that better suits retail needs in the store and corporate dimensions.
“This solution supports the retail industry with more accurate inventory insights, shrink and loss prevention, and provides more tailored consumer experiences,” he says.
“The AI-powered learning capability enables the system to continually scale to the needs of the business,” he adds. “The expected benefits include reduced transaction time by providing the shopper with a smaller and more accurate set of items to select from, resulting in an improved overall shopper experience.”
Based on recent retailer data, Toshiba concluded that ELERA Produce Recognition, on average, can save customers as much as five seconds per produce item lookup during checkout, compared to other traditional self-checkout systems, he says.
“The self-service experience can be riddled with challenges,” says Tsirulnik. “In one study, as much as 23% of shoppers reported avoiding self-checkout due to the inconvenience of checking out produce. But over the last two to three years, consumer interest in frictionless experiences, where they can get in and out of a store as quickly as possible, is growing.”
He says retailers realize that having the system recognize and suggest items at checkout “is vital to helping eliminate slow transaction times, inaccurate store inventory, reduce loss and give shoppers an experience that keeps them coming back.”
REDUCING FOOD WASTE
Afresh, San Francisco, CA, is applying AI in a more direct assault on food waste. The company has established a goal of eliminating food waste and making fresh food more broadly available. Afresh helps stores place accurate orders that mitigate waste and keep shelves stocked. As such, Afresh is helping food retailers not only reach sustainability goals but to command bigger profits.
Matt Schwartz and Nathan Fenner launched Afresh to directly address the lack of purpose-built technology for fresh food.
“Afresh leverages AI to help grocers make smarter decisions in fresh department store ordering and inventory management amid the complexity and uncertainty that comes along with fresh.”— Matt Schwartz, chief executive, Afresh, San Francisco, CA
“Afresh’s first product, an AI-powered predictive ordering and inventory management solution, is the only built-for-fresh solution that intelligently navigates hard-to-predict and error-prone data to drive optimal decisions for grocers in their fresh departments,” says Schwartz, Afresh chief executive. “Afresh enables grocers to reduce waste, empower store teams, and drive profitability across the business.”
The founders began operations by speaking with, and shadowing, people in the food supply chain to learn about their unique challenges, then built their AI-powered solution to address the needs they encountered, says Schwartz.
“Afresh leverages AI to help grocers make smarter decisions in fresh department store ordering and inventory management amid the complexity and uncertainty that comes along with fresh,” he says. “Variables like perishability, changing display sizes, and seasonality make fresh departments particularly tricky to navigate, but the Afresh platform uses cutting-edge machine learning to deliver a unified view of all the factors that influence an optimal decision for fresh ordering.”
The Afresh AI-based system gathers accurate data by using targeted human inputs, which helps assure they enter the system clean.
“Each order day, fresh department managers follow a short list of required inventory checks that help generate AI-managed, pre-filled orders,” says Schwartz. “By asking for targeted counts only where needed, this data is ultimately more accurate, and store associates have more time to focus on value-additive tasks, such as interacting with customers.”
The system has its own approach to demand forecasting as well, leveraging AI to model inventory position, data input quality, perishability and other specific factors.
“The system leverages a new and rapidly evolving field of AI specifically focused on decision-making in uncertain conditions to arrive at order recommendations for hundreds of items that fresh department managers accept, on average, 94% of the time,” says Schwartz, adding grocers who leverage Afresh’s AI for ordering and inventory management “see lighter backrooms, faster inventory turns, more efficient store teams and happier customers.”
“Retailers create 40% of all food waste, and over two-thirds come from fresh categories,” says Schwartz.
He says Afresh grocer partners enjoy a 25% reduction in food waste, which means, since 2019, the company has prevented over 43 million pounds of products getting trashed.
• • •
The Case for AI Content
By Mike Duff
Content production is an easy introduction to AI.
When it comes to artificial intelligence and related technologies such as machine learning, popular attention has been seized by generative AI such as ChatGPT. In the case of building content for marketing and other purposes, generative AI can be an effective tool, and it is an approachable part of the technology.
Google and Microsoft are letting people try it to get some understanding of what AI can do, while also selling relatively easy-to-access browser tools and applications. They and others, including ChatGPT and OpenAI, are also building tools and platforms that make various applications easier to use.
“If you are anticipating using AI anywhere, content production is going to be one of the first places you want to flip over a stone and see if there is some way to do that. It’s extremely affordable,” says Jordan Brannon, president and chief executive of Coalition Technologies, a marketing agency that works with businesses to bring the latest capabilities to consumer outreach.
“If you are anticipating using AI anywhere, content production is going to be one of the first places you want to flip over a stone and see if there is some way to do that. It’s extremely affordable.”— Jordan Brannon, president and chief executive of Coalition Technologies
“There is a lot of helpful content and tools to make it more accessible to non-technical users, and it’s quite good. For any industry, if there is some conversation about where AI today plays a role in marketing, the content piece is one of the first ones we want to touch on.”
PRIME THE PROMPTS (GARBAGE IN, GARBAGE OUT)
The content users want to create has to be shaped by parameters, or prompts, that will narrow results. Just plugging the word “kiwi” into an AI-driven content tool is going to produce content that is too broad to be immediately useful.
Brannon explains that understanding where AI tools are sourcing data is important in setting prompts. In the case of AI for content marketing, the tools available are largely going to use the web as the basis for output. That means, in setting prompts, the user has to keep in mind the diversity and variable quality of the content available.
“We really emphasize understanding it is sourced largely from web-based content,” he says. “It’s poorly policed for accuracy. Generally, people experience more generic responses when they ask for longer form content without a lot of parameters.”
Setting a goal for the content — determining who it is supposed to reach and deciding what points need to be addressed — is a proven way to approach prompting, says Brannon.
“When we’re creating AI-driven content for paid advertising or elsewhere, we’ll begin with a series of prompts that address the who, what, when, where, why of the content,” he explains. “Next, we’ll begin to prompt articles and longer-form content with specific objectives to come up with the individual components, and then we’ll ask the AI in the end to put it all back together in a cohesive argument with a more polished structure.”
The business purpose is the basis of the exercise.
“It might begin with something like: I am trying to sell kiwi to customers in upstate New York, or perhaps more broadly throughout the Northeast, and I want them to understand the health benefits of the kiwi while romanticizing the flavors that are identifiable as only coming from the kiwi,” he says. “My customers are mostly going to be upper and middle class, two-income households, one or two children.”
“I’m creating the who, what, when, where, why type of material,” he continues. “Then, I can begin to prompt the LLM (large language models) with some questions or requests: Please write a 50-word opening paragraph for an email that I would like to send to my subscriber list extolling the kiwi benefits. Then, I want you to highlight in a 50-word paragraph on the specific health benefits of the kiwi as compared to other fruit. Then, I want you to provide three different summer-appropriate recipes that utilize the kiwi that are refreshing and speak to the outdoors in the sun. Then, I want you to talk about the history of the kiwi in the United States.”
An LLM takes all of those different prompts and creates an article.
In budgetary terms, AI can be a cost-saver, not only because it saves staff research time, but because it can also create different versions of the same content appropriate to different channels, such as a blog, social media post or email blast.
The advantages of initially considering Google, Bing, ChatGPT, Open AI and some other popular artificial intelligence providers is that the tools are already built into apps and web browser functions and are simpler to understand and use.
GET SPECIALIZED HELP
In addition, Brannon says businesses can look for specialized AI tools.
”They often have a little more simple focus, so if you’re looking for some image generation or image editing, transcriptions for calls or presentation, if you’re looking for text for marketing collateral, if you’re looking for AI to help with data analysis from marketing activity, make a quick Google search, or you can even go to Bing’s or Google’s AI tool and ask them what they recommend for your specific AI use case,” he says. “You’ll find you can get some very specialized recommendations that are often more focused.”
Brannon cautions it’s important to have humans as filters who understand the content as output of AI has to be evaluated. Also, because the output is what he calls “word soup,” it has to be judged in terms of coherence and effect.
• • •
Instacart Develops AI-Powered Help for Retailers
By Mike Duff
Instacart is using AI throughout its operation, creating new tools and features to improve the customer experience.
For example, the San Francisco, CA-based company just announced multiple updates to Instacart Storefront. The platform powers owned-and-operated e-commerce storefronts for more than 550 retail banners, and Caper Carts, Instacart’s smart carts, to help retailers advance e-commerce, digitize their stores, and build AI-powered omnichannel experiences for customers.
Instacart says it is bringing AI-powered conversational search to the new Instacart Storefront so customers can ask open-ended questions about meal components and food nutritional qualities.
“We’ve long believed the future of grocery, and commerce in general, isn’t online or in-store, it’s both,” says Asha Sharma, Instacart chief operating officer, in a company news release introducing the new Instacart initiatives. “And now, more than ever, it’s being supercharged with AI.”
In addition, it is adding in-store rewards for its Caper Carts so retailers can offer customers points, coupons or badges for completing actions like logging into a loyalty account, adding certain items to the cart, or trying a Caper Cart for the first time. A new Caper Cart dock allows retailers to provide a permanent place to store and charge the wagons.
Instacart also offers the ability to order made-to-specs items, such as deli sandwiches or custom cakes directly from the carts using FoodStorm, an Instacart order management solution.
It is also introducing a new in-store mode, which turns retailers’ apps into companions when customers shop. In-Store mode helps customers see what’s in stock, view important details about items on their list, such as nutrition information or whether they’re EBT SNAP-eligible. Consumers also can get product recommendations, sort items by aisle and access in-store promotions and discounts.
The In-Store mode also provides data to help retailers better understand their customers, no matter if they shop online, in stores, or both.
“We know that omnichannel customers in particular are more valuable to retailers, which is why Instacart is developing more solutions that help retailers serve their customers no matter how they shop,” says Sharma. “And good data is the foundation for good AI solutions for retailers.”
Although in a quiet period due to its initial public offering bid and filing with the United States Securities and Exchange Commission, an Instacart spokesperson told Produce Business that the company is “incredibly excited about the rise of generative AI. It’s worth noting that AI and machine learning are already deeply embedded into everything we do today. These technologies allow us to solve the complex challenges in grocery that come with integrating with thousands of retailers’ catalogs, each with tens of thousands of SKUs, to provide tools like personalized search and suggestions, availability and replacements, as well as order batching to minimize fulfillment costs.”