Retail, historically a top proponent of artificial intelligence, is facing an embarrassment of riches now as a growing crop of platforms crowd the scene armed with data science’s greatest hits, including machine learning, neural networks, deep learning, generative adversarial networks and stable diffusion.
The headlining tech is generative AI, a machine learning-based form of AI that can produce new audio, images, writing and more based on natural language prompts. Learning the ropes, and the right prompts, takes some trial and error, but the bots offer more natural interactions — everyday language, with a contextual boost for back and forth exchanges — and they can revise their work handily, as well as create art that’s virtually indistinguishable from that of humans. They also won’t flinch at complex requests and calculations (although some commands take longer than others).
Related Articles
Think of the tech as a creative genius that can take direction, minus the personal drama, or that know-it-all friend that really does seem to know almost everything, including how not to brag annoyingly about it.
You May Also Like
It’s an obvious fit for retail.
Shopping is a uniquely human experience that often relies on warmth and relationships, and this version of AI can pull off impressively humanistic traits in how it communicates, serves up information, presents analysis and creates or edits content. The most obvious example is ChatGPT, the generative AI bot by OpenAI responsible for setting off the current momentum, though it’s far from the only game in town.
ChatGPT wowed the public last fall as a smarter example of machine intelligence than the often robotic, scripted or automated chatbots and voice assistants that came before. Its reading and writing comprehension and abilities impressed users, and it even understood prompts for surprising requests for subjective or tonal tweaks — think rewriting copy with a funnier or sadder tone, or creating a story using different styles of writing. It can also read and understand uploaded images, though it can’t actually create visual assets.
That’s the domain of the company’s other product, DALL-E2, an AI image generator that can churn out convincingly photorealistic images and artwork. But ChatGPT can help the effort by writing prompts for the image service, so it can create just what the user had in mind.
Since OpenAI released software development tools in March, which allow outside developers to work with ChatGPT, an explosion of new services and feature updates erupted, as well as competitors. Microsoft, a major backer of OpenAI, released Bing Chat based on ChatGPT’s tech. Google’s chatbot, called Bard, took a similar approach, using large language models that send huge data sets through to train and polish the model quickly.
A useful analogy is a garden hose versus a power washer. Both can clean off dirt and debris, but releasing a trickle will take far longer than unleashing a storm that blasts off every unwanted bit of detritus.
Now Google is aiming the blast at retail.
The search giant has been ramping up efforts to court retailers in recent years, and its most recent efforts bank on AI being a powerful new enticement. In March, Google Cloud, the company’s cloud division, partnered with Shopify on a new effort that allows Shopify merchants to use AI-powered Google searches and recommendations in their e-commerce stores. The company also announced its intention to support generative AI inside Vertex AI, its end-to-end, cloud-based machine learning platform.
At its Google IO developer conference in June, it brought the receipts. At an event tied to IO, the company laid out an array of offerings such as Codey, a code generation and completion tool for creating chatbots.
“That may sound technical, and why do I care as a retailer or consumer leader, but the reality is, this fits right in with retail’s core agenda to do rapid test and learn,” Google Cloud’s Carrie Tharp told WWD. Formerly Google Cloud’s vice president of retail and consumer, Tharp recently assumed a new role as vice president of strategic industries. But years working at retail companies, including The Neiman Marcus Group, Bergdorf Goodman and Fossil, among others, continue to inform her perspective.
“Frankly, [retail’s] an industry that does a lot of technology through hacks. So they often have constraints around their technology stack, whether it’s the e-commerce stack, MarTech stack etc, and a business leader can feel constrained by how fast can I develop a new customer interaction?” she explained.
Now brands can create new scripting to automate marketing campaigns, rapidly develop a web experience and other tasks. “So instead of just relying on humans, you’re getting a generative AI assist, where the code can already be completed or partially completed to your specification.…I don’t think I’ve met a single retail executive that’s happy at the speed at which their teams are operating to drive digital transformation.”
There’s a new Imagen text-to-image generator for studio quality visuals. Tharp is already seeing a lot of traction for this tool, which can be applied to areas like concept design, product packaging, campaign imagery and one-to-one personalized images, so companies can express their brand in specific, micro-personalized ways. Chirp is a universal speech-to-text model that supports more than 300 languages. Tharp immediately thought of global brands and retailers that deal in different languages and accents. They can bake Chirp into their own shopping apps for external use or deploy it on internal portals, so employees can use it in-store.
Vertex AI isn’t geared toward one sector specifically, but it’s clear that Tharp had given potential retail applications a lot of thought.
Later in June, at a presser in New York, Google doubled down on retail once more, specifically fashion, telling WWD that it used AI to develop a new apparel try-on functionality for search. The company shot various real-world models of different sizes, genders, ages and ethnicities, then used machine learning to layer clothing on them that portrays wrinkles, creases and other textile details realistically, so people can envision the clothes on models that look like them.
In general, the company wants to make AI development simpler, while also focusing on task-specific solutions. Part of the process is Reinforcement Learning from Human Feedback (RLHF), a tester program that can give human feedback to the AI about whether the machine’s output to various prompts is realistic, safe and relevant to the prompt or question. Generative AI created with large language models, including ChatGPT, often make human reinforcement an essential part of the process.
Google isn’t the only tech-maker to home in on specific jobs or tasks. JasperAI and Copy.ai are popular alternatives for marketers looking for bot-written blog posts, social media blurbs, marketing copy and other text. Others such as Surfer SEO specialize in search engine optimization.
That just scratches the surface for retail. Aarthi Ramamurthy, chief product officer at CommerceHub, can already envision its effect throughout the e-commerce operation. With a platform that connects online shops like those of Tapestry and Nordstrom with more than 40,000 suppliers, the firm already processes an average of 2.4 billion daily transactions and more than $50 billion in commerce annually. But Ramamurthy appears to be bracing for a generative AI-powered retail revolution.
“Generative AI is a game-changer for e-commerce and [the transformation will] happen much sooner than any of us think it will,” she told WWD. Retail experts believe that new shopping experiences are a crucial area of development for the tech, particularly for more personalized, curated features. Ramamurthy agrees, but also sees other benefits, because it can “meaningfully reduce the operating expense for retailers of all sizes by increasing speed to site for new products and reducing the manual effort required with today’s methods.”
Add in other business approaches, and the combination could add up to a transformative stage for retail. “We expect that these trends will combine with asset-light models like drop shipping and marketplaces to help retailers better serve their customers, more rapidly adjust selection, better personalize the ability to serve customers, all while growing sales and margins,” she added.
According to market research firm Tractica, the worldwide tech market for AI will balloon, with sales going from approximately $9.5 billion in 2018 to roughly $118.6 billion by 2025. As for the impact on retail, researchers at IHL Group estimate that the impact of generative AI on the sector will boom over the next several years, hitting $9.2 trillion by 2029.
This rampant acceleration is already starting to happen, with companies like Shopify, Carrefour, Instacart, Mercari, Zalando and others deploying generative AI for customer service, as personalized shopping assistants and more.
After years of irritating customer service bots and hit-or-miss shopping assistants, the new customer-facing offerings really need to distinguish themselves — or at least wipe away the memory of such uninspired interactions, if nothing else. Since AI models are only as good as the data they train on, the quality of the data has to be a key factor.
The new AI ingests massive volumes of information. Where it comes from, who owns it, how accurate or factual it is, if biased material made it through and whether or not it includes personal, confidential or copyrighted assets and information are all crucial considerations, and likely to become more pressing still as additional platforms crowd the space.
The situation ties into an old saying in math and computer science circles: “Garbage in, garbage out.”
For Salesforce, that’s an emerging opportunity in a new digital frontier.
In June, the tech company took the wraps off a new suite of products called AI Cloud, a set of tools and AI models that allow companies to pour in their own data. Chief executive officer Marc Benioff is promoting them as safer tools, because they can prevent sensitive information or assets from going in and training the models, and it can even identify potentially harmful outputs, he said.
In its latest survey, 61 percent of sales, service, marketing and commerce employees currently or plan to use generative AI. However, almost 60 percent don’t know how to ensure that only trusted data sources are used, and close to three quarters believe generative AI poses security risks.
“[Our AI team has] to really be able to use these next generation models,” the Salesforce chief executive officer Marc Benioff told the audience at a June event, “but have that capability to deliver a trusted experience to our customers.”
The solutions may appeal to specific sectors that deal with stringent regulations, such as banking. But the reality is that every business needs to ask the tough questions and understand the nature of the data underpinning their AI, especially in this early stage. There are no industry wide standards yet, so if a customer experience goes haywire, even ChatGPT may not be able to concoct a story good enough to turn things around.