How Did Companies Use Generative AI in 2023? Here’s a Look at Five Early Adopters.

(Illustration: THOMAS R. LECHLEITER/THE WALL STREET JOURNAL)
(Illustration: THOMAS R. LECHLEITER/THE WALL STREET JOURNAL)

Summary

Business technology leaders in construction, travel, retail, healthcare and energy say AI is already improving productivity and changing customer behavior. But they are also sorting through its high costs and limitations.

Generative artificial intelligence emerged this year as the most buzzed-about new technology for businesses, promising to supercharge productivity while transforming the way white-collar work gets done.

But AI’s high cost, need for specialized talent, and legal and privacy risks have stymied attempts to fully realize that promise, with many businesses cautious of moving beyond early experimentation. Still, there is little doubt that generative AI will completely reshape enterprise technology. And some businesses have already begun incorporating AI into their operations by using it to write code, create marketing and sales content, and bolster customer support.

Enterprises worldwide will have spent around $19.4 billion on generative AI solutions by the end of 2023, according to an estimate from International Data Corp. That spending—which includes generative AI software and related hardware plus IT and business services—will reach $151.1 billion by 2027, the research firm forecasts, translating into a compound annual growth rate of 86.1% over the four-year period.

Companies for the most part are testing the limits of the technology, reinvesting in AI and cloud more broadly as they organize their corporate data, assessing cybersecurity and other risks, and setting up guardrails for its safe use. Many chief information officers and chief technology officers are looking to 2024 as the year in which generative AI proves its worth and its high price tag.

Here are five companies that have figured out how to integrate generative AI into their products and operations, and what they learned in the process. Some including online travel agency Expedia have already put generative AI in front of customers, while others, such as healthcare system Mass General Brigham and construction software maker Bentley Systems, are targeting 2024.

Wayfair: Meet AI, your new interior decorator

Wayfair, the Boston-based online furniture seller, is using generative AI to help customers redesign their living rooms. Once customers upload a photo of their living space, Wayfair’s Decorify tool uses the image-generation AI model Stable Diffusion to create new versions of it in styles such as glam and Midcentury Modern.

Aside from its design tool, most of Wayfair’s generative AI applications help improve employee productivity, proving that there is economic benefit to implementing the technology, said Fiona Tan, the company’s chief technology officer. In some cases, Wayfair will keep using its existing machine-learning models, which are cheaper to run and can carry out prediction and optimization solutions more effectively than generative AI.

Wayfair’s design tool, which is free to use, creates photorealistic images based on a customer’s uploaded photo. But because it relies on a generative AI model, that also means the occasional odd glitch such as a table leg that doesn’t look quite right, or instances where the AI can’t identify windows and mirrors, Tan said.

Decorify also gives customers prompts about Wayfair products that are similar to those in the AI image, helping connect the AI-generated world with the real one, according to Tan. The company said customers have created more than 100,000 designs and have purchased products via the tool since its launch in July.

“This could be a very viable way to shop for things like your home and style-based categories, that are really hard for you to articulate what it is you really want. Being able to see that is helpful," Tan said.

Schneider Electric: Picking the right AI with energy use in mind

Schneider Electric, one of the world’s biggest manufacturers of electrical and automation products, is choosing AI models that are smaller—and therefore less energy intensive—for certain applications, said Philippe Rambach, the company’s chief AI officer.

“Large language models are great in summarization, in text generation," he said. “They can do things like forecasting and optimization, but not so great. And at the cost of both dollars and energy consumption, it doesn’t make sense."

Generative AI runs on an immense amount of energy from data centers, which are supplied by electricity from strained power grids. AI could consume up to 3.5% of the world’s electricity by 2030, according to an estimate from IT research and consulting firm Gartner.

Schneider Electric, which makes goods such as light switches, electric-vehicle chargers and home-automation systems as well as energy-management software, chose to use OpenAI’s GPT-3.5 model, which is much smaller and uses less energy than its latest systems, to power its internal company chatbot, Rambach said.

The France-based company is also using generative AI to help its customers compute and analyze their carbon emissions as if they were conversing with the chatbot ChatGPT.

Customers needed a way to use its Resource Advisor tool, which can visualize and track their energy data, in a faster and easier way than traditional software, Rambach said. The “copilot," which was built using Microsoft’s OpenAI service, allows them to ask questions such as “which of my factories has the highest emissions?" he said.

Mass General Brigham: Data-driven diagnostic insight, personalized care

Mass General Brigham, a Boston-based healthcare system that is affiliated with Harvard Medical School, is applying generative AI to help identify patients with similar profiles, with the hope that this information can help doctors treat patients based on what has worked for others in the past.

The healthcare system is working with AI medical-imaging firm Annalise.ai, as well as other companies and startups, to bring AI solutions from development to market as part of its newly launched commercial AI office. It will roll out some of its first commercial AI-enabled diagnostic products in the first half of next year, said Keith Dreyer, Mass General Brigham’s chief data science officer and vice chairman of radiology. “The field is moving remarkably fast."

Using data like diagnostic information, medical records, patient discussions and genomic and genetic data, large language models can better identify “patients like me," Dreyer said.

Previously, the process to create a single “convolutional neural network," or AI algorithms that can analyze images, required several million dollars and several years of training. Now, large language models can be trained in a matter of months via a process called “few-shot learning," that requires far less training data, according to Dreyer.

Though large language models are more expensive to run than convolutional neural networks, a faster development cycle still results in time and cost savings, he said. Generative AI can correlate both text and image data from scans, and extract and correlate it in less time than other deep-learning AI methods.

Expedia: A personalized travel assistant, not a silver bullet

Expedia is embracing the use of chatbots to help travelers ask for recommendations and make bookings, but the online travel agency doesn’t view generative AI as a silver bullet.

“Generative AI is an assistant, it’s going to help you get there, help you explore and discover," Expedia Chief Executive Peter Kern said. But the idea of asking AI to book an entire trip is “a grossly exaggerated opportunity," he said.

Instead, the Seattle-based company will personalize travelers’ booking process and use AI to enhance the experience—especially as consumers aren’t yet fully comfortable using generative AI, Kern said. It is also using AI to automate processes like customer-service-call summaries to reduce costs.

For instance, Expedia will help answer questions based on customers’ travel history and preferences, according to Kern, providing tailored help throughout the entire booking process. “What we have to do is use everything we can to take the worry out, make it personalized so you’re not fishing around in the dark," he said.

Expedia’s competitive advantage is the roughly 70 petabytes of data it has on customer booking patterns, preferences and other information, said Rathi Murthy, the company’s chief technology officer. The company recently completed a multiyear technology overhaul to connect its data sources and power its machine-learning systems. Its chatbot is powered by OpenAI’s ChatGPT.

“Now, we can truly connect all of this with large language models, and drive that through personalization in a much more proactive manner than we could in the past," she said.

Bentley Systems: More resilient infrastructure and automated drawings

Bentley Systems, an Exton, Pa.-based maker of construction-industry software, is developing generative AI-powered tools for making drawings such as site plans, and systems that could propose infrastructure designs with better climate resiliency.

“What will be the impact of two more degrees on the average temperature on the planet, on highways, on bridges?" said Julien Moutte, chief technology officer at Bentley Systems. “We need to be able to run all of those designs through simulations that are going to assess those future conditions so that we can design infrastructure and make it last."

In the near term, Bentley Systems is helping engineers create drawings with generative AI, with the goal of adding the feature to its software next year. Most spend 30% to 50% of their time producing such structural and construction drawings, according to Moutte. A “copilot"-like assistant that helps engineers codesign with generative AI is also being developed.

“In the same way that the AI is able to generate pictures, you can generate drawings that are conveying this important information," he said.

AI is already widely used in construction for detecting cracks and rust on bridges and predicting when roads will need repairs, but generative AI is now capable of combining data from Bentley Systems’ existing tools to suggest climate-resilient designs, Moutte said.

For instance, large language models can ingest data sources including structural analyses, building codes, laws of physics, climate conditions, and the effects of new, more sustainable building materials on infrastructure design.

To put advanced AI systems into action, Bentley Systems is investigating methods of fine-tuning or customizing AI models with its own design data and simulation tools, Moutte said. That will result in smaller and more specialized systems, which are also cheaper than using large models directly from vendors such as Google or Microsoft.

“AI is not only analyzing content and providing insights," Moutte said. “It’s taking your input as a requirement and it’s generating new data based on that."

Write to Belle Lin at belle.lin@wsj.com

Catch all the Business News, Market News, Breaking News Events and Latest News Updates on Live Mint. Download The Mint News App to get Daily Market Updates.
more

MINT SPECIALS

Switch to the Mint app for fast and personalized news - Get App