AWS, PepsiCo and Other CEOs Tell What’s Needed to Scale AI

The CEOs of several Global Fortune 500 companies said artificial intelligence (AI) is revolutionizing their operations, while acknowledging that most companies are still grappling with implementing and scaling it, during a power-packed panel Tuesday (Jan. 21) before the World Economic Forum (WEF) in Davos, Switzerland.

The panel comprised of Matt Garman, CEO of AWS; Ramon Laguarta, CEO of PepsiCo; Julie Sweet, CEO of Accenture; Paul Hudson, CEO of Sanofi; and Amin Nasser, president of Saudi Aramco, the national oil company of Saudi Arabia. They offered insights from across the cloud, consumer goods, consulting, energy and pharmaceutical industries and sectors.

Their high-level discussion revealed a striking contrast between companies already seeing transformative benefits from AI and the majority still struggling to implement it effectively. A new WEF white paper showed that 74% of companies struggle to scale AI, with only 16% prepared for AI-enabled reinvention, said moderator and CNBC co-anchor Sara Eisen.

The focus of the panel was how to operationalize AI, going beyond experimentation in proofs of concept to implementation. “While most companies are still piloting, dabbling, most CEOs that are a little bit apprehensive … are running lots of proofs of concepts because it’s safe to just do that a few years before retirement,” said Sanofi’s Hudson.

Hudson said most companies stumble in AI implementation today because they relegate it to the company’s chief data officer (CDO) or other tech leader. He said it’s one of the “biggest mistakes” a company can make with AI due to the “nature of change and the courage needed to change business processes at scale, to better optimize resources and go chase miracles.” At Sanofi, he is the one who takes charge of AI, not the technologists.

Hudson said Sanofi’s use of AI is making its “probability of success” in discovering new drugs “so much higher.” He said AI is also helping discover drugs for diseases that are considered “undruggable” or those currently still without any treatments. Sanofi is using AI to validate targets with over 90% accuracy, helping to shorten the typically long and expensive drug development process. The drugmaker also has integrated AI deeply into its processes, such that its drug development committee now begins sessions with AI recommendations on whether drugs should advance to next phases.

Thus far, Hudson said no pharmaceutical company has used AI to go the full route from drug discovery to FDA approval, a process normally taking 12 to 15 years. “We’re still waiting for the first medicine that’s done the entire journey,” he said, adding, “In the race of turtles, we are the lead turtle.”

It’s About Data and People

Garman of AWS said many companies have not even moved their data to the cloud. He said data that is organized, labeled and set up in the cloud is critical because it feeds into generative AI’s large language models. Without it, it’s “actually hard to get value out of those AI projects.” In his view, the biggest hurdle for companies is “just being in legacy, on-premise, by far.”

As for enterprise clients with their data in the cloud, AWS provides free AI training to them. Garman also said it’s important for clients to understand that while it could take 12 to 48 months to deploy AI in their companies, engineers have to already be thinking ahead to future capabilities two to four years out since advancements in AI are moving so quickly.

Saudi Aramco’s Nasser said AI has been a “transformative force” for the energy industry. He said his organization receives 10 billion data points daily to improve operations. For example, seismic analysis that once took months to run now takes weeks, days or even hours to help Aramco find more oil. Also, they use AI to predict equipment failures based on the data they are getting. It reduces downtime and increases their efficiency.

But to get to this point, Nasser said Aramco needed to have an infrastructure in place for AI to thrive. One critical factor is having people with the right skills to identify and develop the appropriate use cases. Nasser said Aramco itself has 450 use cases at present. “Each use case is a project of its own, and it ends up with huge benefits,” he said. Nasser noted that it is important to pair an AI expert with a subject matter expert to work on use cases.

Upskilling the workforce is one way to have an AI-skilled staff. PepsiCo’s Laguarta said the company is training its 330,000 employees on AI. The company’s divisions range from agriculture and manufacturing to logistics and consumer engagement. PepsiCo is using AI to enhance crop yields, optimize transportation routes and develop personalized nutrition solutions. “It’s a huge transformation,” he said. However, companies must be careful not to create a divide between digital and analog workers.

AWS’ Garman added that employees must be convinced that AI will not take their jobs for them to embrace it. One of the things AWS did was to showcase how AI can make employees’ lives better. It took a boring, tedious internal task — Java upgrades of the company’s software — and automated it with an AI tool it built internally. Normally, it would have taken a team of five AWS people around more than 4,000 years to do it, he said. The AI tool let them do it in “a couple of months,” Garman said. It was “taking away work that no one wanted to do in the first place.”

In addition to upskilling employees, they should seek external expertise. Accenture’s Sweet said collaborations and partnerships are also important to innovating with AI. “Innovation happens only with partnerships,” she said. “Any company that says they can innovate only internally is almost by definition not innovating.” Collaborating can help unlock and accelerate advances in AI.

Three Trends in Implementing AI

For companies that are already deploying AI, Sweet identified three major trends where she sees traction. The first is hyper-personalization for growth. She pointed to the example of Radisson Hotels, which can show a traveler different photos of the same hotel depending on their interests. If they want to go skiing, Radisson can show them a hotel in the Swiss Alps. If they want to go to a spa, different hotel photos show up in real time.

“Now we’re seeing [this trend] in banking, we’re seeing it in retail, we’re seeing it increasingly in health,” she said.

The second trend is physical AI for efficiency, such as reducing manufacturing defect investigation time from weeks to hours. Sweet said a global tire manufacturer usually had to wait weeks to investigate an equipment failure on the production line. With AI, it merely took hours. “It makes a huge difference [to business] if that line is stopped for a week or two weeks versus a few hours,” she said.

The third trend is using AI to accelerate legacy system modernization. “GenAI is helping us be able to get off the mainframe faster,” Sweet said.

She also encouraged companies to adopt AI responsibly, given all the potential harm it could create. She said only 2% of companies globally have robust responsible AI programs in place, a percentage that has not changed in the past year. “We’ve really got to move that number,” she added. This builds trust in AI systems and is the “key to scaling.”

Looking ahead, AWS’ Garman dismissed concerns about AI hitting an innovation wall. “If you look at how these technologies evolve over time, oftentimes you’ll make a lot of progress on one path, and then that path will run out of steam. Then you’ll just find a different path and you continue to go on from there.”

The GenAI path that is probably out of steam is the technique of giving an AI model increasingly large amounts of data to make it more intelligent. That method has stopped working. “For the most part, that path is gone,” Garman said.

“However, if you look at all the latest models, they’re doing things like reasoning loops … [where] you ask the models to just actually think about the answer. … They can actually get much better.”

He predicted AI will work more like human employees, maybe even taking a day or a week to iterate and collaborate before delivering results, instead of the instant responses AI chatbots produce today.

For all PYMNTS AI coverage, subscribe to the daily AI Newsletter.

The post AWS, PepsiCo and Other CEOs Tell What’s Needed to Scale AI appeared first on PYMNTS.com.