JPMorganChase, NFL, Georgia-Pacific share GenAI ups, downs

The ROI in AI

JPMorganChase expects a $2 billion return on its AI investments in the coming years. The bank's work to date includes developer tools for code generation, LLM Suite and IndexGPT. LLM Suite is a virtual assistant available across business lines to help employees write, summarize documents and generate ideas. IndexGPT is an OpenAI GPT-4-powered tool that formulates thematic investment baskets for institutional investors.

JPMorganChase is investigating agentic architectures, which are characterized by autonomous agents strung together to perform more complicated tasks. Experts describe this architecture as a paradigm shift in AI and software design.

"I think it's just going to blow up how we all think about FinOps," said Lori Beer, the bank's global CIO, in an interview. "It's going to be more complicated."

In the meantime, the bank is preparing for AI's IT demands. The institution plans to turn on new data center infrastructure at the end of next year to support traditional IT growth predicted two years ago. Now the bank is looking four years out to support generative AI (GenAI), which needs new chips, more power and liquid cooling.

"We have had to extend the timeline in which we think about planning for demand," Beer said.

Also, the bank has to recalculate what it runs on-premises and in the public cloud as it decides where to do model inferencing and training while balancing price, performance and capacity, Beer said.

The Manchester Airport Group (MAG) operates three airports, making it Britain's largest homegrown airport operator. The company serves over 61 million passengers and contributes billions of dollars to the U.K. economy annually.

MAG plans to grow its business with AI, starting with the multilingual virtual assistant Virtual Aiport Liaison (VAL), built using models on AWS. VAL provides immediate answers to travelers' questions through kiosks that went live across the company's three airports in November.

Since then, travelers' questions have provided information that could help MAG grow its business. For example, at London Stansted Airport, many travelers asked for the fastest and cheapest route to London, allowing MAG to sell single and two-way tickets one day.

MAG is already working on a mobile version of VAL that could help MAG get into the business of offering even broader travel services and in-app purchases, said Ryan Cant, MAG's chief digital officer, during a presentation.

"We see this as a really strategic tool whereby our airport customers can start to interact with us in a mobile way," Cant said. "VAL becomes a really important hook."

AI's benefits bring challenges

National Football League teams use AI-powered software to process game data on AWS, analyze opponent strategies, and make real-time decisions during games. The data also feeds chatbots and virtual assistants on mobile apps, giving fans immediate information on games and players.

But as the league expands internationally, the NFL must re-evaluate its cloud dependency, NFL deputy CIO Aaron Amendolia said. That's because the NFL can't control the technology available within foreign stadiums the way it does in the U.S.

"We're going to decide what we can do on-premises and what we can do with the cloud," Amendolia said in an interview. "If we're playing in a new region and can only trust the local providers to connect us, then we have to make sure we have enough access [to data] within the stadium to run the basics."

Satellite connections, for example, could provide a backup to what is available in the local stadium. However, having more AI processing within the facility could be the best option to avoid delays in delivering data, Amendolia said.

The league played a total of five games this year in three countries: Brazil, Britain and Germany. Next year, the NFL plans to do eight international games, Amendolia said.

Georgia-Pacific, one of the world's largest manufacturers, has accumulated vast amounts of data on AWS since becoming a customer in 2016.

The maker of tissues, paper, packaging and building products runs 30,000 machine-learning models on AWS to provide predictive analytics for machinery on factory floors in 150 locations worldwide. The models process data from 90,000 vibration sensors to track machinery's overall condition and determine whether maintenance is required.

Maintaining low latency in cloud connections is the biggest headache for any project, said Carter Smith, GP's GenAI product leader, in an interview.

"Latency sucks," she said. "It's a challenge because we have a lot of data."

To reduce latency, GP has experimented with different models, streamlined the process of making queries to the cloud, and reduced the number of queries as much as possible.

"It's been a process," Smith said. "When we first started, latency was terrible. I mean, it was terrible."

The early cloud projects weren't critical, so latency was less of an issue than it is now as GP's newer AI initiatives scale. Carter expects it to continue to be an ongoing problem that will be addressed in multiple ways.

"It's an evolution, and I don't think there's one answer," she said.

Challenges aside, the organizations interviewed are confident that GenAI will significantly enhance the value they've received from machine learning and other traditional forms of AI.

For example, JPMorganChase's cybersecurity teams want to use GenAI to map architectural diagrams of IT systems to threat models. This would speed up their understanding of the paths cybercriminals could take to access computer systems.

"That's an example of how GenAI, we believe, will help transform the effectiveness of some of our cybersecurity team members," Beer said.

Those types of advancements would have a significant return.

"I think the higher value stuff is what's ahead," she said.

Antone Gonsalves is an editor at large for TechTarget Editorial, reporting on industry trends critical to enterprise tech buyers. He has worked in tech journalism for 25 years and is based in San Francisco. Have a news tip? Please drop him an email.