Beyond the hype: 4 use cases that show what’s actually working with gen AI

If 2023 was the year of experimentation with gen AI, 2024 was when companies zeroed in on use cases and started putting pilot projects into production.

In a survey of 2,300 IT decision makers that IBM released in December, 47% say they’re already seeing ROI from their AI investments, and 33% say they’re breaking even on AI. Only 14% say they’re losing money, and 66% of companies plan to increase their AI investments compared to 5% that plan to decrease it.

The early part of 2024 was disappointing when it comes to ROI, says Traci Gusher, data and analytics leader at EY Americas. “But now we’re actually starting to see real benefits,” she says. “Some of the best use cases I’ve seen have been in marketing, and that’s just one area.”

According to experts and other survey findings, in addition to sales and marketing, other top use cases include productivity, software development, and customer service.

Use case 1: productivity

Mike Baker is CITO at PGIM, a $1.4 trillion asset management firm, formerly known as Prudential Investment Management. Baker says productivity is one of the main areas of gen AI deployment for the company, which is now available through Office 365, and allows employees to do such tasks as summarize emails, or help with PowerPoint and Excel documents.

This is a use case that’s been rolled out widely, he says, though not all tools are available to all employees. Registered investment advisors, for example, have to jump over a few hurdles when deploying new technologies.

“Part of it has to do with things like making sure we’re able to collect compliance requirements around AI,” says Baker. And then there are guardrail considerations. Gen AI is still in its early days and the company is concerned about safely integrating the technology. “Today, we’re not using gen AI in any investment decision, trading decisions, or even back office areas without a human in the loop,” he adds. “We’re taking that part very slowly.”

But employees who are able to use gen AI do take advantage of it, he says. “Once you get Copilot for Office 365, you go through training, and that’s driven up our utilization to around 93%.” Plus, according to a recent survey of 2,500 senior leaders of global enterprises conducted by Google Cloud and National Research Group, 34% say they’re already seeing ROI for individual productivity gen AI use cases, and 33% expect to see ROI within the next year.

Another organization using Microsoft Copilot for productivity is Oral Roberts University in Tulsa, Oklahoma.

“It’s accelerating the learning process, improving research, and helping students with assessments,” says Mike Matthews, the university’s VP for innovation and technology. “I’ve not met many professors who don’t feel overworked and exhausted,” he says. “To be able to have the opportunity to do research and grade faster, and to assess students faster, it’s all perfect for them.” For example, a faculty member might want to teach a new section of a course. Copilot can create an outline in seconds that the faculty member can use as a starting point, he says.

As a result of using AI for productivity, marketing, and to help process applicant transcripts, says Matthews, the time it takes to respond to applicants has fallen from weeks to hours, the number of leads from new countries has increased by 267%, and enrollment has grown by nearly 11%.

Elliott Franklin, CISO at Fortitude Re, a global reinsurance company, says his firm is also using enterprise subscriptions to ChatGPT and Copilot to integrate gen AI into operations.

“With these paid versions, our data remains secure within our own tenant,” he says. The tools are used to extract information from large documents, to help create presentations, and to summarize lengthy reports and compared documents to find discrepancies. “We save time, minimize human error, and allow teams to dedicate attention to higher-value tasks,” he adds.

Use case 2: software development

PGIM also uses gen AI for code generation, specifically using Github Copilot. This use case is in full production, says Baker, with about 1,000 developers using it.

“We see about 60% of our developers using it on a day-to-day basis,” he says. “And about 70% of the code that’s recommended by Copilot we actually adopt. So a pretty high adoption rate for AI code generation.”

According to the survey by Google Cloud and National Research Group, 28% of leaders report positive ROI for gen AI in developer productivity and engineering, with another 34% expecting to see ROI within a year.

At Emburse, an expense reimbursement company, software developers use code generation tools like Github Copilot and Amazon Q Developer, which integrate directly into developer environments, says Ken Ringdahl, the company’s CTO. “They’re used to automate code generation, improve code structure, and even identify potential defects.”

Software development was also the area where financial services firms see highest productivity improvements, according to a 2024 survey by Bain & Company.

EY’s Gusher says she’s seeing gen AI value in code debugging and testing. “We’ve also seen some significant benefits in leveraging it for productivity in data engineering processes, such as generating data pipelines in a more efficient way.”

Use case 3: sales and marketing

Gen AI is further used at PGIM to help its sales staff help users interact with documents in a more user-friendly way.

“Our salespeople go out to meet with financial advisors,” says Baker. “We have a ton of documents we can talk about. With AI, they’re able to get the right information at the right time for business conversations, which was somewhat hard and laborious to get in the past.”

And that’s just the start of it. They also use AI to help with website content, speech writing, and client communications. “Marketing communications is a great area for AI,” he says.

And then there’s taking requests for proposals.

“We get a lot of RFPs from a lot of clients,” he adds. “We use AI to generate the first draft of the response to the RFP by using past RFPs and other data sets.” So gen AI can help the firm respond to more RFPs, he says, or respond to RFPs that employees wouldn’t take on in the past because they were too complicated, or the deals were too small.

In the Google Cloud and National Research Group survey, 33% of leaders report positive ROI for gen AI in sales and marketing, with another 30% expecting ROI within a year.

Gartner analyst Arun Chandrasekaran says there are several business functions like marketing that show success, and EY’s Gusher confirms that marketing is one of the biggest areas for gen AI value.

“The most common pattern I’m seeing is custom-building capabilities and leveraging other systems for data,” she says. For example, companies can use data from their CRM systems to get data to create personalized communications. “To get to ROI requires data from several systems,” she adds. “The real value comes from enabling the entire process.”

That’s exactly what SS&C, a financial services and healthcare technology company, is doing with gen AI. “We generate customer communications based on core data in our systems,” says Brian Halpin, the company’s senior managing director of automation. “It’s live, it’s real, and it’s working at scale.”

Use case 4: customer service

Another top use case for gen AI is in customer service, including omni-channel support and customer call centers. For example, Verizon’s customer call centers receive hundreds of millions of calls a year. Another two billion customer interactions take place over digital channels. So there’s a lot of opportunity for gen AI to have a business impact, and the company is already seeing benefits, says Brian Higgins, Verizon’s chief customer experience officer.

Verizon began working on developing gen AI use cases a couple of years ago, starting with Google Bard, and moving up to Gemini 1 and then Gemini 1.5, using RAG to provide the model with relevant information.

The gen AI is used to quickly summarize complex documents, listen to conversations and automatically pull up relevant information, answer questions, and provide other functionality. Take for example the average 18-minute conversation a customer might have with a Verizon representative.

“We can shave off about a minute on every call,” says Higgins. “When we think about the number of calls coming in, that has a material impact.”

And the amount of time a customer has to wait before they can talk to someone has come down as well, he says.

“We don’t release financial ROI numbers,” he says, but Verizon does have internal measurements in place. For example, customers are surveyed right after a call to determine if their interaction was positive. “All of those measurements have increased,” he adds.

Today, all customer service representatives use the gen AI tool, which is over 40,000 people. The fine-tuning and contextualization that Verizon adds to the base model means that accuracy is currently over 90%, says Higgins. Still, it’s not perfect, and Verizon has been cautious about how the technology is rolled out, says Vivek Gurumurthy, Verizon’s consumer CIO.

“Today, we use it to primarily explain what happened, but we’re not in the business yet of taking actions on behalf of the customers,” he says. “For actions, we want to make sure there’s a human involved. And for the explanation, we want to be very careful and validate, verify, and certify everything.”

In addition, some of the more powerful tools, like the research assistant that actively listens to calls and provides real-time information and insights, are currently only available to agents, not directly to customers.

“The personal research assistant is only for Verizon employees,” says Higgins. “We’re still taking the time — and getting everything through our AI council — before we start expanding it to our digital footprint as well.”

The public-facing chatbot only answers low-risk queries, like what the latest offer is. “Most of what’s on the digital side is what’s on the web pages,” he says. “And if you want to add a line or upgrade a device, we do a warm transfer over to a human.”

According to the EY report, 72% of business leaders report positive ROI for gen AI in customer satisfaction, and that number goes up to 75% for companies that spend more than 5% of their budget on AI. And in the Google Cloud and National Research Group survey, 34% of leaders report positive ROI for gen AI in customer service, with another 33% expecting to see ROI within a year. That makes customer services tied for first place with individual productivity when it comes to ROI.

Another thing that makes customer service a particularly good area for gen AI deployment is companies tend to have a solid sense of their KPIs, says Ricardo Madan, SVP at TEKsystems, a technology consulting company. Companies know what the good scores are for particular contact center functions. “You’ve got that down to a science,” he says. Knowing how to define success is a big advantage, too. “And we’re seeing real results in seeing resolution time and tickets go down, and call volumes shrink.”

Choosing deployment strategies

There are many ways to roll out gen AI at a company, all requiring varying degrees of investment and effort. A company could give employees access to a secure version of ChatGPT, or enable AI features in the tools they’re already using. Or it could build its own models from scratch, or develop its own agentic AI platform instead of waiting for a big vendor to come out with one. At PGIM, for instance, it all depends on what differentiates it as a company, and where investments would yield the most value.

“We’re not going to create our own coding LLM,” says PGIM’s Baker. “We’ll use Github for that. For marketing, we use Jasper, a platform that’s specifically built for communications and marketing. But in other areas, like our document Q&A, a lot of data is specific to the materials, and we want to customize the prompts and fine tune a model.”