Generative AI efforts at Walmart last year centered around discovery, David Glick, Walmart’s SVP, Enterprise Business Services, told CIO Dive. But this year it’s all about execution.
Most enterprises can relate.
The arrival of ChatGPT spurred a feeding frenzy, with businesses and consumers rushing to use the technology. But most organizations struggled to move past the generative AI experimentation stage, running into implementation obstacles that slowed progress toward achieving scalable, quantifiable benefits for the business.
More than 2 in 5 executives say their organizations are piloting generative AI, according to an October Gartner survey. Only 10% of organizations are reaching the production stage.
Enterprises are investing heavily in employee reskilling, training, restructuring infrastructure, solution selection, experimentation and pilots tied to generative AI. Now, the pressure is on to bring ROI.
Walmart took a “diverge, then converge” approach to generative AI, Glick said. By first surveying the landscape and getting a pulse on a plethora of possible use cases, leaders then turned their attention to picking a few for a deeper dive.
The retailer launched a generative AI-powered assistant for employees to use through its internal app in August. Since its launch, My Assistant has helped employees draft content, eased information gathering for internal help desks and sped up the process for creating financial documents for quarterly reports, Glick told CIO Dive.
“We’ve seen associates on their own use it for this kind of thing, and now we’re trying to … scale that across the organization, taking these ideas that we’ve crowdsourced and scaling up,” Glick said.
In the year ahead, the company will expand employee access to My Assistant, taking the tool global to 10 more countries, Walmart announced earlier this week.
“What we’ll see both at Walmart and in the industry is people will pick a few things and go deep on those rather than just sort of talking about lots of different things,” Glick said.
Identifying use cases to scale
Most enterprises approached generative AI experimentation with a wide net, creating dedicated space for employees to explore where generative AI could benefit them the most.
Having employees solve their own problems is a big benefit of generative AI tools, according to Vincent Yates, chief data scientist at Credera and founder of the AI Global Council.
It does, however, create a roadblock.
“How do you actually give them the guardrails to have this thing scaled to the size it needs to scale, have the right measures in place to ensure security, privacy, data consistency and all of the other things you need to do as a CIO?” Yates said. “That will be the thing that they have to start wrestling with next year because I suspect every organization will have an explosion of these types of applications.”
As they enter the next stage of adoption, CIOs will establish clear goals and metrics to track indicators of success.
They’re not alone. CEOs are tasking a majority of leaders from across the business with identifying the return on investment of implementing generative AI, according to an Insight Enterprise survey of more than 600 decision-makers.
Part of the recipe for making any implementation project or initiative successful is knowing the end goal. It’s up to leadership to determine what kind of ROI the business needs from generative AI.
Leaders are focusing on measuring customer satisfaction gains, enhancing service availability, reducing operational costs and improving productivity, according to an Insight Enterprise survey.
“What I’m really focused on internally at Insight is, how do I take something that maybe took two days to do and now can be done in five minutes and add that into the workflow,” said David McCurdy, chief enterprise architect and CTO at Insight Enterprises.
Once businesses identify what they want to come from generative AI adoption, leaders can select solutions and identify use cases with those goals as a guiding light.
Glick said Walmart always starts evaluations with the user experience in mind.
“It has to pay dividends and has to make financial sense, but starting with the experience that our associates get on a day-in, day-out basis and how we can make that better is where I always start,” Glick said. “Then we can say, ‘What’s our ROI on this? Is there a cost-benefit analysis? Is this going to save us money, make us more money, or allow us to provide higher quality services to both associates and customers?’”
From strategy to action
Since generative AI tools hit the scene in late 2022, people have used these tools for research, coding, image generation and more.
The most common way employees access generative AI is through chatbot-style tools and many organizations are deploying conversational tools across their workforce.
PwC laid out plans for a broad internal rollout of the company’s generative AI tool ChatPwC in August. JLL gave its 103,000-member workforce access to its proprietary generative AI tool, JLL GPT, in the summer as well.
P&G rolled out a generative AI chatbot-style tool chatPG to employees in September. And Deloitte began giving 75,000 employees access to PairD, an internal generative AI platform, for coding and researching in January.
Zillow, another enterprise early and eager to embrace AI, experimented with off-the-shelf solutions and built capabilities in-house early last year. The experiments resulted in the broad implementation of an internal conversational generative AI tool using OpenAI’s technology, which is open to all employees, and a code generation tool for its developers.
“As a leader, one of the top things that I’m thinking of is how do we improve our developer experience and improve our productivity,” Lakshmi Dixit, VP of tech engineering and operations at Zillow. “There’s a lot of tools that are coming out, and our engineering team is actively exploring, but we also want to make sure it seamlessly integrates with our Zillow ecosystem.”
Zillow’s ZG Chat is open to all employees with use cases ranging from task prioritization to meeting scheduling and writing content. The code-generation tool has an acceptance rate close to 50%, according to Dixit.
“The key thing for us is the ability to scan our code repository, so we are definitely looking at what works best for our unique use case and how do we make sure that our employees or our developers find the most value,” Dixit said.
