CIOs still grapple with what gen AI can do for the enterprise

A year since ChatGPT burst on the scene, and despite demonstrable benefits, IT leaders have yet to go all in on gen AI, choosing instead to proceed with caution.

Most CIOs have begun exploring generative AI to make sure they stay relevant. But many are finding that the technology on the market doesn’t yet live up to the hype. “After experimenting with both GitHub copilot and ChatGPT for over six months, I’m amazed by the pace at which generative AI is evolving,” says Yves Caseau, global CIO of Michelin. “But in its current state, it’s just a toolbox.”

There’s indeed a lot of hype around the latest wave of large language models (LLM) and associated tools, yet beneath the noise, there’s a whisper about how the technology will one day become indispensable. “Once it’s matured, generative AI will perform many of our mundane tasks — and this will free us to focus on new things,” says Caseau.

Yves Caseau, global CIO, Michelin

Michelin

Some technology leaders, including Patrick Thompson, former chief information and digital transformation officer of Albemarle, go so far to say that generative AI will become the most disruptive technology in our lifetimes. “It will be more disruptive than what Apple did with the iPhone for consumers,” says Thompson. “And for business users, it will surpass what Microsoft did for workforce productivity.”

The big question is what to do with it now.

A boost to traditional AI

While generative AI is new, AI is not. One of the first use cases of artificial intelligence in many companies, including both Michelin and Albemarle, was predictive maintenance, which at its most basic level is an algorithm trained on data collected by sensors. Once trained, the model looks for indicators that have led to failures and alerts human operators, who can then prevent manufacturing outages.

One common shortcoming of the basic setup of predictive maintenance is that rare events are underrepresented in the training data. As a result, the algorithm might not learn enough about the patterns in sensor output that, while infrequent, may forebode failure. To fill the gap, many companies complement the real data with synthetic data.

AI is being used in other ways in the enterprise as well, to do things like improve the efficiency of the supply chain, facilitate customer interactions, and help employees perform office tasks. Albemarle has been using AI as a virtual assistant since the recent pandemic lockdowns. “We were a little ahead of the game, mainly out of necessity,” says Thompson. “The pandemic forced us to find ways of self-servicing 7,000 employees at home.”

The self-service chatbot developed at Albemarle evolved into a tool to help with other corporate functions, which then developed into a virtual personal assistant that manages federated workflows, making it easier for employees to work with several systems at once without having to log into all of them. An employee, for instance, can participate in workflows and make inquiries by just communicating with the bot using natural language, and the bot interfaces with the enterprise business systems.

Patrick Thompson, former CIO and digital transformation officer at Albemarle

Albemarle

But in a few short months, generative AI is beginning to take traditional AI to another level for applications like predictive maintenance. “Interactions become more conversational so you can ask questions and get different insights about the state of equipment,” says Thompson. “It can be used to curate internal and external industry data that’s then used to train traditional algorithms to deliver agile results.”

Moreover, generative AI offers an entry point for companies in sectors yet to use traditional AI. Sectors, such as finance, where most companies began developing data platforms years ago to use with analytical tools, are now experimenting with the newest AI technology using the same platforms.

“Generative AI can also be used to parse publicly available data on markets and companies to help make investment decisions,” says Chris Herringshaw, global CIO of Janus Henderson, the British-American global asset management group. “Rather than spend a lot of time manually researching all of that information, we want to use generative AI to summarize what’s out there, tell us where the signal in the noise is, and suggest areas for us to look into.”

The challenges and rewards of early adoption

Aside from a lack of maturity of the underlying technology, several other obstacles need to be overcome before enterprises further embrace generative AI. The first challenge is the lack of skills both in-house and among vendors that sell traditional applications.

The lack of in-house expertise affects the build versus buy decision every IT leader has to make. “‘Buy’ certainly gets you up the curve faster,” says Herringshaw. “You don’t need to figure out how to productize it, scale it, and support the underlying infrastructure. And prices are so low now that it costs very little to do exploratory work.”

Vendors are keeping the prices down to encourage adoption. But over time, companies will start putting more data into the models, which locks them in with a vendor, and they’ll start creating offshoots that are specialized in certain areas. Instead of using the general version of ChatGPT, for example, they’ll use versions for specific industries, like financial services.

“Once you have different models tailored for different use cases, you wind up with several versions running at the same time, which multiplies the subscription price,” says Herringshaw. “Our hope is that business revenue will scale with the cost. If we really find a way to revolutionize our investment process, the return should more than outweigh the cost.”

Chris Herringshaw, global CIO, Janus Henderson

Janus Henderson

For the short term, subscribing to cloud-based models is less expensive than building in-house — and that may even hold true over the long term. Another advantage of buying is it makes adoption quicker and easier. But over the long-term, building in-house may be the better option for organizations that need to tailor models to their industry, or those that want to push the AI out to the edge and run inference on devices that aren’t connected to cloud-based services.

For the time being, though, very few enterprises have the skilled staff to build an AI model or tweak an existing one. Most companies don’t even have the expertise to be good users. To get the most out of what you buy, you need to first curate your enterprise data to train the models, and then during the inference phase, ask it questions in the right way. Above all, you need to know when to doubt the model.

While generative AI will probably increase the value a company can extract from data, and ultimately change the way businesses are run, it’ll also increase the gap between the digital savvy companies and the digital laggards. So regardless of whether organizations choose to build or buy, they should start developing some level of in-house expertise. “We’re starting to put together formal training to improve how we use the technology,” adds Herringshaw. “The first thing we want to get better at is asking questions.”

Not only is the lack of skills affecting how people use the models, but it’s also affecting the quality of third-party products, which often claim to include AI algorithms. CIOs who are buying the latest version of an enterprise application should check this claim, because there’s still confusion among the traditional application vendors as to how to integrate generative AI.

“Traditional technology vendors are partnering with companies that develop generative AI to deliver virtual assistants that unlock the value of enterprise business systems,” says Thompson, who sits on the advisory boards of several application vendors. “They’re having to balance security and data privacy with speed of delivering on the generative AI value promise.”

While many of the organizations that are now experimenting with generative AI are large enough to have the resources to investigate new things, use of this technology doesn’t have to be limited to big enterprises.

“If you get your governance, security, and your data ingestion right, generative AI can help scale a small company into a big company — and a lean one,” says Thompson. “My prediction is generative AI will be the most disruptive innovation in business. It’ll help consolidate, optimize, and integrate industries, which will result in new industry performance benchmarks that raise the bar and create greater shareholder value. Companies that don’t embrace generative AI will become obsolete.”

SUBSCRIBE TO OUR NEWSLETTER