Like many chief information officers, IBM CIO Matt Lyteson is approaching artificial intelligence both carefully and optimistically. In a recent online interview, he observed that no CIO can afford to switch focus on every new “it” thing, yet AI is certainly anything but a passing fad. “My approach to AI includes traditional AI, generative AI, agentic AI, and business automation — using any and all of these tools in concert to help transform IBM’s business into a digital operating model,” Lyteson explains. “While there are some incredible new AI capabilities, one thing hasn’t changed — it’s best to take a holistic view rather than simply deploy another tool.”
Lyteson notes that generative AI in particular has the potential to impact every business area. “Out of the box it can often be a blunt tool — a sledgehammer for the work of a chisel.” To gain full value out of generative AI, he believes that new adopters should focus on the specific business areas that are ripe for transformation.
An Initial Project
Lyteson’s first AI project was deploying generative AI, trained on the firm’s own data, to the entire employee population, and doing so in a secure, trusted way. “We already had a number of AI-powered assistants and automations, and users were familiar with these for specific tasks or functions,” he explains. While relatively simple and uncomplicated, the new project represented a significant step forward, providing a general, enterprise-wide solution with context-specific generative capabilities.
Matt Lyteson
The project helped Lyteson move forward on three basic goals. “First, we wanted to give IBMers a space to play — a safe, trusted, and secure environment to use and experiment with AI on our own business data,” he says. The approach was designed to minimize risk. It also helped Lyteson equip employees with a user interface that worked specifically for them. “This helped build on our cultural transformation at IBM, pointing employees toward how AI can augment their work, and signaling our expectation for this to happen day-to-day in every corner of the business.”
The approach, Lyteson notes, also established an important reference point for future expansion. “We’ve introduced additional assistants and agents since then [yet] it remains the single-entry point for employees looking to access the continually expanding knowledge and capabilities we’re rolling into the back-end technology.”
Effective Planning
Successfully implementing AI demands thoughtful intentionality at the outset, Lyteson says. “Our plan began with getting key stakeholders around a table and zeroing-in on crystal clear objectives,” he explains. “We didn’t allow for mission creep, and we also didn’t dive into complex environments with no specific plan.”
Lyteson says project planning began in November 2023 and was completed in less than 90 days, resulting in the first iteration’s release. “Because the capabilities of the technology keep evolving, as does our acumen, we continue to iterate based on feedback and solving specific needs,” he notes. “For example, routing between different assistants.”
Expectations Met
Lyteson says that in many respects the project was about showcasing the incredible potential of generative AI to the full breadth of IBM’s employee — more than 200,000 team members doing all sorts of work worldwide. “That was a key first step to having AI be an enabler, and we have now deployed more and more AI-powered agents and business automations across the enterprise.”
“Thanks to an upfront commitment to intentionality, our project met initial expectations, and we’ve been able to continually learn and iterate since then,” Lyteson says. “A CIO’s work is a team sport, especially at large companies.”
Lyteson notes that it was also interesting to see how generative AI often spits out nondeterministic answers when some business functions absolutely require deterministic responses. “This was a key learning for us,” he says. “As a result, we took steps to help our various user bases understand [the concern] — proactively communicating, offering education sessions, and most importantly embedding the right messaging within the user interface itself.”
A Final Thought
A single project such as this one should be table stakes for CIOs, Lyteson says. “Learning to initiate AI projects, experiment rapidly, and capitalize on data-backed learnings is key to ongoing success,” he observes. “This was a nice example of all of those elements rolled into a single package.”