How Are AI Super-Spreaders Changing Big Companies?

At Citigroup, thousands of employees are doing AI enablement work outside their formal job descriptions. Many are not engineers. Most do not sit inside a central technology group. They are peers.

Citi’s AI Champions and Accelerators program launched in early 2024. It now includes roughly 4,000 volunteer “accelerators” supported by about 25 to 30 “champions” embedded across business lines according to reporting by Business Insider. Accelerators help colleagues understand and use internal AI tools in their day-to-day work. Champions coordinate activity within their divisions and act as liaisons to Citi’s technology organization. Participation is voluntary, and the program does not impose formal performance targets.

Citi says its proprietary AI tools are available to 182,000 employees across 84 countries. Adoption is above 70 percent, according to chief executive Jane Fraser, who told analysts on a recent earnings call that AI will “reshape how work gets done.” In a January memo to staff, Fraser said she expects a more disciplined and more confident workforce in 2026, explicitly tying that expectation to AI use.

Carey Ryan, chief of staff for Citi’s technology organization and one of the program’s leaders, told Business Insider that a small central team would not be able to reach the entire workforce. The accelerators model is intended to address that reach problem by distributing AI knowledge through peer networks.

Citi is not alone. Large organizations are increasingly framing AI adoption as a workforce-scale change-management effort rather than a simple tooling rollout. Bank of America has reported more than 23 million employee interactions with its internal AI assistants, Ask Merrill and Ask Private Bank, in 2024.

Walmart has rolled out AI tools intended for roughly 1.5 million U.S. associates, saying some pilots reduced shift-planning tasks from about 90 minutes to roughly 30 minutes.

Accenture and Microsoft have also emphasized train-the-trainer approaches that pair central AI platforms with distributed learning.

The common thread is leverage. Rather than hiring thousands of AI specialists or relying solely on centralized centers of excellence, companies are mobilizing existing employees to translate AI into role-specific workflows. The bet is that peers move adoption faster than mandates.

Why Central AI Teams Aren’t Enough

The accelerators model is indicative of how some enterprises are organizing AI. Central AI teams still build tools, set policy, and manage risk. What they struggle to do at scale is context. Job-specific use cases vary across functions, geographies, and regulatory environments. Peer networks are intended to help fill that gap.

At Citi, accelerators meet twice a month to attend demos, learn about new tools, and share what is working in their teams. They also surface friction. Program leaders say feedback from accelerators has led to product changes, including feature notifications and expanded file upload limits. More than 100 internal “Citi AI Days” have been hosted by volunteers who demonstrate tools and answer questions from colleagues.

The emphasis is cultural as much as technical. Josh Goldsmith, AI champion for Citi’s internal audit unit, told Business Insider that adoption accelerates when employees see colleagues using the tools in real work. The legitimacy comes from proximity. Seeing a peer apply AI to a real task lowers the barrier to experimentation.

What is harder to find are hard metrics. Citi does not publicly disclose engagement measures at the accelerator level. One accelerator might have a handful of small interactions. Another might present to hundreds of employees. The company emphasizes adoption rates and tool availability rather than time saved or output gains. That contrasts with organizations like Walmart, which has highlighted specific operational improvements, or Bank of America, which reports interaction volumes.

There is a tradeoff embedded in that choice. Cultural adoption is easier to scale than return-on-investment measurement. It is also easier to defend internally.

AI Adoption Has a Labor Problem

The volunteer model shifts labor in ways companies rarely quantify. Champions at Citi report spending between three and five hours a week on accelerator work in addition to their core responsibilities. Participation is optional, and people can leave if the commitment becomes too heavy. The role is not tied to formal compensation.

Academic research on organizational citizenship behavior shows that voluntary, extra-role work often drifts into expectation. Over time, that can lead to role overload and burnout, particularly when recognition and workload adjustment lag behind responsibility.

The burden is not evenly distributed. Employees with more flexible schedules are more able to volunteer. Frontline or heavily regulated roles have less room. Career visibility can accrue to those who participate, even when the work is unpaid. At Citi, retention among internal audit accelerators is reported at 70 to 80 percent. That suggests commitment, but it also suggests attrition.

Still, none of this invalidates the model. Volunteer AI work is not free. It is subsidized by employee time and goodwill. Companies are making a calculated decision that the speed and reach outweigh the cost.

For now, the approach appears to be spreading. As AI moves from pilot to infrastructure, enterprises are discovering that the hardest part is not building the system. It is getting thousands of people to use it.