Inside AT&T's Copilot Rollout: How 18,000 Employees Adopted AI in Six Weeks

Most organizations investing in AI are spending significant money and not seeing a return. Our latest Prosci research across more than 1,500 workers found that motivation to use AI is nearly identical in organizations that are succeeding and those that are stalling. People want to use these tools. What separates results from failure is how organizations prepare and enable their workforce. Last month, I sat down with Stephanie Harvey, Principal Technology Strategist at AT&T, and Patrick Martin, Senior Customer Success Account Manager at Microsoft, for a panel-style webinar on this topic. AT&T had successfully scaled Microsoft 365 Copilot to more than 18,000 active users in six weeks, reaching a 96.4% sustained adoption rate among assigned users. Nearly 2,000 people attended the webinar, with lots of questions, some of which we were able to address on the call. If you want to listen to the full conversation, the recording is available on our website. I’ve also captured below the most frequently asked questions with more depth than we had time for on the call. How do you get a senior leader to care about change management without overwhelming them with the methodology? Stop leading with the methodology. Executives are focused on outcomes, and the most effective way to earn their attention and commitment is to frame the conversation around risk to their goals, not around the change management process you're planning to use. AT&T's team leveraged the Prosci Risk assessment to identify areas of risk and leveraged those insights in discussions with executives. Leaders seeing the cost of getting it wrong is more effective to drive alignment than being educated on methodology. Stephanie put it plainly: "Help them understand the risk of doing nothing and show them the specific outcomes their business unit needs. Make the sponsorship feel like a leadership role, not just an ask." A few practical notes from what AT&T did: bring options to that conversation, not a finished plan. Leaders who help shape the approach own it differently than those who are asked to approve it. How do you define adoption, and how do you know when it's real? AT&T’s initial measure was at least one active Copilot use within a 30-day window, which they monitored daily. That definition gave the team something specific to manage, and daily monitoring enabled them to intervene quickly when usage dropped, rather than discovering the problem weeks later. We received several questions asking whether active usage is really adoption or just a usage indicator, which is a fair challenge. Real adoption shows up as sustained behavior change. Usage that builds over time, spreads across the tool's capabilities, and holds without constant reinforcement. Stephanie was clear that active usage was the floor, not the ceiling, and her team used surveys and focus groups to gather more usage information. How did you handle employee resistance, especially fear of job displacement? AT&T saw minimal broad-based resistance, but job displacement fears and other concerns about the environment and the ethical use of AI surfaced in pockets. For any organization, especially one as large as AT&T, the best approach to managing these concerns is to route them to the team lead or manager. When employees raised questions the change team wasn't positioned to answer on the company's behalf, Stephanie's team pointed them to the company's official stance on AI and gave supervisors specific talk tracks to use with their teams. "With change management, you've got to be careful as a practitioner," Stephanie said. "Not everything is yours to solve." One of the key messages delivered during this rollout was that workers who build AI proficiency now will be more marketable than those who don't, which did contribute to an increase in training attendance. How do you build the ROI story when you don't have the numbers yet? Most organizations make one or two mistakes here: they either wait too long to start building the story, or they manufacture numbers too early to satisfy executive pressure. Both create credibility problems. Stephanie set honest expectations upfront, then built the methodology before she had the results. In the early months, the team ran surveys and focus groups, asked employees to estimate time saved on specific tasks before and after Copilot, applied AT&T's loaded labor rate to translate that into financial terms, and cross-referenced everything against Microsoft's usage dashboard. The result was a conservative number with a clear, auditable methodology. "Leaders trust you if they can understand how you got there," Stephanie said, "far more than they trust some big number without a story behind it." What makes an AI rollout structurally different from other technology changes? Several things make an AI rollout structurally different from other technology changes, and underestimating them is where most organizations get into trouble. First, the nature of the change itself is different. Most technology rollouts ask people to learn a new system or process with defined inputs and outputs. AI tools, especially generative AI, ask people to develop judgment about how and when to use them, how to evaluate the outputs, and how to integrate them into work that often doesn't have a clear before-and-after. That's a fundamentally different kind of learning. Prompt engineering is a skill that develops over time through practice and experimentation, and needs vary significantly by role. AT&T ran more than 200 live training sessions, with dedicated sessions on prompt engineering, because one-size training wasn't sufficient. Second, the emotional stakes are higher. Prosci's research found that organizational change management, specifically resistance to change and user adoption challenges, is the single largest barrier to enterprise AI adoption, representing nearly a third of all executive-reported challenges. Job displacement fears are real and widespread. They don't disappear because leadership issues a reassuring message. They require sustained, honest communication at the right level, with the right people, over time. AT&T saw this directly. Addressing it required routing those conversations to supervisors with specific talk tracks, not managing it from the change team. Third, the pace of change won't slow down. Whatever playbook your organization builds for one AI tool will need to flex as the tools evolve. That's exactly why methodology matters more than platform familiarity. A team that understands how to move people through Awareness, Desire, Knowledge, Ability, and Reinforcement can apply that framework to whatever comes next. A team that built a Copilot-specific rollout plan will be rebuilding it in a matter of months. Is it too late to introduce structure if your organization has already started informally? No. But the approach changes depending on where you are. If AI use is scattered and informal, the first priority is understanding what's actually happening. Which groups are using which tools, for what, and with what results? That picture shows you where the highest-value opportunities are and where the biggest risks are quietly accumulating. From there, the work is largely the same as it was at the start: build executive alignment, map your personas, address Awareness and Desire before pushing more Knowledge and Ability, and put structural reinforcement in place so adoption doesn't depend on individual motivation. The difference is that you're working with an existing reality rather than a blank slate, and that reality includes employees who may have formed habits, positive or negative, around AI tools already. One thing you should avoid is launching a formal program that feels like a correction of what employees were already doing, which will likely encounter resistance. The better framing is that the organization is now investing in making what's already started more effective, more supported, and more connected to business outcomes—which is true. The structure exists to make the work better and communicating it that way is what brings people along rather than putting them on the defensive. It is never too late to introduce the structure. Your workforce is ready What our latest research shows, and what AT&T demonstrated, is that business results from enterprise AI initiatives come from the model you build around organizational change, and a people-focused approach is critical to drive outcomes from AI investments. If you're facing these same questions in your own organization, a Prosci consultant can help you apply this thinking to your specific rollout.

Reinforcement is the final building block of the Prosci ADKAR® Model, a simple but effective model for individual change made up of five building blocks of change: Awareness Desire Knowledge Ability Reinforcement® It is a natural tendency to revert back to what we know. In fact, there is research emerging about how the brain functions that suggests it is not just a natural tendency but a physiological tendency. While making a change is difficult, sustaining a change can be even more difficult. This is why reinforcement is such a critical component of successful change. It encompasses the mechanisms and approaches so that the new way stays in place. Taking the Time for Reinforcement One of the reasons reinforcement can be difficult, from a change management perspective, is that once a change is finished, we are often already moving on to the next change. Acknowledge this fact, and the tendency to simply "move on" once a change goes live, and build the necessary mechanisms to reinforce a change. It takes concerted effort and time to make sure a change stays in place. Given the scarce resources and change saturation that many organizations face, reinforcement efforts can often fall short. As the final building block of successful change, the focus on reinforcement needs to remain strong so that changes are sustained and deliver the expected results over time. You only know if a change was successful if you take a step back afterward and see if employees are actually doing their jobs differently. Building Reinforcement Successful reinforcement may require: Celebrations Rewards and recognition Feedback Corrective actions Visible performance measurement Accountability mechanisms Potential challenges and resisting factors Rewards not meaningful or not associated with achievement Absence of reinforcement for accomplishments Negative consequences including peer pressure for desired behavior Incentives that directly oppose the change Tactics for fostering reinforcement Publicly visible performance scoreboards that positively show compliance to a new process Feedback from supervisors directly to employees, including saying "Thank you" Visible recognition by senior level sponsors Project-sponsored celebrations for employees Compensation and appraisal systems designed to support the change Most organizations make a significant investment in helping people build the awareness, desire, knowledge, and ability to make a change successfully. But without understanding and proactively planning to reinforce a change after it has happened, you run the risk of that investment being wasted and the expected results from the change not fully being realized.