AI Implementation: How Citi, Zapier, Microsoft, and Others Did It

Giving everyone a ChatGPT or Microsoft Copilot license does not deliver 40 to 50 percent productivity gains. The companies that get real results treat AI as a people and change challenge, not a software rollout. They redesign workflows, build AI fluency, and spread adoption through trusted peers.

That is the gap between a pilot that fizzles and a transformation that sticks. So rather than rehash general best practices, this overview breaks down how specific companies actually implemented AI: what they built, how they drove adoption, and what you can borrow for your own organization.

This is a living resource, continuously updated by the team behind Lead with AI, the executive-level AI training and community. Last update: July 15, 2026. For submissions, please contact us.

Current cases covered:

• Citi
• Doist
• Zapier
• Hearst
• Apple
• Amazon
• BCG X
• Microsoft HR
• Morgan Stanley
• PwC Netherlands
• Roblox
• ServiceNow

Adobe

Adobe's Head of Specialist GTM Strategy and Practice, Amy Wowak, joined a Lead with AI PRO session to share the six-step playbook she used to roll out AI fluency across Adobe's global sales team. Her framing was blunt: "I can give you a tractor, but that doesn't make you a farmer." AI is a skill, not a tool, and the failures she has seen are organizational, not technical.

• Start From the Work, Not the Tool: Wowak's first move was to audit the three or four moments where AI actually changes a sales outcome, then choose tools to fit that work, rather than dropping a portfolio of apps on the team. This is Workflow Literacy in practice, the prerequisite skill we teach before any tool decision.
• Platform Your Renegades as Champions: She looks for the curious "cowboys" who push to the edge of what AI can do, and they are rarely the top sellers. Her rule, "work with them, don't build on them," is exactly how effective AI champion programs operate, pairing top-down sponsorship with bottom-up energy in what we call the AI Implementation Sandwich.
• Get in the Game and Learn in Public: Wowak insists leaders do not need to be the tip of the spear, but they do need to use the tools and show their own broken, half-built agents. Leaders who model that build the credibility and psychological safety that turn quiet, shadow AI users into an open culture, and it starts with the leader's own AI fluency.
• Build on Clean Data, and Measure With AI: An agent is only as good as the data underneath it, so she starts with the reliable "low-hanging fruit" and runs AI fitness checks twice a year to track fluency instead of firing off more surveys. She even has teams ask their AI to describe their own usage patterns, letting the model surface adoption signals no survey could capture.
• Decide What the Recovered Time Is For: Wowak's sharpest point is that time saved is not time captured. Unless a leader sets a North Star for where reclaimed hours go, the admin of a busy day simply takes them back and the time dividend evaporates. It is the same reason we treat Impact per Hour as the guiding metric of the Co-Pilot Economy, not raw hours saved.

Citi

Citi is one of the strongest enterprise proof points that AI adoption scales through people, not platforms. Over roughly two years, the bank built an internal network of more than 4,000 AI Accelerators, supported by a smaller group of 25 to 30 AI Champions, and reached over 70% adoption of firm-approved AI tools across 182,000 employees in 84 countries. We covered it in our breakdown of how Citi scaled AI adoption socially.

• People Copy People: Champions and Accelerators are embedded inside teams across functions as local guides rather than centralized trainers. They show colleagues how AI supports real tasks, from summarizing documents and drafting internal notes to analyzing datasets and supporting development work. That peer-led transmission lowers the psychological friction that stalls most rollouts, and it makes AI feel practical and relevant.
• Distribute Capability, Do Not Concentrate It: Rather than keeping AI knowledge inside a small expert group, Citi's model puts thousands of employees in a position to support others. For a regulated financial institution operating across dozens of countries, reaching more than 70% adoption is a level many enterprises never hit even with heavy investment.
• Put Champions in the Middle Layer: Champions sit squarely in the middle of the AI Implementation Sandwich, translating strategy into daily practice and surfacing real constraints back to leadership. Credibility, context, and consistent in-work examples matter more than tool access.
• Start With a Critical Mass, Then Scale: Citi's 4,000-person network was built over two years, not launched overnight. The lesson for leaders chasing scale is that adoption falters when organizations focus on tools alone, and succeeds when they build the human systems that help technology spread. Our guide to AI Champion Programs covers how to design one.

Doist

Doist, the 100-person, fully remote company behind Todoist and Twist, is one of the clearest examples of an AI-native organization. In a Lead with AI PRO session, Head of Operations Chase Warrington walked us through Doist OS, the shared AI workspace their whole team uses every day. It became the highest-scoring product in the company's history on their own product-market-fit survey.

• Build the Floor, Not the Ceiling: When Doist opened AI access to everyone, the most technical people raced ahead while the non-technical half fell behind, creating a widening capability gap. The reframe that changed their approach: what shows how leveraged an organization is not the ceiling, it is the median employee. Doist made "company-wide leverage" a formal goal and designed everything, including Doist OS itself, for the people least likely to adopt on their own.
• Turn Your Existing Knowledge Base Into an AI Asset: Doist has been async-first since 2007, so almost everything it knows already lives in writing: an 1,800-page handbook, thousands of Twist threads, project data, and code. Doist OS connects all of it, so any employee can ask questions and take actions across every tool from a single interface. The lesson for more office-based companies is that the same knowledge already exists in email and chat, waiting to be connected.
• Make Advanced Usage Visible: Before building any tool, Doist raised the baseline with monthly AI lightning talks, biweekly office hours, dedicated AI channels, and parallel beginner and pro tracks at company retreats. Leadership committed to discussing adoption in every leadership meeting. This is the same instinct behind structured AI Champion Programs: make peer-to-peer learning visible rather than leaving people to figure it out alone.
• Give the Shared Tool a Shared Identity: Doist OS works because it knows the company, not just the individual user. A personal profile, a "soul" document that encodes company values and communication standards, and an index that knows where information lives are what make it smart rather than merely searchable.
• Keep Humans Accountable for Outputs: As adoption scaled, Doist introduced one clear rule: if you would not put your name to it, do not post it. Peer review handles governance at the structural level, and individual accountability handles quality at the output level. Both are necessary.

Zapier

Zapier's Chief People Officer Brandon Sammut joined a Lead with AI PRO session to break down what separates AI adoption from real AI transformation. The stakes were real: nine months after calling an internal "Code Red" on AI in early 2023, this healthy, well-funded company ran its first-ever layoff, not because of cash flow, but because the talent it had no longer matched the talent it believed it needed going forward.

• Separate Adoption From Transformation: Adoption produces individual productivity gains of 20 to 40 percent and spreads grassroots. Transformation is a different order of magnitude, 3x to 10x, and it requires redesigning the work itself, more sophisticated builds, and real change management. Confusing the two is where most AI strategies stall.
• Redesign the Work, Not Just the Speed: Zapier's customer support team is the clearest proof. Its leader committed publicly to cutting average handle time in half within 12 months while holding customer satisfaction, and the team hit both targets. What made it work was not the technology. They rebuilt the staffing model and the job descriptions around AI, and employee engagement scores actually rose 20 to 30 points in the process.
• Choose the Transformation Lead for Credibility, Not Title: An executive sponsor is non-negotiable, and the CEO's voice is irreplaceable but not sufficient on its own. Zapier put a head of learning and development in charge day to day, chosen less for the title and more for a deep read on human behavior and the trust colleagues already placed in them.
• Run Hackathons Where Leaders Build in Public First: The single most effective tactic for building a culture of experimentation is dedicated, structured time where senior leaders demo their own unfinished, imperfect work alongside everyone else. That removes the stigma of showing something that does not fully work, and it gives people the time, the use case, and the person to ask that experimentation requires.
• Answer Job-Security Fears Honestly: Asked whether AI would take people's jobs, Sammut's honest answer to his own team was that he did not know. He pairs that candor with a real commitment: Zapier cannot promise headcount, but it can promise to be the place where you learn to become an elite, AI-supported professional. Honesty builds more trust than false certainty.

Hearst

Hearst Corporation spans 20,000 employees across hundreds of businesses, from newspapers and magazines to television stations and data services. When leadership announced enterprise-wide AI upskilling, the company set out to train 7,000 employees on its internal tool, HearstGPT, in roughly seven months. Future of work strategist Phil Kirschner recapped the rollout on his newsletter, The Workline, drawing on an interview he conducted for Lead with AI with Hearst learning and development leader Maris Krieger.

• Reframe AI as a Skill, Not a Threat: The initial "transformation" messaging failed and surfaced genuine anxiety about job security and surveillance. The team regrouped and positioned AI literacy as one more skill in the toolbox and a form of career development that increases individual agency. Removing the existential threat is what got people willing to experiment.
• Map Where Authority Actually Sits: Hearst runs a federated structure where business units hold their own P&Ls and local leaders hold real power. Rather than pushing from the center, the team built communication through local business leaders and HR, because employees trust the people closest to them.
• Build Champions to Compensate for Sponsor Gaps: A compressed timeline of 107 live sessions across 16 functional groups forced the L&D team to operate as both change agents and de facto sponsors. They launched a grassroots AI Champions network from the start, which gave local advocates permission to act and lent credibility inside their own business units where central authority could not reach.
• Acknowledge the Other Changes Competing for Attention: Employees never experience one initiative in isolation. Hearst explicitly named the concurrent disruptions its people were living through, from reorganizations to external industry events, and adjusted its messaging accordingly.

The results: 7,000 employees completed training, active HearstGPT usage increased 175%, and business conversations with the Chief AI Officer grew 200%.

Apple

As Apple is rolling out its own AI called “Apple Intelligence, it makes sense that there isn’t a lot of insight into how the most valuable company in the world uses AI internally.

However, two projects are known (based on mostly rumours):

• Engineering Chatbot: Apple created an internal chatbot that some engineers call "Apple GPT." This chatbot requires special permission for employees to access, is used for product prototyping, and can’t be used to develop customer-facing product features
• AppleCare Support Tool: Apple is testing a ChatGPT-style generative AI tool called "Ask" with its AppleCare support employees. This tool generates responses to technical questions from Apple's internal knowledge base to speed up support replies.

Amazon

Through its AWS division, Amazon has been integrating generative AI into various aspects of its operations, transforming internal processes and customer engagement strategies. Here are some key takeaways from their approach:

• The Amazon Q Platform: Amazon developed Amazon Q, a generative AI-powered assistant designed for work purposes. Amazon Q can be tailored to a company's business and integrates with internal systems and data repositories. An example is the AI Sales Assistant, which is reported to save 35 minutes per summary, leading to a 4.9% increase in the value of opportunities created. 
• Strategic Model Selection: Amazon’s multi-model approach uses different generative AI models, including Amazon Titan and Anthropic Claude, tailored to the specific needs of different tasks. This flexibility allows for optimizing accuracy, response time, and cost-efficiency, ensuring that the right model is used for the right task.
• Mitigating Hallucinations and Ensuring Quality: Amazon is preventing AI hallucinations by implementing robust prompting strategies, specific and detailed instructions, and a comprehensive feedback loop involving automated metrics and human review. This multi-faceted approach ensures the quality and reliability of AI-generated outputs, making them more trustworthy for business use.

BCG X

BCG X, Boston Consulting Group's AI-focused division, has been at the forefront of integrating AI into its operations. Led by CTO Matt Kropp, it’s been driving AI adoption across the organization, focusing on enhancing employee experience and unlocking new growth opportunities.

Here are the key strategies BCG X used to implement AI effectively, as shared by Matt in our interview:

• Focus on Enhancing Employee Joy Through AI: BCG X is committed to using AI to minimize toil and maximize joy for its employees. By identifying repetitive, low-value tasks that AI can handle, it aims to free up employees to focus on more fulfilling work. This approach not only improves productivity but also enhances employee engagement and satisfaction.
• Co-Design and Community Engagement: BCG X emphasizes involving employees in the AI implementation. By co-designing AI solutions with the people who will use them and fostering a culture of peer sharing and community engagement (e.g., through hackathons and ambassadors), BCG X ensures better adoption and more effective use of AI tools.
• Discovery Sprints for Innovation: BCG X conducts "discovery sprints" to accelerate AI adoption and innovation. During these sprints, specific project teams are encouraged to explore and experiment with AI tools. The insights and techniques developed during these sprints are then shared across the organization, promoting a culture of continuous learning and improvement.
• Strategic Platform Selection: BCG X has strategically adopted ChatGPT Enterprise across the entire organization to empower its employees. This decision was guided by the need for a robust, general-purpose tool that integrates well with its existing infrastructure and supports the diverse needs of its workforce.
• Beyond Cost-Saving’Focus on Growth: BCG X's internal AI strategy is not solely focused on reducing costs. Instead, it views AI as a means to drive business growth by improving work processes, increasing output quality, and enabling new opportunities. This broader perspective allows BCG X to leverage AI for innovation and competitive advantage rather than just cost efficiency.

Microsoft HR 

Microsoft, maker of Copilot and backer of OpenAI, the company behind ChatGPT, has, of course, been working hard to roll out AI. 

In a fantastic case study shared by Global VP Chris Fernandez, we got a peek into how they pulled it off successfully:

• Human-Centered AI Adoption: Microsoft’s HR team emphasized the importance of keeping human judgment at the core of AI implementation. AI is treated as a tool that augments human decision-making, not as a replacement for it. Companies should focus on creating AI adoption plans that prioritize enhancing the human experience and maintaining a balance where AI supports, rather than overrides, human judgment.
• Empowering Non-Technical Employees to Create: Microsoft empowered HR professionals, who were not traditionally technologists, to become “citizen developers using low-code platforms like Microsoft Power Apps. This democratization of technology enables domain experts to create tailored solutions without needing extensive coding knowledge. Companies can foster innovation by enabling employees across various functions to develop their own AI-driven tools with the help of low-code platforms.
• Iterative and Inclusive Innovation: Microsoft’s approach involved creating an innovation intake process and an AI community of practice, allowing ideas to be collected, prioritized, and refined continuously. This iterative process encourages cross-functional collaboration and ensures that AI-driven innovations are practical and aligned with business needs. Companies should establish mechanisms for ongoing innovation and collaboration, ensuring that AI developments are inclusive and meet the entire organization's needs.
• Focus on Practical Applications and Use Cases: Microsoft’s HR team used AI to address specific, practical challenges, such as automating routine tasks and improving employee interactions through AI-powered bots. This focus on practical use cases ensured that AI implementations delivered tangible benefits. Companies should start their AI journey by identifying specific areas where AI can provide immediate value and gradually expand as they gain experience and confidence.
• Commitment to Responsible AI: Microsoft grounded its AI efforts in its Responsible AI Standard, which includes principles like accountability, inclusiveness, and transparency. This commitment to ethical AI ensures that AI implementations are trustworthy and beneficial to all stakeholders. Companies should adopt similar frameworks to guide their AI initiatives, ensuring that AI is used responsibly and ethically across the organization.

Morgan Stanley

In June 2024, Morgan Stanley launched the AI @ Morgan Stanley Assistant. 

This generative AI-powered chatbot provides financial advisors with quick access to Morgan Stanley's intellectual capital. The tool's adoption rate has been impressive, with 98% of Financial Advisor teams using it.

CEO Ted Pick said these tools could save financial advisers 10-15 hours per week, allowing them to focus more on high-value activities like customizing investment strategies and deepening client relationships. 

PwC Netherlands

In a fascinating interview with HR Tech Director Marlene de Koning, we learned about how PwC Netherlands's clever approach to rolling out AI: 

1. Phased Scaling for AI Adoption: PwC started with a pilot of 300 AI enthusiasts and gradually scaled to all 6,000 employees in the Netherlands. This phased approach allowed for continuous learning, adjustment, and feedback, ensuring that AI adoption was effective and manageable. During this pilot, PwC launched three AI platforms: Microsoft Copilot, ChatPwC, and Harvey.
2. Integration of AI into Daily Workflows: PwC focused on embedding AI tools like Copilot into employees' daily workflows to ensure widespread and consistent use of AI. Integrating AI into familiar tools like Teams, Excel, and PowerPoint, PwC made it easier for employees to incorporate AI into their routine tasks. This approach highlights the importance of making AI accessible and relevant within existing work processes.
3. Identifying and Leveraging Specific Use Cases: PwC took a strategic approach by identifying specific use cases across the company where AI could provide substantial value. Examples include automating the summarization of client workshop notes and analyzing open comments from surveys. By focusing on practical applications that address real needs, PwC demonstrated the immediate benefits of AI, making it easier for employees to see its value and integrate it into their work.  
4. Cultivating AI Influencers: PwC identified and nurtured natural influencers within the organization using organizational network analysis. These influencers were critical in driving AI adoption by sharing their experiences, providing support, and encouraging others to embrace AI. This strategy underscores the value of leveraging internal champions to foster a culture of AI adoption.
5. Balancing Experimentation with Strategy: PwC emphasized the need for a clear AI strategy while encouraging hands-on experimentation. This balance ensures that AI initiatives are aligned with organizational goals while allowing flexibility to learn and adapt as the technology evolves. Companies should adopt a dual approach of strategic planning and practical experimentation to maximize the benefits of AI.
6. "AI in the Loop" Approach: PwC introduced the concept of "AI in the Loop," where AI supports human processes rather than replacing them. This approach helps address potential biases and ensures that AI enhances, rather than diminishes, the human element in workflows. Organizations can benefit from this mindset by ensuring that AI implementations are designed to complement and augment human decision-making.

Roblox

In an interview for our Lead with AI podcast, Roblox CHRO Arvind KC shared how they integrated Generative AI into their business:

• Integration of HR and Systems for Seamless AI Adoption: Roblox merged HR and Systems under a single leadership to create a unified, AI-driven employee experience. This integration allowed for smoother implementation of AI tools, ensuring that technology and people management strategies were aligned. Companies can learn that merging traditionally separate functions like HR and Systems can enhance AI adoption by creating a more cohesive environment where technology directly supports employee needs and organizational goals.
• Strategic AI Deployment in High-Impact Areas: Roblox focused on deploying AI in areas where it could provide immediate and significant value, such as engineering (which accounts for 80% of the business) and customer support. For engineering, Roblox opted for Github Copilot (#7 in our AI Top 100) and Saxon for customer service. This approach ensured that AI was effectively integrated into existing workflows, enhancing productivity without requiring significant changes to how work was done. Companies should identify specific, high-impact areas where AI can be most beneficial and focus their initial efforts to maximize early returns on AI investments.
• Balancing AI Hype with Realistic Expectations: While recognizing AI's transformative potential, Roblox emphasized the importance of setting realistic expectations for its impact. They avoided overhyping AI's short-term benefits, instead focusing on gradual, sustainable improvements. Companies can benefit from this approach by being patient with AI implementation, starting small, scaling over time, and maintaining a long-term perspective on AI’s potential to reshape their operations.

ServiceNow

Like Microsoft, ServiceNow sells AI services, which integrate with HRIS Software like Workday and other AI platforms like Copilot. 

In an interview with Fortune, the company's Chief Customer Officer, Chris Bedi, shared how they integrated through ServiceNow's internal processes, with over 25 use cases in production:

• Company-Wide AI Integration and Adoption: ServiceNow mandated that every department develop an AI roadmap, ensuring that AI was integrated into every aspect of the business, from software engineering to customer service. This approach led to 84% of the workforce using AI daily. Companies should consider a holistic, top-down approach to AI adoption, where every department is encouraged or required to explore how AI can enhance their operations.
• Impact-Driven AI Use Cases: ServiceNow focused on identifying and scaling AI use cases that delivered tangible business impact. They prioritized high-impact applications while discontinuing those that were less effective. This strategic focus on impactful AI implementations helps ensure that AI investments deliver measurable value. Companies should identify and scale AI use cases that provide the most significant benefit to their operations, ensuring a clear return on investment.
• Leadership and Employee Engagement: ServiceNow emphasized the role of leadership in setting the tone for AI adoption, making it clear that AI was a tool to enhance jobs, not replace them. They also actively measured employee sentiment toward AI tools, ensuring that the technology was genuinely helpful and embraced by the workforce. Companies should focus on clear communication from leadership about the role of AI, coupled with efforts to gauge and respond to employee feedback to foster a positive AI adoption experience.

Ready for your AI Implementation? Some guiding principles:

• Strategic AI Integration: All companies emphasized aligning AI tools with their business goals and processes. Whether through phased scaling, strategic platform selection, or focusing on high-impact areas, AI implementations were carefully tailored to enhance core operations without disrupting existing workflows.
• Human-Centered Approach: AI was implemented to augment human capabilities, not replace them. Organizations like Microsoft and PwC prioritized AI tools that support human decision-making, emphasizing the importance of keeping employees engaged and ensuring AI enhances rather than diminishes the human element in workflows.
• Iterative and Inclusive Innovation: Companies adopted an iterative approach to AI adoption, encouraging continuous learning and improvement. This included engaging employees in the development process, fostering cross-functional collaboration, and maintaining a feedback loop to refine AI applications over time. This approach ensured that AI implementations were practical, widely accepted, and continuously improved.

Need more to implement AI successfully? Check out our guide to AI change management.