Navigating the AI Revolution: A Leader’s Guide to Enterprise AI Implementation

By Marilyn Markham, AI & Automation Strategy Vice President at American Express Global Business Travel

The conversation around Artificial Intelligence (AI) has shifted from a futuristic whisper to a present-day roar. For leaders, the question is no longer if we should integrate AI, but how. The journey to bring generative AI into the enterprise is less about flipping a technology switch and more about cultivating a new, human-centric leadership methodology.

Our own generative AI journey at American Express Global Business Travel (GBT) was not a sudden mandate – it has been a continued evolution in our solutions for years, bolstered by acquisitions of Egencia and 30SecondsToFly. Our efforts progressed into a dedicated initiative by January 2024. We identified key pillars for transformation—Productivity, Service, Finance, and Engineering. While these areas were right for our business, the framework we developed for this journey can provide valuable insights for any leader.

Formalization: The Bedrock of Responsible AI The initial excitement surrounding generative AI often leads to “shadow IT” and fragmented R&D. To truly harness its potential and mitigate risks, the first crucial leadership step is formalization. This isn’t about stifling innovation but creating a safe and transparent environment for it to flourish. This formalization must address the core concerns of your clients, leadership, and your tech-curious workforce.

Another critical component of this journey is bringing everyone along. An AI revolution cannot be dictated from the top down; it must be nurtured from the ground up. We made a conscious effort to offer AI training as an opportunity for personal and professional growth, giving our people the tools to remain competitive and relevant. By providing access to learning platforms and general use AI tools, we replaced fear with excitement and curiosity. When everyone is part of the revolution, it generates a groundswell of ideas, making AI useful and valuable for the entire organization.

Our approach also led to a clear management policy that can be scaled globally and ensures adherence to international laws and technology best practices. Amex GBT’s roots in a bank holding company means we have the highest standards of data privacy, security compliance and governance – so this consideration is paramount when submitting data into AI applications and more. We created an AI Oversight Committee, which is a cross-functional team tasked with ensuring ongoing compliance. Furthermore, we updated our third-party and vendor management policies in collaboration with procurement, legal, and cybersecurity teams to establish new evaluation criteria and contract language.

Finally, we established a technology “playground”—an in-house, secure environment with approved large language models (LLM). This empowers our engineers to experiment safely, with any new functionality being built once and shared across the community. By creating this structured framework, you remove guesswork and signal a clear, responsible approach to all stakeholders.

Prioritization: From Hype to Tangible Value Once a solid foundation is in place, the next challenge is to channel the flood of ideas into a focused investment strategy. The promise of AI can lead to many requests, and a structured intake process is essential. We required a business case for each idea, but our AI Initiatives Business Review team still had to triage many proposals where AI was not the right tool, the cost outweighed the benefits, or the idea was not core to our strategy.

The ideas that gained the most traction were those that applied generative AI’s core strengths or involved partnerships with vendors offering plausible solutions at a reasonable cost.

To action these ideas, we created a dedicated AI Studio team. This allowed our mainstream engineering teams to maintain focus while we cultivated specialized expertise. A key learning for us was that while a proof-of-concept might seem straightforward, the journey to a productized, scalable solution requires testing at scale as this is where biases and hallucinations in the training data come to light. The takeaway for leaders is clear: be selective, but don’t relegate AI to a side project. A dedicated team can build momentum and disseminate knowledge, systematically embedding AI capabilities across the organization.

Validation: The Rigorous Path to Production

Validating a generative AI use case is a familiar process with a new twist. Best practices in product development still apply, but with unique considerations:

• Usability: We must never forget that we are crafting human experiences. Is the technology intuitive and genuinely helpful?
• Quality of Output: The inherent variability of generative AI requires scaled testing with real world, often imperfect, data. We discovered, for instance, that our AI could “hallucinate” flight details if key information was missing from a request.
• Cost-Benefit Analysis: Enterprise-grade AI is not free. High data volumes can lead to prohibitive costs, so it’s essential to be pragmatic. Thankfully, processing costs are decreasing, and smaller, open-source models are becoming viable options.
• Strategic Risk Management: Before production, any solution must be presented to your AI oversight body to ensure all risks have been addressed. This goes beyond technical compliance. HR involvement is necessary to navigate the potential for job displacement and the need for upskilling. Likewise, commercial teams must assess market impacts and how AI-driven changes will be perceived by clients.

It’s also vital to time-box your R&D. Once you have your answers from the validation phase, make a swift decision: productize or pivot. The pace of progress is astonishing, so focus on what is achievable now.

Productization: Bridging Potential and Reality

Bringing an AI-enabled solution to life at scale introduces new design questions. A common one is how to present AI to the user. Our rule of thumb is simple: for complex but context-driven prompts, use a button. For conversational interactions, a chat window is more intuitive.

Regardless of the interface, always initialize the conversation with the LLM with a prompt that establishes context and reinforces safeguards. For example, our AI Contract Assistant is pre-set to act as a paralegal, instructed not to invent information, and to reference clause numbers in its answers.

The journey of integrating AI is a marathon, not a sprint. It demands a leadership approach that is both visionary and pragmatic. By building a strong foundation, prioritizing wisely, and validating rigorously, you can navigate the complexities of the AI revolution. But it’s crucial to remember that technology alone is not a silver bullet. True adoption and transformation are only possible when you invest in your people through training, upskilling, and thoughtful change management, unlocking the full potential of human and artificial intelligence working together.