The Meeting That Never Ended
We had the right set of stakeholders gathered for a meeting that was supposed to result in approval to move beyond a pilot using AI, an example of innovation and adoption that aligned with the company’s broader AI transformation objectives. We had product, engineering, data science, legal, and senior leadership. We had a prototype and a deck prepared.
At the end of the meeting, we scheduled another meeting.
The prototype performed as expected, but legal wasn’t looped in early enough and raised risks that had to be addressed before expanding. The data the model needed required access to a system owned by a team that was reluctant to participate. Engineering raised concerns about the current use case because they wanted to prioritize another. We learned that another team had been running a parallel pilot, and we were on a collision course with a platform migration that wouldn’t be able to handle the technology we were proposing.
It’s easy to place the blame on anyone in the room, but everyone believed they were doing their job. And yet the project that was supposed to be aligned with AI transformation seemed destined for failure.
Sound Familiar?
At this point, most organizations — companies, government agencies, non-profits, international development agencies — have launched an enterprise AI transformation initiative in the past two years. In this Fortune piece, a number of leaders describe some version of the challenges they faced. The context and details differ, whether the industry or sector, a civilian or national security-focused agency, the types of customers, or the application of AI, but the pattern is the same.
Intelligent, experienced practitioners do work that tries to satisfy leadership and corporate or mission objectives. But they work in silos, with data scientists building models, product managers identifying use cases, engineers building prototypes, legal and compliance managing risk, and everyone attending meetings. Lots of meetings. Everyone is making progress on what matters to each, but the pieces never connect.
Prototypes and pilots succeed from a technical perspective, but they are too often stopped because of organizational barriers. Then weeks turn into months and progress is elusive. The transformation initiative that was supposed to propel the organization forward becomes a collection of disconnected experiments that leaves everyone feeling deflated.
It’s not a technology problem. It's a structural problem, and it shows up across organizations and sectors, both private and public.
Even the companies selling AI have lived it. From the same Fortune article, Salesforce’s SVP Andy White, who oversees adoption of Salesforce products for internal operations, admitted the company built hundreds of agents before cutting back to a few: “We chased too much all at once and didn’t have enough focus.” Kyndryl’s CIO described automating processes that were “already clunky” and found it “didn’t fix anything. It actually made it more complicated.”
A Consistent Pattern
Over nearly twenty years of innovation and transformation work, including at the FBI, the World Bank, and an AI mobility scaleup, I’ve run into the same failure modes.
In government at the FBI, the challenge was how to share intelligence with private industry to help protect companies and the U.S. economy from criminal and national security threats. The FBI already had an answer, which was to build a portal to share information with companies so they can defend themselves. This was solution-first thinking. As I researched the challenge, the underlying problem was that the intelligence being shared didn't meet companies' needs. The real barriers were structural and organizational, such as operating models, incentives, priorities, and ways of working that cut against cross-domain collaboration. A web portal wouldn't fix these.
In Brazil, I worked with the World Bank to help small farmers adapt to climate change using emerging technologies. In this case, we had an executive mandate and donor buy-in, but a mandate doesn't replace understanding. Doing the deep diagnostic work of mapping the ecosystem, observing firsthand what farmers and entrepreneurs needed and not relying on second-hand assumptions from consultants or the agriculture ministry, was critical in building a solution to the right problem.
At the AI mobility scaleup, I was told the company had spent two years trying to define the second product. I realized this was due to fragmented efforts with piecemeal meetings and no cohesive strategy or process to define what a second product could be. The company was spinning its wheels. Before I joined, a new CEO made defining the second product an imperative, so it made sense to pursue a focused, rigorous, cross-disciplinary effort. I teamed up with two others, one in customer success and the other in product design, and interviewed key stakeholders, mapped the opportunity space, surfaced fifteen candidate use cases, identified knowledge gaps, then brought the right cross-disciplinary team into the room with a structured methodology to work through decisions together.
What’s consistent across these examples is that teams and organizations often skip past identifying the underlying problem to solve or lack a rigorous approach to do so. The tendency is to pursue and develop solutions as quickly as possible, but without adequate diagnosis that includes gathering a range of perspectives, especially diving into the details with those living the problem, the result is often a solution in search of a problem, with many wondering why it never reached production and scale.
Three Principles for Innovation and Transformation
In these and other examples, what helps is the application of three principles working together as an integrated methodology.
1. Diagnose Before You Design
Before building any solution, define what problem you’re trying to solve. Where do people make judgments today? What information do they use? What are the real constraints and not the assumed ones? How are we measuring success? Who owns the outcome? In a government or national security context, what other agencies and partners have a stake in the outcome, and how do we balance each of those competing objectives?
Most people underweight this diagnostic phase, including when it comes to AI transformation. We all want to build the product or solution, but without mapping workflows and decision points by rolling up your sleeves with those living the problem, you build technically brilliant AI that doesn't align to the desired outcome.
At the AI scaleup, the preparation we did prior to the workshop accomplished four things. It surfaced real opportunities, highlighted gaps in our knowledge, identified who needed to be in the room, and gave everyone a common baseline to work from. Without doing so, we would have been spinning our wheels building solutions for problems we hadn’t collectively defined.
2. Build With, Not For
I’ve seen that when the C-suite drives AI transformation, the temptation is to dictate top-down, announcing decisions to the people who'll be affected rather than building with them. This results in compliance rather than co-creation. Teams race to respond to the mandate, because they want to show that they’re responding to leadership. You get pilots that succeed in demos and fail as they head into production. Instead, leaders should engage employees in what they’re seeing as problems and challenge their thinking to sharpen their ideas. Teams should use that engagement as an opportunity to uncover and define the underlying problem. Doing so helps to formulate the strategy and execute on it, because you built buy-in throughout the organization.
At the FBI, co-creation was not the theatrics of a few meetings to demonstrate stakeholder buy-in. It meant aligning analysts, data scientists, attorneys, and agents from headquarters and the field around the needs of companies, gathering insights directly from the practitioners, teams, and leaders in industry living the problem, and working through iterations to build muscle memory.
At the World Bank, we convened stakeholders across the full ecosystem, from government agencies, universities, extension agents, cooperatives, and entrepreneurs. We used a structured approach over time so the model emerged from their input. The result of that integrative approach was something none of us could have done without engaging each other. At the AI scaleup, we iterated and refined how we prioritized use cases for the product, driven by the team’s collective input. When someone from go-to-market flagged an operational hurdle to overcome, we refined our thinking. That flexibility created buy-in. It became our framework, strategy, and execution plan.
People support what they help create.
3. Think Big, Start Small, Adapt to New Insights
With AI, we are living in a different moment now (putting aside the debate on pushing the frontier of AI). The risk of not moving fast enough, falling behind, missing transformative opportunities, and playing catch-up can outweigh the risk of waiting to have all the pieces in place. AI capabilities evolve week to week. Waiting to prove value with small pilots means you could be behind before you know it.
But speed without strategy produces fragmentation only faster. The discipline is to define a north star, challenge assumptions and what’s taken for granted, identify how use cases connect, invest in shared governance and infrastructure, and pursue high-value use cases in parallel, each with the same rigorous methodology that builds toward a coherent strategy.
At the scaleup, we moved from use case definition to customer trial discussion in days and weeks, not months. We chose one primary use case, remaining disciplined while maintaining our north star. This meant engaging internally on how this use case aligned with our technical capabiliites and other promising use cases. We balanced this while listening deeply to the needs of our customers.
What Becomes Possible
Progress on FBI government-industry intelligence-sharing that stalled for years changed when we bridged the divide between industry and the FBI by deeply understanding their needs and folding them into the intelligence production cycle — there was also a significant amount of trust building, part of which is following through on your promises. Valuable sharing increased dramatically, resulting in the mitigation of national security threats. The effort produced a 29-fold improvement in delivery and sharing.
In Brazil, the World Bank work didn't launch as a pilot. It launched as a platform called NITA, the Nucleus for Technological Innovation for Family Agriculture, a multilateral collaborative with the governor of Santa Catarina as a key supporter. The platform has since grown to become part of Agate, a leading player in Latin America’s emerging tech ecosystem, as measured by venture capital investment.
At the AI scaleup, rigorous sessions over one week with structured preparation and follow-through did what two years of fragmented efforts could not accomplish. We produced a clear product vision for the second product, a strategy, and a path to execution, with full cross-functional alignment and C-suite buy-in to execute on that vision.
These examples represent the ability to move from pilot to a coherent platform when you apply a rigorous methodology of diagnosis, integrative thinking, and collaboration with sustained commitment. Add to this visionary leadership, executive sponsorship, and the will to see it through, and the organization can move and scale more quickly.
The pace of your competition is only moving faster, and those who apply these principles can build AI-native workflows and compress their product lifecycles more quickly. Over time, that muscle memory results in a sustained organizational capability. The alternative is continued fragmentation while the competition increases their lead.
What You Can Do
As you think through your organization, whether it’s a company, government agency, or an international development organization, ask the following when it comes to your AI transformation efforts:
Do you have the right people in the room who understand the problem and are willing to challenge the conventional wisdom?
Do these people and teams have decision-making authority?
Are teams doing the diagnostic work and is there a willingness to integrate those perspectives to find common ground and alignment?
Is there an overarching strategy that various efforts align to, or is it a collection of pilots that lack a vision?
Does executive sponsorship exist, or is AI transformation just one competing priority among many?
Have you created the conditions for constructive collaboration, where people check their ego and title at the door?
If the answer to many or most of these is no, then what is likely required is less about the technology and more about the approach and a better way of working.
It means commitment from the seniormost leaders to remove barriers, provide resources, and encourage challenging the norm. It requires a structured methodology that begins with diagnosis, encourages co-creation, and maintains a strategic vision while adapting to new insights.
It can feel like a more difficult path to do things in a way that challenges the status quo, but if the organization can set the right conditions, there will be a greater chance of real innovation and success in scaling the AI transformation.
The views expressed here are mine and not those of the FBI.