Your AI Strategy Found the Problem It Wanted to Solve

I was sitting in a meeting about a year ago where an executive described the organization's new AI strategy. Forty-five minutes in, I noticed that nobody had named a problem they were solving. The conversation was entirely about what the AI tools could do, which vendor to choose, what the rollout timeline should look like, and which teams would be "AI-enabled" first. The problem that this strategy was supposed to address kept shifting. First it was efficiency. Then it was competitive positioning. Then it was talent attraction. The solution was fully formed before the problem was.

Cohen, March, and Olsen (1972) published a paper called "A Garbage Can Model of Organizational Choice" that describes exactly this. My recollection is that they were studying university governance, which they characterized as an "organized anarchy": an organization with problematic preferences (goals that are unclear, inconsistent, or discovered only through action), unclear technology (the organization does not fully understand its own processes and how they produce outcomes), and fluid participation (the people involved in decisions change from moment to moment). In such an organization, rational decision-making, the kind where you identify a problem, generate alternatives, evaluate them, and choose, is not how decisions actually get made.

Instead, the garbage can model proposes four independent streams that flow through the organization simultaneously. Problems are unresolved concerns looking for attention. Solutions are ideas, tools, capabilities, and technologies looking for problems to attach to. Participants are the people available to be involved in decisions at any given moment. Choice opportunities are occasions when an organization is expected to make a decision: meetings, budget cycles, committee reviews, crises. In Cohen, March, and Olsen's framework, these four streams do not get assembled rationally. They collide. A decision happens when a solution, a problem, a set of participants, and a choice opportunity happen to be in the same place at the same time. The collision is temporal, not logical.

The word "garbage can" refers to the choice opportunity itself. It is a container into which problems, solutions, and participants get thrown. What comes out depends on which elements happened to be present when the can opened. The decision is less a product of analysis and more a product of timing.

I need to flag that the garbage can model is not in my local study file sources, so I am working from memory and general knowledge of the theory here. I need to verify the exact claims before citing this in an exam context. But as an analytical lens for understanding what I observe in organizations, it is one of the most useful I know.

The AI strategy wave of 2023 and 2024 is a near-perfect illustration. Large language model tools became publicly available in late 2022. By early 2023, every enterprise technology company, consulting firm, and software vendor had a generative AI announcement. The solution was in the room before most organizations had identified a problem they wanted to solve with it. The typical sequence was not: we have a customer service problem, we explored AI as a potential solution, and we concluded the fit is good. The typical sequence was: AI is clearly going to be important, our competitors are announcing AI initiatives, our board is asking about our AI strategy, so we need one. The solution found the meeting. The problem got filled in after.

This is not cynicism. Cohen, March, and Olsen were making an empirical observation about how organizations actually function, not a moral judgment about how they should. Under genuinely uncertain conditions, where nobody knows which problems matter most or which solutions will work, the garbage can process is a reasonable adaptive strategy. You cannot develop a strategy for problems you have not identified yet. Solutions get developed based on what is technically feasible, which is often ahead of organizational readiness to name the problems those solutions address. The garbage can is not dysfunction. It is how complex organizations make decisions in ambiguous environments.

What it does explain is why so many AI strategies feel unmoored. The solution arrived with its own momentum, driven by mimetic pressure from other organizations, vendor enthusiasm, and genuine technical excitement. The problems that got attached to it, efficiency, competitive positioning, talent, experience, were real problems, but they were attached after the solution was already committed to. The mismatch between solution characteristics and the actual texture of the problems creates the implementation difficulties. You adopted a solution that can do certain things, and the problems you attached to it require different things.

The connection to what I wrote about sensemaking and how organizations create coherent narratives after the fact is direct. Weick's sensemaking framework says that organizations act first and create coherent stories about why they acted second. The garbage can model explains the moment of action: the solution attached to the problem at a particular choice opportunity. Sensemaking explains what comes after: the organization constructs a narrative of strategic intent around a decision that was actually produced by a temporal collision of solution, problem, and participants. "We adopted AI because we recognized the strategic necessity of transforming our customer experience capabilities" is the sensemaking account of what was actually: "The vendor pitched us, the board asked about our AI plans, the IT director was enthusiastic, and the budget cycle was open."

The mimetic pressure I wrote about in my post on institutional isomorphism and AI adoption feeds directly into the garbage can. Mimetic isomorphism delivers solutions into the can before problems are identified. When everyone in your field is announcing an AI strategy, the solution, AI adoption, is already circulating in the organizational environment. It arrives at choice opportunities, quarterly planning sessions, board meetings, budget reviews, ready to attach to whatever problems happen to be present at that moment. Compatible innovations diffuse faster, as Rogers would say. But the garbage can model explains the mechanism: when the solution is already in the room, the compatibility question is about whether the problems can be framed to fit the solution, not whether the solution fits the problems.

I think the practical implication is uncomfortable for strategy teams. If the garbage can model is accurate, then the right question after adopting any technology is not "did we make the right decision?" but "what problem did this solution actually attach to, and is that the problem we should have solved?" The post-hoc problem identification is where the real strategic work is. Once the solution is committed, the organization needs to be honest about what problems it is actually being asked to solve, and whether the match is good enough to produce value.

My read is that the garbage can model is not cynical. It is accurate. The most honest thing a strategy team can do is admit when a solution arrived before the problem was named, and then work backward to figure out whether the problem that got attached is real and whether the solution can address it. Pretending the decision was rationally sequenced is the part that actually creates dysfunction, because it forecloses the honest evaluation the organization needs to do after the can has opened.