Organizational Reasoning · Article

The Problem You're Solving May Not Be the Problem You Have

Strong execution cannot rescue a problem that was framed incorrectly

David BrysonOctober 19, 2026 · 9 min read

In 1985, Coca-Cola did not make the New Coke decision without evidence. By then, the U.S. share lead of its flagship cola had been declining for years amid growing competitive pressure. Coca-Cola had a real business concern, and it responded with a substantial research effort.

The company developed a sweeter reformulation and tested it extensively with consumers. Coca-Cola’s own historical account says the new formula was preferred in taste tests involving nearly 200,000 people. In April 1985, the company made a decision that went well beyond introducing another product: it replaced the traditional Coca-Cola formula in the United States.

The public reaction exposed relationships the taste tests had not represented adequately. Consumers were not relating to Coca-Cola only as a flavored liquid. The existing product also carried habit, memory, identity, and attachment to the brand. Opposition was intense enough that Coca-Cola restored the original formula seventy-nine days later.

Good Evidence Can Still Answer a Smaller Question

What makes New Coke important is that the evidence was not obviously weak. The company had data, a clear competitive concern, and a reformulated product that performed well in a meaningful kind of test. The difficulty appeared in the relationship between what the research established and what the eventual decision required.

A taste test can answer a legitimate question: under the conditions of the test, which formulation do people prefer? Replacing an established flagship product asks something larger. That decision also depends on what customers believe they are losing, what the product represents to them, how they may react to its removal, and whether preference for the alternative is strong enough to justify changing the relationship they already have with the incumbent product.

The research did not become worthless because it failed to answer all of those questions. It answered one of them. The reasoning problem begins when evidence that is adequate for a narrower proposition silently becomes evidence for a broader commitment.

The same distinction appears in many forms of organizational work. A company can know that customers like a feature without knowing whether the feature changes purchasing behavior. It can know that a process consumes substantial labor without knowing whether removing that work improves the larger system. It can know that employees dislike an organizational arrangement without knowing whether the structure itself is causing the outcomes leadership wants to change. In each case, the observation can be valid while the implied problem remains incomplete because the organization has begun treating an inference as though the evidence established it directly.

Not Every Bad Outcome Is a Framing Failure

The argument needs a boundary, because a framework that explains every disappointing result as bad framing explains very little. Some problems are framed adequately and still fail for reasons that belong elsewhere.

NASA launched the Mars Climate Orbiter in December 1998 to study Mars from orbit and to serve as a communications relay for other Mars missions. As the spacecraft approached the planet in September 1999, controllers lost contact with it. The subsequent investigation traced the navigation error to a failure in the transfer and handling of measurement units: one part of the system used English units while another expected metric units.

Nothing about that failure requires the conclusion that NASA was pursuing the wrong scientific objective or had misunderstood what the mission was for. The mission could have been framed correctly and the unit-conversion failure could still have occurred. The relevant problem belonged much more directly to engineering interfaces, verification, and execution.

That distinction suggests a practical counterfactual: when an outcome is poor, ask whether the same failure could plausibly have occurred even if the larger problem had been framed correctly. If the answer is yes, the organization may not need a deeper strategic reinterpretation; it may need the appropriate specialist discipline to address a specific mechanism. Reframing is most useful when the frame itself is constraining the reasoning, not when it merely surrounds a failure whose primary cause lies elsewhere.

The Frame Also Determines What You Can See

Problem framing does more than influence diagnosis. It also shapes which possibilities are available to consider.

NVIDIA was founded in 1993 around 3D graphics for gaming and multimedia. Graphics processing units, or GPUs, were built to handle the large numbers of parallel calculations required to render increasingly complex visual scenes. Over time, those processors became more programmable, and the underlying parallel-compute capability had uses that extended beyond graphics.

In 2006, NVIDIA introduced CUDA, a programming platform and model that allowed developers to use GPU throughput for general computational workloads rather than reaching the hardware only through graphics interfaces. Several years later, GPU computation became important to major advances in deep learning. The 2012 AlexNet work, for example, used GPUs to train a large convolutional neural network that produced a major ImageNet result.

The important shift was not that a graphics processor stopped being useful for graphics. The capability became representable and accessible more broadly. Programming tools, libraries, compatible hardware, and a growing developer ecosystem allowed people outside the original graphics context to discover additional uses. If a GPU is represented only as a graphics component, the natural search is for better graphics; when its underlying parallel-compute capability becomes visible, a different possibility space opens.

NVIDIA did not need to foresee every eventual use of GPU computing for that shift to matter. Opening the capability allowed other people to discover applications the original product frame did not need to contain. The example shows why a narrow representation can constrain an organization even when the representation is not false: it can describe what something has been used for without exhausting what the underlying capability can do.

The same problem appears in inherited products, processes, technologies, organizational designs, policies, and roles. Their names already imply what they are and what should be done with them, so teams naturally begin by improving the form they inherited. A process gets optimized as a process; a structural problem leads to structural alternatives; an architecture problem leads to architecture optimization; a feature gap leads to feature comparison. Those responses may be correct, but they all preserve the current representation before asking whether it is necessary.

A useful way to reopen such a problem is to ask why something exists in its current form, what function it is actually performing, and whether that function is inseparable from the form that currently provides it. Regulation, physics, safety, economics, contractual obligations, or technical dependencies may make the answer yes. In other cases, the form is historical rather than necessary. Preserving a required function does not always require preserving the architecture that currently performs it.

New Coke illustrates the danger of reducing a product to one measurable attribute when the product is performing additional functions in the customer’s relationship with the brand. GPU computing illustrates the opposite opportunity: separating an underlying computational capability from the product category in which it first became commercially important can reveal uses that were difficult to see from inside the inherited frame.

Separate What You Observed From What You Think It Means

Organizations rarely encounter a problem as a neutral collection of facts. The observation usually arrives already attached to an explanation: sales are falling because the product is outdated; delivery is slow because the team lacks capacity; customers are leaving because the price is too high; employees are frustrated because the structure is wrong.

Any of those explanations may be correct, but once an explanation becomes embedded in the language of the issue it begins determining where the organization looks next. A ‘capacity problem’ invites hiring, outsourcing, automation, and productivity improvement. A ‘pricing problem’ invites discounting, packaging changes, and commercial analysis. A ‘structural problem’ invites reorganization. The label therefore creates a search space by making some explanations and interventions easier to see while pushing others farther from view.

One practical discipline is to keep four things distinguishable: what the organization actually observed, what it inferred from those observations, what problem definition followed from the inference, and what intervention that problem definition made seem natural. The point is not to manufacture doubt around every decision. It is to preserve enough separation that another explanation can still be considered before the proposed solution begins reinforcing the frame that produced it.

Reframing Has to Be Proportional

The same discipline can be overused. Once people learn to question frames, every problem can become an invitation to rethink the entire system, even when the obvious response is inexpensive, reversible, and easy to test. In those situations, acting may produce better information than continuing to reconstruct the problem at greater levels of abstraction.

Reframing deserves more attention as the commitment becomes more consequential: when the proposed response is expensive or difficult to reverse, when capable people have repeatedly tried to solve the same condition without resolving it, when different groups hold materially different explanations of what is happening, when important evidence does not behave as the prevailing explanation predicts, or when all of the available options appear to preserve the same underlying assumption.

The objective is therefore not perfect understanding before action. It is enough understanding to know whether the organization is solving the problem its evidence actually supports, with reasoning effort proportionate to the consequences of being wrong.

A Better Question Before a Bigger Solution

Organizations are rewarded for execution. Once a problem has been declared, progress becomes visible through budgets, plans, staffing, vendors, milestones, launches, and completed work. Reopening the frame can feel slower because it interrupts that movement, but faster execution does not compensate for a problem definition that points the organization in the wrong direction.

The cases above establish different boundaries around the same discipline. Evidence can be sound yet too narrow for the commitment being made. A serious failure can occur even when the larger problem is framed adequately. And a representation can be accurate enough to describe a capability’s current use while still obscuring what else that capability could become. The question is not whether every frame should be rejected, but whether the frame is adequate for the decision or search it is being asked to support.

Before making a solution larger, faster, more efficient, or more sophisticated, an organization should be able to explain what it actually observed, what it inferred from those observations, which assumptions connect that interpretation to the proposed action, and which materially different explanations or possibilities have been excluded by the current frame. Sometimes that examination will strengthen the existing solution; sometimes it will reveal a different problem; and occasionally it will show that what looked like a problem contained an opportunity that could not be seen from inside the original representation.

A useful question to carry into consequential work is: What would have to be true for this to be the problem we actually have?