Not All Decisions Are Created Equal

Imagine an industrial company debating whether to pilot an AI-enabled engineering assistant. The proposed trial would be limited to one product team, run for a few weeks, and could be stopped with little disruption. Yet the proposal spends months moving between committees.

At the same company, leadership may be under pressure to select an enterprise-wide product lifecycle management (PLM) platform. That decision could shape processes, data structures, supplier relationships, and technology investments for a decade. The discussion is driven toward a deadline, even though critical architectural assumptions remain unresolved.

These are illustrative scenarios, but they expose a familiar management problem: organizations often apply similar decision-making processes to fundamentally different decisions. Some receive far more scrutiny than their consequences justify. Others receive far less.

The answer is not simply to make decisions faster. Nor is it to demand more data, add approval layers, or pursue consensus at any cost. It is to design the decision process around the nature of the decision itself.

Better decision-making is not about making every decision faster or gathering more information. It is about applying the right decision-making approach to the right decision, at the right time, to achieve the right outcomes.

Start with the outcome, not the decision

Before deciding how to decide, define what the decision is meant to achieve. Otherwise, an organization can optimize its approval process without improving the business.

A consequential decision should have an explicit value hypothesis: the desired outcome, the realistic alternatives (including doing nothing), the expected benefits and costs, the important downside risks, and the evidence that would trigger continuation, adjustment, or termination.

For example, “implement a new PLM platform” is not an outcome. Shorter engineering-change lead times, fewer configuration errors, improved reuse, and faster industrialization are outcomes. The platform is one possible means to achieve them.

The distinction matters. Framing the problem around an outcome makes different solutions comparable and provides a basis for measuring what happens after the choice is made.

Figure 1. Decision-to-Value Architecture: intended outcomes define the purpose of the decision; realized outcomes and learning close the loop.

A good decision is not the same as a good outcome

In Thinking in Bets, Annie Duke describes a cognitive error she calls “resulting”: evaluating a decision primarily by how it turned out. A sound decision can produce a poor result because of uncertainty, execution failures, or external events. A weak decision can succeed because circumstances happened to be favorable. Duke later revisited the same idea in her essay, “Don't be a resulter.”

Figure 2. Outcomes provide evidence, but a single favorable result does not establish that the original decision was sound.

This is not an argument for ignoring outcomes. It is an argument for separating two assessments:

• Ex ante decision quality: Given the information, alternatives, incentives, and uncertainties at the time, was the choice well reasoned?
• Ex post performance: What happened, what caused the result, how well was the decision executed, and what should change next time?

An enterprise that judges leaders only by the immediate result risks rewarding lucky bets and punishing intelligent experiments. An enterprise that judges only the process risks excusing poor execution. Both are incomplete.

Four dimensions that should shape every consequential decision

To connect decision quality with business outcomes, I propose assessing four dimensions: impact, reversibility, uncertainty, and urgency.

Dimension	Executive question	Implication
Impact	How consequential is the decision for value, risk, and stakeholders?	Determines attention, rigor, and oversight.
Reversibility	How expensive or difficult would it be to change course?	Determines how much optionality and downside protection matter.
Uncertainty	What do we not know, and would knowing it change our choice?	Determines the need for evidence, experiments, and staged commitments.
Urgency	What is the actual cost of waiting?	Determines the available decision window and how analysis must be time-boxed.

These dimensions are not independent. High uncertainty is more dangerous when a commitment is hard to undo. High urgency reduces the time available to gather evidence. High impact raises the cost of both action and inaction.

That interaction is exactly why one standardized approval process is inadequate.

Lens 1: Impact × Reversibility. How much rigor is justified?

Figure 3. Not every decision warrants the same management attention.

High-impact, hard-to-reverse decisions deserve serious scrutiny. These may include major acquisitions, factory network redesign, core enterprise platform commitments, or fundamental product architecture choices.

High-impact, reversible decisions are different. They can often be approached through controlled experiments, bounded commitments, and rapid feedback. Low-impact decisions, meanwhile, should rarely consume executive bandwidth, even when sensible safeguards are needed.

In his 2015 letter to Amazon shareholders, Jeff Bezos drew a related distinction between one-way-door decisions, which demand deliberation, and two-way-door decisions, which can be made and reversed more quickly. The organizational failure occurs when heavyweight processes designed for the first category become the default for the second.

The critical implication is that the level of governance should be proportionate to the decision, not automatically inherited from the organizational hierarchy.

Lens 2: Uncertainty × Reversibility. Should we experiment, investigate, or commit?

Figure 4. Uncertainty determines the need to learn; reversibility determines how safely we can learn by doing.

When uncertainty is high and the decision is easy to reverse, a pilot may be better than another workshop. When uncertainty is high and the commitment is difficult to reverse, moving immediately to full implementation may be reckless. Here, the task is to identify the few uncertainties that could actually change the decision.

A proof of concept should therefore not merely demonstrate that technology works. It should test a decision-critical hypothesis. If the evidence would not change the choice, the experiment may be theater rather than learning.

But the opposite error is equally common. Organizations keep collecting data long after additional information is unlikely to justify its cost or delay. The relevant question is not “Do we have enough information?” in the abstract. It is “Would the expected value of further information outweigh its cost, including the cost of waiting?”

This is also where reversibility becomes a design variable. Through modular architectures, contractual exit rights, pilots, migration waves, and staged capital allocation, enterprises can sometimes turn a seemingly binary commitment into a sequence of smaller decisions. Real-options thinking has long examined the value of preserving such choices under uncertainty; see, for example, Copeland and Tufano's discussion of staged investments.

Lens 3: Urgency × Reversibility. How fast must we act?

Figure 5. Urgency should compress the decision process selectively, not eliminate discipline.

A genuinely urgent, reversible decision should usually be made quickly, close to the people with the best information. A genuinely urgent, hard-to-reverse decision is different. It requires escalation, a clear deadline, focused examination of the most consequential unknowns, and safeguards against catastrophic downside.

But urgency must be challenged. A board meeting, quarterly steering committee, or executive preference is a date, not necessarily an economic reason to decide. The right test is the cost of delay: lost value, increased risk, expiring opportunities, regulatory obligations, or operational harm.

Urgency should determine the time budget for analysis, not automatically its absence.

Three matrices, one practical decision sequence

The matrices are not competing frameworks. They are lenses within one decision architecture, applied in a logical order:

1. Define intended outcomes. What business value are we pursuing? Against what baseline, alternatives, and constraints?
2. Assess impact. Is this a routine choice or a genuinely consequential commitment? Set the level of attention accordingly.
3. Assess reversibility. What would it cost to change course? Can we design the commitment to preserve options?
4. Assess uncertainty. Which assumptions could change the choice? Is an experiment, targeted analysis, or staged investment justified?
5. Assess urgency. What is the real cost of waiting? How much time can we afford to spend on steps 3 and 4?
6. Choose the approach and decision rights. Act, experiment, investigate, stage, or commit. Name a decision owner and clarify who advises versus who approves.
7. Execute and measure. Track delivery, adoption, intended benefits, unintended consequences, and conditions that require a change of course.
8. Learn and adapt. Review the original assumptions against what actually happened, separating judgment, execution, and external effects.

This sequence is a management discipline, not a rigid waterfall. Impact, reversibility, uncertainty, and urgency should be reassessed when new evidence or deadlines change the decision context.

What this looks like in an industrial enterprise

Consider a manufacturer deciding whether to replace its enterprise PLM landscape. The desired business outcomes might be improved configuration consistency across engineering and manufacturing, reduced engineering-change cycle time, and fewer downstream data errors.

At first, the decision appears straightforward: choose a platform and plan the rollout. The four-dimensional assessment reveals a different picture:

Dimension	Illustrative assessment	What it means
Impact	High	The decision influences multiple functions, budgets, and years of operation.
Reversibility	Low	Migration, integration, process changes, and vendor commitments are expensive to unwind.
Uncertainty	High	Data quality, configuration logic, adoption, and integration complexity may be insufficiently understood.
Urgency	Moderate	A transformation milestone creates pressure, but an immediate enterprise-wide commitment is not necessarily required.

The recommended approach is not simply to postpone the decision until certainty arrives. It is to change the structure of the commitment.

Rather than selecting a full replacement and immediately committing to a large rollout, the organization could test one representative product family and its end-to-end information flow. The test would focus on whether product structures, variants, engineering changes, and downstream manufacturing data can be managed as required. Decision gates would determine whether to expand, adapt, or stop.

That approach will not remove every dependency or transition cost. But it may reduce uncertainty before the most expensive commitments are made. It also creates evidence that executives can use to compare a full replacement, phased modernization, and improvement of the current landscape.

The lesson extends beyond PLM to manufacturing automation, digital twins, enterprise AI, new product launches, and capital investment: an architecture that preserves technical options can also preserve strategic options.

Governance should follow the decision, not the other way around

The framework has direct implications for organizational design. Decision rights, evidence requirements, escalation paths, and investment gates should not be identical across every decision.

A practical governance model would establish a few clear rules:

• Delegate reversible, low-impact choices to the teams closest to the information, within defined boundaries.
• Require explicit hypotheses and learning criteria for uncertain experiments, instead of treating pilots as small implementations with no decision to follow.
• Reserve rigorous executive review for high-impact, hard-to-reverse commitments and material exposure.
• Use staged funding and decision gates where uncertainty is material and optionality has value.
• Record assumptions, owners, and review triggers before commitment, making subsequent learning possible without rewriting history.

Governance is valuable when it improves decision quality or controls material risk. It becomes waste when it adds time without changing the choice, the accountability, or the outcome.

Close the loop: Evaluate decisions and outcomes separately

The last step is often the most neglected. Once a decision has been approved and a program launched, the rationale can vanish into meeting minutes, while attention shifts to delivery milestones.

A mature decision architecture preserves a short decision record: the intended value, alternatives considered, key assumptions, confidence levels, owner, commitment level, and explicit conditions for revisiting the choice. This record then informs three different reviews:

Review	Core question	Typical indicators
Decision quality	Was the choice sound given what was known at the time?	Assumption quality, alternatives considered, evidence relevance, decision latency.
Execution quality	Was the choice implemented effectively?	Adoption, implementation reliability, time-to-value, operational readiness.
Outcome quality	Did the intended business value materialize?	Financial returns, productivity, quality, customer value, strategic capabilities.

A poor result should trigger investigation, not automatic condemnation of the original decision. Equally, a favorable result should not excuse weak reasoning. Over repeated decisions, forecasts and realized outcomes can help the organization calibrate its judgment, improve its methods, and distinguish genuine capabilities from luck.

This is what turns a decision process into an enterprise learning system.

The real competitive advantage is knowing how to decide

Better decision-making does not come from treating every choice as a crisis, nor from forcing all choices through the same governance machinery. It comes from recognizing the different economics of different decisions.

Start with the outcome. Understand the impact. Assess what can be reversed. Identify what must be learned. Respect the real cost of delay. Match governance to the stakes. Commit deliberately, execute effectively, and learn from what happens.

The ultimate question for leaders is not simply, “Are we making decisions fast enough?”

It is this: “Are we making different decisions differently enough, and are those decisions creating the value we intended?”

---

Further reading

• Annie Duke, Thinking in Bets (2018), on uncertainty, decision quality, and resulting.
• Annie Duke, Don't be a resulter (2025), on why one outcome cannot establish decision quality.
• Jeff Bezos, 2015 Letter to Amazon Shareholders, on one-way and two-way door decisions.
• Tom Copeland and Peter Tufano, A Real-World Way to Manage Real Options, Harvard Business Review (2004), on staged investment under uncertainty.