People in organizations (ideally) do three things: they inform and educate, they solve problems, and they make decisions. The three are more interleaved than organizations typically acknowledge. From the board down, different levels have different needs for each, and the characteristics change at each level. Informing a board is different from informing a team. Solving a supply chain problem is different from solving a strategic positioning problem. Problem solving and decision making, in particular, are often treated as the same activity. They are not, and the distinction matters.
This post is about making decisions.
What Decisions Are
A commitment to a course of action under certainty is a plan. Under uncertainty, it is a decision. Uncertainty means the output of analysis is a distribution, not a point estimate. If you are analyzing a decision and producing a single number, you have discarded the uncertainty, which is what makes it a decision rather than a plan. Decision analysis is statistical at its foundation: distributions, probabilities, conditional outcomes. Statistical reasoning is unintuitive. It requires training that most people have not had, and even with training it requires discipline that competes with faster cognitive shortcuts. The domain where analytical rigor matters most (decisions under uncertainty) is the domain where unaided human intuition is least reliable.
Before a decision comes a question, and before a question comes an issue: the recognition that something needs attention. “Our largest customer is unhappy and the margin is eroding” is an issue. “Should we renegotiate, restructure, or exit the relationship?” is a question. “If we renegotiate, the relationship might be salvageable” is a hypothesis. It is not yet testable. You don’t know the terms. You don’t know under what conditions the margin recovers, or to what level. You don’t know if the customer would accept. You may not have the systems in place to calculate any of it. The hypothesis has to be refined: what terms, what margin threshold, what the customer’s alternatives are, what the contract allows. Each refinement step requires its own analysis, and each analysis may change the question. “If we renegotiate at these terms, the customer stays and margin recovers to X” is a hypothesis you can test. Getting from the first formulation to the second is itself a process, and much of the analytical work lives there. The issue surfaces naturally. Converting it into a well-framed question is hard. Formulating a testable hypothesis is harder. Refining it to the point where it can actually be evaluated is where most of the work is. Most organizations stay at the issue level, or jump from issue to recommendation without the intermediate steps.
But a refined hypothesis is still a single proposition in a single state of the world. “The customer stays” and “the customer leaves” are not different points on one distribution. They are different scenarios, each with its own distribution and its own consequence chain. Stress testing probes the boundaries: under what conditions does this decision fail, what assumptions must hold for it not to be catastrophic, what happens in the tails. Evaluating hypotheses across scenarios and stress conditions at once is what Monte Carlo simulation does: parameterize the uncertainties, run tens of thousands of samples, get the full distribution of outcomes. Sensitivity analysis tells you which parameters matter most. The methods exist. They come from statistics, operations research, quantitative finance, and decision science.
Most people in organizations have never encountered them. The vocabulary itself (parameterize, distribution, scenario, sensitivity) is specialist. The gap is not only that statistical reasoning is unintuitive and requires training. Most people do not have the language to describe what good decision analysis would look like. They know how to build a spreadsheet and make a slide deck, not how to parameterize an uncertainty and evaluate a distribution. And even when the analysis is done, assumptions are set once, buried in a footnote or a spreadsheet cell, and never revisited. When an assumption breaks, nobody connects the failure back to it because nobody remembers it was an assumption.
Between the hypothesis and the decision sits problem solving: applying known methods to the identified problem. This is the part organizations are generally capable of. The expertise exists, the methods are known. The gaps are at both ends of the chain: upstream (framing the right question, formulating the hypothesis) and downstream (evaluating distributions, making the commitment under uncertainty). The parts that require methodology are not the hard parts. The parts that require statistics and judgment are.
Not all decisions are equal. They vary along several dimensions, and the dimensions interact in ways that strategy books and consulting frameworks tend to flatten into 2×2 matrices. What follows is the actual complexity of what makes a decision hard, and it is qualitative. Before analysis can become quantitative (distributions, scenarios, simulations) it must first be structural: the dimensionality, the dependencies, and the hierarchical structure of the decision.
Good executives already do a version of this intuitively. They reduce dimensionality. They identify the two or three variables that matter and hold the rest constant or irrelevant. They flatten a ten-dimensional problem into one they can reason about. This is not a shortcut. Each additional dimension multiplies the space of possibilities, so reducing dimensionality does not make problems incrementally easier. It makes them categorically easier.
But some problems are irreducible. The dimensions interact. The dependencies are circular. The hierarchy has feedback loops. Flattening discards the interactions that determine the outcome. The executive who reduces a non-decomposable decision to its two most salient dimensions has made the problem tractable at the cost of making it wrong. Experience tells you when you can safely reduce. It does not always tell you when you can’t. The decisions where dimensionality reduction fails are the ones that need quantitative methods: holding ten interacting variables in a single analysis is something computation does and human intuition does not.
Characteristics of Decisions
Frequency and consequence. Some decisions are high-frequency and low-consequence: what to prioritize this sprint, which vendor to use for a commodity service. You get many repetitions, and the cost of being wrong on any single one is low. Other decisions are low-frequency and high-consequence: should we enter this market, should we do this acquisition, should we change the capital structure. You get very few repetitions, the cost of being wrong compounds over years, and there is no rapid feedback loop to learn from. But the interesting failures are the misclassifications. Decisions that look low-consequence but compound: which customers you say yes to early on shapes what kind of company you become, and by the time you notice you’ve drifted into serving a market you didn’t intend, dozens of “small” decisions have already set the trajectory. Decisions that look high-frequency but are precedent-setting: the first discount you give, the first exception you make to a policy, the first time you let a deadline slip, each one becomes the reference point for the next one. Decisions that are high-consequence but get treated as low-consequence because they’re embedded in routine processes: budget allocation is often done annually through a process that feels operational (fill out the template, negotiate with finance) but is actually one of the most consequential strategic decisions the organization makes. And decisions that are low-frequency but get made by default: not deciding is a decision. Renewing a contract without reviewing whether the relationship still serves the strategy. Continuing to invest in a product line because stopping requires an affirmative choice that nobody makes.
Reversibility. Some decisions are more reversible than others (which tool to use for an internal project, which meeting cadence to adopt). Some are effectively irreversible (a major acquisition, a market exit, an IPO). But reversibility is a spectrum with its own complications. Some decisions are technically reversible but practically irreversible: you can undo a reorg, but the trust damage, the departures it triggered, and the institutional knowledge that walked out the door don’t come back. Some are reversible in theory but the reversal costs more than the original decision: you can exit a market you entered, but the sunk cost of entry makes the exit more expensive than if you’d never entered, so you delay the reversal until it’s much worse. Some are reversible in components but not in combination: you can change the technology stack, you can change the org structure, you can change the business model, but if you did all three at once, reversing any one while the other two are in flight is nearly impossible. Some have time-decaying reversibility: a partnership that could be exited cleanly at year one has contractual, relational, and operational dependencies by year three that make exit far more costly, and nobody tracks when the window closed. And some are reversible for you but irreversible for others: you can decide to stop serving a customer segment, but those customers made plans based on your presence, and your reversibility creates irreversibility for someone else that comes back as reputational cost or loss of future options.
Decomposability. Some decisions can be broken into independent sub-decisions that can be evaluated separately. Some can’t, and the ways they can’t be decomposed vary. Linear chains: capital allocation affects growth rate, which affects organizational capacity, which affects technology choices, which constrains what markets you can serve. Circular dependencies: pricing affects volume, volume affects unit economics, unit economics affects pricing. Shared resource conflicts: the same engineering team is needed for the new product launch, the platform migration, and the security overhaul, and you can’t evaluate any of those independently because they’re competing for the same constraint. Timing coupling: the decision to raise capital depends on market conditions, but the decision to launch the product depends on having the capital, and the market conditions depend partly on whether you’ve launched the product. Organizational identity: the decision to enter an adjacent market changes what kind of company you are, which changes who you can recruit, which changes what you can build, which changes which markets are accessible. The sub-decisions aren’t separable because they all feed back into identity. The decisions that can’t be decomposed are the ones where KPIs fail, because the KPI assumes the dimension is independent.
Ambiguity. Some decisions have clear right answers if you do the analysis. Some have no right answer even with perfect information, because they involve genuine tradeoffs between incommensurable values: retaining a high-performing executive who is toxic to the team vs. losing their output; investing in a market where the upside is large but the regulatory risk could be existential; honoring a commitment to a partner that is no longer in your strategic interest; accepting a lower valuation now vs. waiting for conditions that may not materialize; cutting a product that customers love but that loses money and distracts from the core business. You can’t put “team health” and “individual performance” on the same axis. The second kind of decision is where executive judgment actually matters and where analytical rigor matters most, because the judgment needs to be informed by the full landscape of tradeoffs, not just the one axis that feels most urgent.
Time horizon. Some decisions play out in weeks. Some play out over years or decades. The longer the horizon, the more the assumptions can drift and the harder it is to evaluate the decision against outcomes. But time horizon has its own complications. Some decisions have multiple time horizons simultaneously: an acquisition has a near-term integration horizon (6-18 months), a medium-term value-creation horizon (2-5 years), and a long-term strategic positioning horizon (5-10+ years). The decision can look right on one horizon and wrong on another, and which horizon you evaluate against determines whether you think the decision was good. Some decisions have hidden time horizons: a technology choice that looks like a 2-year decision turns out to be a 10-year decision because the migration cost means you’re living with it long after the original rationale expired. Some decisions create time pressure on other decisions: committing to a delivery timeline accelerates every downstream decision (staffing, procurement, design tradeoffs) whether those decisions are ready to be made or not. And the people who made the decision are often gone before the consequences arrive. A CEO who commits to a capital structure lives with the quarterly results; the successor lives with the structural constraints. The feedback loop between decision-maker and consequence is broken by tenure, which means the organization can’t learn from the decision even if it tracks the outcome.
Stakeholder complexity. Some decisions affect one group. Some affect many groups with competing interests (investors, customers, employees, regulators, partners). The more stakeholders, the more tradeoffs that can’t be optimized simultaneously. But the real complexity is not the number of stakeholders; it’s the structure of their relationships. Some stakeholders have veto power they rarely exercise until the moment it matters: a regulator who has been permissive for years can change posture overnight, and every decision you made assuming the prior posture is now exposed. Some stakeholders have interests that are aligned in the short term and divergent in the long term: investors and employees both want the company to grow, until the growth requires layoffs or margin compression, at which point you discover the alignment was contingent. Some stakeholders are invisible until the decision is made: a pricing decision surfaces a customer segment you didn’t know you had, a partnership decision surfaces a competitor relationship your partner didn’t disclose. Some stakeholders are internal but act like external ones: a business unit leader whose team is affected by a strategic decision will optimize for their unit’s survival the same way an external partner would, and the fact that they’re inside the organization doesn’t make their incentives aligned with the whole. And the stakeholder map changes over time. The investors who supported the IPO have different interests than the investors who bought in afterward. The customers who signed up for one product have different interests than the customers who arrived for the pivot. The decision was made for one set of stakeholders and is now being evaluated by a different set.
Information completeness. Some decisions can wait for more data. Some can’t because the window closes. The decision to act on incomplete information is itself a decision. But information completeness has its own pathologies. Sometimes more information makes the decision harder, not easier: additional data introduces contradictions, surfaces edge cases, and reveals complexity that the simpler picture didn’t show. The team that waits for one more quarter of data sometimes discovers that the new data made them less certain, not more. Sometimes the information exists but is distributed across people who don’t talk to each other: the sales team knows the customer is unhappy, the finance team knows the margin is eroding, the engineering team knows the architecture won’t scale, and nobody has assembled those three facts into the conclusion that the product line should be shut down. Sometimes the information is available but politically expensive to surface: the analysis that shows the CEO’s initiative is failing exists in the data, and nobody wants to be the person who presents it. Sometimes information completeness is an illusion: the data is comprehensive but the model is wrong. You have complete information about the past and the wrong framework for interpreting what it means for the future. And sometimes waiting for more information is itself the decision, made by default. The option to enter a market, acquire a company, or recruit a key person has a window, and the team that is “still gathering data” when the window closes made a decision without ever acknowledging it as one.
Interdependence. Some decisions are independent. Some are coupled to other active decisions in ways that aren’t visible until one of them changes. A capital structure decision constrains growth options which constrain technology choices which constrain talent requirements. But interdependence has several forms that are easy to miss. Latent coupling: two decisions that were made independently years apart turn out to share a common assumption, and when that assumption breaks, both decisions fail simultaneously. The pricing model assumed a cost structure that the technology migration changed, and nobody connected the two until margins collapsed. Temporal interdependence: the order in which decisions are made constrains the options available for subsequent decisions, and the organization often doesn’t realize that making decision A first foreclosed the best option for decision B. The sequence felt natural but was actually a commitment. Cross-organizational interdependence: a decision made by one company in a partnership or supply chain constrains the decisions available to the other. Your vendor’s architecture choice becomes your migration cost. Your customer’s budget cycle becomes your revenue timing. Interdependence with decisions not yet made: committing to a strategy implicitly commits you to a set of future decisions you haven’t analyzed yet. The decision to build a platform implies decisions about governance, pricing, partner management, and ecosystem investment that will arrive whether you’ve thought about them or not. And interdependence with decisions made by competitors: your strategy assumes a competitive landscape, and your competitor’s decisions change that landscape. The decision to enter a market at a certain price point assumes the incumbent won’t respond, and the incumbent’s response changes the economics of every assumption in your pre-decision document.
Precedent effect. Some decisions set precedent that constrains future decisions. The first deal you do at a certain price becomes the reference point for every subsequent deal. The organizational structure you choose creates constituencies that resist restructuring. But precedent operates through several mechanisms that are worth distinguishing. Anchoring: the first number in a negotiation, the first term in a contract, the first exception to a policy sets a reference point that every subsequent conversation orbits. You can argue that the first deal was unique circumstances, but the other side will always start from what you agreed to last time. Constituency creation: every decision creates beneficiaries who will resist changing it. A business unit that exists because of a strategic decision three years ago will fight for its survival regardless of whether the original rationale still holds. The decision created the constituency, and the constituency now defends the decision. Narrative lock-in: the story the organization tells about why it made a decision becomes part of the culture. “We’re the company that does X” started as a strategic choice and became an identity. Challenging the decision now means challenging the identity, which is a much harder conversation. Expectation setting with external parties: customers, investors, partners, and regulators all form expectations based on your decisions. A company that has always prioritized growth over profitability has trained its investors to expect that; switching to profitability is not just a strategic decision but a re-education of the market, with the stock price as the tuition. And precedent compounds across decisions: each precedent-setting decision narrows the corridor for the next one. After enough precedents, the organization discovers that its “strategic options” are actually a single path that was determined by a sequence of individually reasonable choices that nobody evaluated as a sequence.
The hardest decisions occupy the extremes on all of these simultaneously: low frequency, irreversible, non-decomposable, high ambiguity, long time horizon, many stakeholders with competing interests, incomplete information with a closing window, coupled to other active decisions, and precedent-setting. Should we go public. Should we do this acquisition. Should we restructure the organization. Should we exit this market. Should we change the capital structure.
But the extremes are potentially the rarest. They are almost theoretical. The decisions that actually cause damage are the messy ones in the middle, and there are a lot of them. They don’t show up on the agenda as “strategic decision requiring analysis.” They show up as budget reviews, contract renewals, staffing discussions, product roadmap prioritization. The analytical depth they require doesn’t match the process they’re embedded in.
How Decisions Fail
The decision that was made by default. Nobody decided to keep investing in the product line. It just kept getting funded because the budget process is incremental: last year’s number plus or minus a percentage. The product lost its strategic rationale two years ago when the market shifted, but stopping requires an affirmative choice with visible costs (layoffs, customer migration, write-downs), while continuing requires no choice at all. The decision to continue was never made. It was never examined. The cost compounds quietly until a crisis forces the question, at which point the exit is far more expensive than it would have been two years earlier.
The decision where the information existed but was never assembled. The sales team knows the largest customer is unhappy and evaluating alternatives. Finance knows the margin on that customer’s contract is eroding because of scope creep that was never repriced. Engineering knows the custom features built for that customer are creating technical debt that slows development for everyone else. Legal knows the contract renewal is in six months with an auto-renew clause. Each team holds one piece. Nobody has assembled the four pieces into the conclusion: we are about to auto-renew a money-losing contract with an unhappy customer on terms that were set before the scope changed, and the technical debt we’re carrying for them is taxing every other product. The decision to renegotiate, restructure, or exit the relationship needed to be made three months ago. It wasn’t, because it didn’t belong to any single team.
The decision where the precedent narrowed the corridor. The first enterprise deal was priced aggressively to land the logo. That price became the reference point for the next deal, and the next. Each negotiation started from “what did we charge the last customer?” rather than “what is this worth?” By the tenth deal, the pricing was structurally below the cost to serve at scale, but raising prices meant renegotiating with every existing customer who would point to what they were already paying. The corridor narrowed with each deal. The original pricing decision felt tactical (“let’s win this one account”). The compounding effect was strategic (the entire pricing architecture is now anchored to a number that was never meant to be permanent).
The decision where alignment was contingent. The board and the management team agreed on the growth strategy. That alignment rested on an implicit assumption: that the growth would be self-funding within four quarters. It wasn’t. At quarter five, the board’s priority shifted from growth to capital preservation, but the management team had already made hiring commitments, signed leases, and entered markets based on the growth mandate. The alignment that authorized those commitments no longer exists, but the commitments do. The management team is now executing a strategy that its governing body no longer supports, and the conversation about the change hasn’t happened because nobody wants to admit that the original assumption was wrong.
The decision where the model was wrong. The analysis was thorough. The data was comprehensive. The scenarios were well-constructed. The conclusion was clear. And the underlying model of how the market works was wrong. The company modeled the competitive landscape as a market share game (invest to grow share, then extract margin) when it was actually a platform dynamics game (the winner is determined by ecosystem effects, not share). Every number in the analysis was correct given the model. The model didn’t correspond to reality. The decision looked rigorous and was rigorous within its framework. The framework was the wrong one. This is the most dangerous failure mode because it is invisible from inside the analysis. The numbers all check out. The logic is internally consistent. The conclusion follows from the premises. The premises are wrong.
The decision where the organization optimized for the wrong level. Each business unit optimized its own P&L. Each executive hit their KPIs. The company missed the strategic opportunity because the opportunity required cross-unit investment that would have hurt every individual P&L in the short term while creating value at the portfolio level that no individual unit could capture. The decision to pursue or not pursue the opportunity was never made at the right level. It was implicitly rejected by the incentive structure, one budget meeting at a time, by executives who were behaving rationally given their scorecards.
The decision where the timing was the decision. The analysis said to enter the market. The analysis was correct. But the team spent four months refining the analysis, and by the time they were confident enough to commit, the window had moved. A competitor entered. Talent that was available was hired elsewhere. A regulatory change made the entry more expensive. The decision to enter the market was right. The decision to spend four months making sure was the actual decision, and it was wrong. The cost of certainty exceeded the cost of being early and approximately right.
The decision where the wrong people were in the room. The technology migration decision was made by the technology leadership team. It was a sound technical decision. It did not account for the commercial implications (migration downtime during the highest-revenue quarter), the contractual implications (SLAs that would be violated during the transition), or the organizational implications (the team with the most institutional knowledge about the legacy system was the team being restructured). The decision was well-made within its frame. The frame was too narrow because the people who understood the other dimensions weren’t consulted, and the process didn’t require them to be.
A 2×2 matrix tells you the decision is “high impact, high uncertainty” and then leaves you alone with it. The actual structure of why it’s hard is what you need to understand before you can do any useful analytical work. Otherwise you’re doing analysis on a simplified version of the decision while the real decision, with all its hidden mechanisms, proceeds without you.
These are the decisions that define the trajectory of the company. They are also the decisions that get the least analytical rigor, because the pre-decision analysis required to do them well was too expensive to produce.
The Recurring Decision Sets
People often mistake “a” decision for “the” decision. Most strategic decisions are not one-time events. They are recurring questions that the organization revisits as conditions change. The market entry decision from three years ago is now the “should we still be in this market” decision. The org structure from last year is now the “is this structure working” decision.
Most organizations have a common set of these. They are worth naming because they are the decisions that actually determine the trajectory of the business, and most of them are being made implicitly rather than through any structured process.
Market and product. What markets are we in? Should we enter a new one? Should we exit one? What is our product strategy (build, buy, or partner)? How do we price, and what is the pricing architecture? Who is the customer? That last one gets revisited more than people admit; the answer drifts over time and the organization sometimes doesn’t notice. The hidden dynamics: the market you entered three years ago has changed, and the decision to stay is a different decision than the decision to enter, but nobody treats it that way. The customer you optimized the product for is no longer your most valuable customer, but the product roadmap still reflects their priorities because the original decision created a constituency (the team that serves them, the features built for them, the revenue attributed to them) that resists reorientation. The pricing architecture was set when the cost structure was different, and repricing means renegotiating relationships that were built on the old terms. Build vs. buy vs. partner looks like a product decision but is actually a capability decision: what you build in-house determines what you can do next, what you can hire for, and what you understand about your own business. Outsourcing a capability is easy; rebuilding it when you realize you need it is not.
Capital and resource allocation. How do we allocate capital across competing priorities? What is the capital structure (debt vs. equity vs. cash flow)? Build vs. lease vs. acquire? When do we raise, and at what terms? The hidden dynamics: capital allocation is where strategy actually lives, regardless of what the strategy document says. The budget reveals the real priorities; the strategy deck reveals the aspirational ones. When they diverge, the budget wins. There is an art to converting strategic intent into a budget that actually reflects it, to making the implicit explicit and keeping the two aligned. The gap between what the strategy says and what the budget funds is where strategic decisions get made by default. These decisions interact with each other in ways that are not visible when they are made independently by different executives: the debt structure constrains the growth rate, the growth rate determines the hiring plan, the hiring plan determines what capabilities exist, and the capabilities determine which markets are accessible. Changing any one of these requires revisiting all of them, but the organization makes them in sequence across different meetings with different people, so the interactions go unexamined. The decision to raise capital at a particular valuation sets expectations that constrain every subsequent operating decision: you have now committed to a growth trajectory that justifies the valuation, whether or not the market supports it. And the allocation process itself creates distortions: the business unit that is best at making the case for capital is not necessarily the one where capital creates the most value, and the process can reward narrative skill as much as analytical rigor.
People and organization. What is the organizational structure (functional, business unit, matrix, something else)? Who leads what? What do we do ourselves vs. outsource? When does a team need to be restructured vs. when does it need more time? The hidden dynamics: the executive team composition question is itself a strategic decision, not an HR decision. Who sits in which seat determines what information flows where, which perspectives are represented in the room, and what the organization pays attention to. A CTO who came from infrastructure sees different risks than a CTO who came from product. A CFO who came from banking has different instincts about capital structure than a CFO who came from operations. The organizational structure creates information channels and blocks others: a functional structure means the product person talks to all the engineers but no single engineer sees the full product; a business unit structure means each unit sees its own customer but nobody sees the cross-unit opportunity. Every reorg is an attempt to fix information flow, and every reorg creates new blind spots. The decision to outsource a function looks like a cost decision but is actually a learning decision: you stop learning about the thing you outsourced, and that learning gap compounds over time until you’re dependent on a vendor for something you no longer understand well enough to evaluate.
Technology and infrastructure. What platforms do we build on? When do we migrate vs. maintain? What do we standardize vs. where do we preserve optionality? Build internal capability vs. buy a vendor solution? The hidden dynamics: technology decisions have the longest half-lives and the highest switching costs of any decision set, which means getting them wrong compounds more than getting any other category wrong. But they are often made by the technology team alone, without the commercial, financial, or organizational context that determines whether the decision is right for the business rather than right for the technology. A platform migration that is technically correct can be commercially catastrophic if the timing conflicts with revenue commitments. A standardization decision that reduces engineering complexity can eliminate the optionality that the business strategy depends on. The build vs. buy decision has a structural bias: building creates work for the team evaluating the decision, and buying requires acknowledging that someone else solved the problem. The incentives of the evaluator are not neutral. And technology decisions create path dependencies that outlast the people who made them: the architecture chosen five years ago constrains what the current team can build, and the current team may not understand why the architecture was chosen because the people who chose it are gone and the decision was never documented.
Risk and governance. What risks do we accept vs. mitigate vs. transfer? What is our regulatory strategy (comply, shape, avoid)? What is our IP strategy? When do we litigate vs. settle? The hidden dynamics: risk decisions can easily become reactive rather than strategic. The organization discovers a risk and responds to it rather than maintaining a model of which risks it is carrying and why. The regulatory strategy is often implicit (whatever the legal team recommends on a case-by-case basis) rather than explicit (we have decided to comply in these areas, shape in these areas, and avoid these areas entirely, and here is why). The IP strategy is often downstream of what the engineering team happened to build rather than upstream of what the business needs to protect. Litigation decisions can default to the legal team optimizing for legal risk, when the decision also involves commercial relationships, reputational impact, and management attention that the executive team needs to weigh. And the biggest risk decisions are the ones that don’t look like risk decisions at all: the decision to concentrate revenue in a small number of large customers, the decision to depend on a single supplier for a critical component, the decision to operate in a jurisdiction with an unstable regulatory environment. These are business decisions with risk implications that are rarely evaluated as risk decisions.
Growth and trajectory. Organic vs. inorganic growth? When do we go public (if private)? Should we go private (if public)? What is the partnership strategy? When do we divest? The hidden dynamics: the organic vs. inorganic growth question tends to be influenced by the organization’s history. An organization that has grown through acquisition has the muscle memory, the integration playbooks, and the deal team to do it again. An organization that has grown organically has the culture, the internal development capacity, and the identity tied to building from within. Neither is wrong; the risk is that the historical pattern becomes the default rather than the analysis. The IPO decision is treated as a milestone rather than a strategic choice, when it is actually one of the most consequential structural decisions the company will make: it changes the governance, the information disclosure requirements, the time horizon of the investor base, the compensation structure, and the management attention allocation. Going private after being public is even harder because the public market created expectations, employee equity plans, and a valuation reference point that all have to be unwound. Partnership strategy is often reactive (someone proposes a partnership and the team evaluates it) rather than proactive (we have identified the capabilities we need and the partners who could provide them). And divestiture is the growth decision nobody wants to make, for the same reason stopping is hard: the business being divested has people, customers, and revenue, and the case for keeping it is always easier to make than the case for the organizational focus that divestiture enables.
Stopping. What do we stop doing? What do we sunset vs. sell vs. wind down? When is a strategy not working vs. not working yet? This is almost always the most underasked set, and the hidden dynamics explain why. Starting things has champions: someone’s career is tied to the launch, the team is energized, the board presentation is optimistic. Stopping things has costs that are visible (layoffs, write-downs, customer migration, admitting the strategy was wrong) and benefits that are diffuse (organizational focus, freed capital, reduced complexity). The asymmetry means the bar for starting is lower than the bar for stopping, so the organization accumulates commitments over time. The question “when is a strategy not working vs. not working yet” is almost impossible to answer honestly inside the organization, because the people closest to the strategy are the people whose careers depend on it working. The sunk cost is not just financial; it is emotional, reputational, and political. And the longer you wait to stop, the more expensive stopping becomes, because the strategy has created constituencies, dependencies, and expectations that all have to be unwound. When stopping is deferred until a crisis forces it, the organization pays the maximum cost for the minimum optionality.
These decisions don’t all happen at the same speed. Some take months because you need to see how elements play out before committing. Some have to be done quickly (“should we stop doing this?” can have real urgency). They don’t all have to be actively in play at any given time, and some don’t need to be revisited for years. But they all need to have answers, even if the answer is “we looked at this and decided not to act.” And they have a lifecycle: should we investigate this further? Should we commit? Now that we’re executing, should we adjust? Should we stop? Each transition in that lifecycle is itself a decision, and each one can be supported by its own analytical work.
A useful exercise: put the full list in front of an executive team and ask, for each one, “do we have a clear, current, documented answer?” Not whether the decision was made, but whether the team can articulate what was decided, when, on what basis, and what would have to change to revisit it. The questions that don’t have clear answers are the decisions that are being made by default.
The Future of Executive Decision Support
Most of what passes for decision support in organizations is not decision support. It is a recommendation. Someone does analytical work, arrives at a conclusion, and presents it to leadership or a board as a single proposal: we should enter this market, we should do this acquisition, we should restructure the organization. The decision-maker is asked to approve or reject a conclusion they cannot independently evaluate. They have no visibility into the assumptions the analysis rests on, the alternatives that were considered and why they were dismissed, the dimensions of the decision that the analysis did not address, or the interactions with other active decisions the analyst may not have known about. A single recommendation is not decision support. It is a request for ratification.
The prerequisite for actual decision support is being able to see the decision: naming its dimensions, its hidden dynamics, and the mechanisms by which it interacts with other decisions. That is what this post has done. The question is what becomes possible once you can see it.
Executive decision support today means dashboards, slide decks, and the occasional consultant engagement. For operational decisions, that is adequate. For the decisions described in this post (the nine dimensions, the messy middle, the recurring sets with their hidden dynamics) it is not. A dashboard can tell you revenue is down. It cannot tell you that the revenue decline is coupled to a capital structure decision made two years ago, that the pricing precedent from the first enterprise deal has narrowed your corridor, and that the team evaluating the problem doesn’t include the people who understand the contractual implications.
What executive decision support could become is different. The analytical work described across these dimensions (research, data architecture, quantitative modeling, cross-domain analysis, scenario construction, assumption tracking) is becoming feasible at a speed and cost that it wasn’t before. The data exists in the organization’s own systems. The reasoning capability to hold a board deck, three quarters of financial data, a delivery pipeline, and a competitive landscape in a single context and identify interactions across them exists. The computational tools to turn assumptions into distributions of outcomes rather than point estimates exist.
But this is not one thing called “AI.” It is two layers. Deterministic computation (simulation, optimization, financial algebra) produces verifiable answers. Frontier model reasoning interprets what those answers mean, formulates what to model next, and synthesizes across domains that no individual analyst holds in context simultaneously. Multiple independent frontier models from different families cross-validate each other’s analysis: what one misses or overweights, another catches. The rigor comes from the structure: computation that proves itself, reasoning that is checked by independent perspectives.
The possibility is real decision support for people who exercise judgment: not a system that makes the decision, but analytical infrastructure that helps the decision-maker see the full dimensionality of what they’re deciding. The constraints, the scenarios, the hidden dynamics, the assumptions, the interactions between this decision and every other active decision. The pre-decision analysis that was too expensive to produce, done before the decision rather than as a rationalization afterward.
This does not replace judgment. The ambiguity doesn’t disappear. The tradeoffs between incommensurable values still require a human willing to choose: which questions to ask, which axioms to set, which direction matters. What changes is whether the person making the call can see the full landscape or is working from a 10-page deck that has already flattened it. And because the frontier reasoning sits in the critical path rather than behind an abstraction layer, this capability improves when the models improve. The infrastructure amplifies model capabilities. It does not replace them.
The structure of what the decisions require has not changed. The cost of doing the analytical work has. The question is whether the people who exercise judgment will have the support that matches what the decisions actually demand.