Decision speed as a variable

The speed of the decision is itself a decision

You already know how to make better decisions. You can list pros and cons, weight criteria, consult stakeholders, run pre-mortems, and gather data until your confidence hits 95%. None of that is the problem.

The problem is that you apply the same deliberation process to every decision, regardless of whether speed or accuracy is the dominant variable. And in a startling number of cases, the cost of slow is higher than the cost of wrong.

This lesson is about treating decision speed not as a personality trait or a reckless impulse, but as a deliberate variable — one you set explicitly before analysis begins, based on reversibility, stakes, and the cost of delay.

The 70% rule: Bezos and the two-way door

In his 2016 letter to Amazon shareholders, Jeff Bezos drew a distinction that reshapes how you think about decision velocity. He categorized all decisions into two types:

Type 1 decisions are irreversible — one-way doors. You walk through, the door closes behind you, and you live with the consequences. These deserve heavyweight process: slow deliberation, broad consultation, deep analysis. Bezos wrote that Type 1 decisions "must be made methodically, carefully, slowly, with great deliberation and consultation."

Type 2 decisions are reversible — two-way doors. You walk through, look around, and if you don't like what you see, you walk back. These should be made quickly, by individuals or small teams, with minimal process overhead.

The pathology Bezos identified was not that people make bad decisions — it's that organizations treat most decisions as Type 1 when they're actually Type 2. They apply irreversible-decision process to reversible-decision situations, and the result is institutional paralysis.

Then came the rule that makes senior leaders uncomfortable: "Most decisions should probably be made with somewhere around 70% of the information you wish you had. If you wait for 90%, in most cases, you're probably being slow."

This isn't anti-intellectual. It's a recognition that information has diminishing returns and delay has compounding costs. The difference between 70% confidence and 90% confidence might take weeks to close — and during those weeks, the opportunity is shrinking, the team is stalled, and the competitive landscape is shifting. Bezos added the essential corollary: "If you're good at course correcting, being wrong may be less costly than you think, whereas being slow is going to be expensive for sure."

The implication is structural: you need to build systems that make course correction cheap. When reversing a decision costs almost nothing, the rational move is to decide fast, observe the results, and adjust. When reversing costs everything, the rational move is to slow down and get it right. The variable isn't your confidence — it's the cost of being wrong relative to the cost of being slow.

Operating inside the loop: Boyd and the OODA cycle

Colonel John Boyd was a U.S. Air Force fighter pilot who earned the nickname "40-Second Boyd" because he could defeat any opponent in air combat simulation within forty seconds. His advantage was not superior aircraft or physical reflexes — it was decision tempo.

Boyd formalized his insight into the OODA loop: Observe, Orient, Decide, Act. Every agent — a pilot, a company, a person navigating a career — cycles through this loop continuously. You observe the environment, orient yourself within it (filtering through experience, culture, and mental models), decide on a course of action, and act. Then the cycle restarts with new observations shaped by the consequences of your action.

Boyd's core insight was not that faster is always better. It was that operating at a higher tempo than your adversary creates a compounding advantage. When you cycle through OODA faster than the other side, you "get inside their loop" — your actions change the environment before they've finished orienting to the last change. Boyd described the effect precisely: "The ability to operate at a faster tempo or rhythm than an adversary enables one to fold the adversary back inside himself so that he can neither appreciate nor keep up with what is going on. He will become disoriented and confused."

This applies far beyond combat. In markets, the company that ships, learns, and iterates faster than competitors forces those competitors into a reactive posture. In your personal epistemology, the person who makes a decision, observes the outcome, extracts the lesson, and makes the next decision faster than their own doubt cycle is the one who accumulates experience at a rate others cannot match.

But Boyd was precise about something most popularizations miss: the critical phase is not "Decide" or "Act" — it's Orient. Orientation is where your mental models, prior experience, and pattern recognition converge to make sense of what you've observed. A faster OODA loop isn't about acting before thinking. It's about building such deep orientation capability that the thinking happens in compressed time. Speed without orientation is just flailing. Speed with orientation is mastery.

Eisenhardt: fast decision makers use more information, not less

Kathleen Eisenhardt's 1989 study, "Making Fast Strategic Decisions in High-Velocity Environments," published in the Academy of Management Journal, demolished the assumption that fast decisions are sloppy decisions. She studied eight firms in the microcomputer industry — an environment where product cycles were measured in months and strategic windows opened and closed within quarters.

Her findings were counterintuitive. The executive teams that made the fastest strategic decisions:

• Used more real-time information, not less. They tracked operational metrics obsessively — not as analysis paralysis, but as orientation infrastructure. They always knew where they stood, which meant when a decision point arrived, they didn't need weeks of data gathering. The data was already in their heads.
• Considered more simultaneous alternatives, not fewer. Fast deciders didn't narrow to one option and agonize. They generated multiple options in parallel, compared them against real-time data, and selected. Parallel evaluation is faster than serial evaluation because comparison reveals differences that isolated analysis obscures.
• Relied on experienced counselors. Fast decision makers had one or two trusted advisors — not large committees, not consensus processes, but specific individuals whose judgment they'd calibrated over years. This is a speed optimization: instead of polling a room, you consult someone whose pattern recognition you trust.
• Used active conflict resolution. When the team disagreed, fast deciders didn't wait for consensus or avoid the conflict. They surfaced the disagreement, argued it out, and then the leader decided. "Consensus with qualification" — try for agreement, but if it doesn't come quickly, the senior person calls it.

The performance correlation was stark. Fast-deciding firms outperformed slow-deciding firms across the board. Several of the slow firms failed entirely. Eisenhardt's conclusion was that in high-velocity environments, speed and quality are not tradeoffs — they're correlated. The practices that make you fast (real-time information, parallel alternatives, experienced counsel) are the same practices that make you accurate.

Recognition-primed decisions: why experts decide fast and well

Gary Klein's research on naturalistic decision making explains the cognitive mechanism behind expert speed. Klein studied firefighters, emergency medical technicians, military commanders, and other professionals who routinely make high-stakes decisions under time pressure.

He found that experienced decision makers don't compare options. They don't weigh pros and cons. They don't run decision matrices. Instead, they recognize patterns. A fireground commander looks at a burning building and immediately sees a situation type — the fire behavior, structural cues, and environmental conditions trigger a pattern stored in years of experience. That pattern comes with a course of action attached. The commander mentally simulates that action, checks whether it will work in this specific case, and if the simulation holds, acts. Klein called this the Recognition-Primed Decision (RPD) model.

In his studies of firefighters, 78% of decisions were made in under one minute. Not because the firefighters were reckless — because their pattern libraries were so deep that recognition replaced analysis. The 70% information rule works for experts precisely because their orientation (in Boyd's terms) or their pattern library (in Klein's terms) fills in the remaining 30%.

This has a direct implication for your own decision practice: the more decisions you make and review, the larger your pattern library grows, and the faster your future decisions become without sacrificing accuracy. Speed is not the enemy of quality — it's the product of accumulated, reviewed experience. The person who has made and reflected on a thousand hiring decisions recognizes the pattern faster than the person agonizing over their fifth.

The cost of delay: what slow decisions actually cost

Most people evaluate decisions by outcome quality alone. They ask: "Did I choose correctly?" They rarely ask the more important question: "What did the delay cost?"

Consider a product team choosing between two feature architectures. Option A has a 60% chance of being the right long-term choice. Option B has a 40% chance. But the team has been deliberating for three weeks, and every week of delay costs $50,000 in lost user engagement and competitive positioning. The expected cost of choosing wrong (say, a $100,000 migration later) is $40,000 (40% chance times $100,000). The cost of the three weeks already spent deliberating is $150,000. The math is obvious in retrospect, but in the moment, the team was focused on "making the right call" without accounting for the cost of the call taking too long.

This is a failure of framing. When you think about decisions only in terms of outcome quality, speed looks like recklessness. When you include delay costs, speed becomes rational — sometimes the most rational option available.

Here's a useful heuristic: for any decision, estimate the cost of one additional day of deliberation and the expected improvement in outcome quality that day would produce. When the cost exceeds the improvement, you've passed the optimal decision point. Every day after that, you're paying for certainty you don't need.

The AI parallel: latency versus accuracy

If you work with AI systems — or even if you just use them — you've encountered this same tradeoff in engineered form. Every large language model faces a fundamental tension between latency and accuracy.

A model can take more time — running more computation, considering more tokens, applying more reasoning steps — to produce a more accurate response. Or it can stream the first token fast, delivering partial results immediately while continuing to generate. The optimization target is not maximum accuracy. It's the right balance of accuracy and responsiveness for the specific use case.

This is called the latency-accuracy tradeoff, and it mirrors the human decision speed variable precisely. Time to First Token (TTFT) — how fast the model begins responding — determines perceived responsiveness. Total latency determines how long you wait for the complete answer. Techniques like streaming inference, model distillation (training a smaller, faster model to approximate a larger one), and speculative decoding all sacrifice marginal accuracy for substantial speed gains.

The engineering insight maps directly to your epistemic practice: the optimal decision process is not the one that maximizes accuracy in isolation. It's the one that maximizes value delivered per unit of time, accounting for both the quality of the output and the cost of waiting for it. Just as an AI system that takes ten minutes to return a perfect answer is often less useful than one that returns a 90% answer in two seconds, a decision that arrives with 95% confidence next month is often less valuable than one that arrives with 70% confidence today.

The speed-accuracy dial: a calibration framework

You don't need to choose between fast and careful as a permanent identity. You need a dial — a way to set the speed-accuracy balance for each specific decision. Here's the calibration:

Turn the dial toward speed when:

• The decision is reversible (Type 2 / two-way door)
• The cost of delay is high and visible (market window, team blocked, opportunity expiring)
• You're at 60-70% confidence and the remaining information would take disproportionate time to gather
• Your pattern library for this type of decision is deep (you've made similar calls before)
• The downside of being wrong is bounded and recoverable

Turn the dial toward accuracy when:

• The decision is irreversible or extremely expensive to reverse (Type 1 / one-way door)
• The cost of delay is low relative to the cost of being wrong
• You're below 50% confidence and critical information is readily available
• This is a novel decision type with no pattern library to draw from
• The downside of being wrong is catastrophic or affects many people

The meta-skill is setting the dial before analysis begins. If you start analyzing without deciding how long to analyze, you will default to your personality — slow deliberators will over-deliberate, fast movers will under-examine. Neither is correct. The correct speed is the one matched to the specific decision's reversibility, stakes, and delay cost.

What this looks like in practice

You wake up on a Tuesday with three decisions pending:

1. Which health insurance plan to choose during open enrollment (closes Friday). The plans differ by $40/month and have slightly different networks. This is a Type 2 decision — you can switch next year. The cost of delay is low but bounded by the deadline. Set the dial to speed. Spend thirty minutes comparing the two most relevant dimensions (your current doctors and the premium difference), pick one, and move on. Confidence: 65%. Good enough.

2. Whether to accept a job offer that requires relocating your family. This is a Type 1 decision — irreversible in practice, affecting multiple people, with life-altering consequences. The cost of being wrong vastly exceeds the cost of taking an additional week to think. Set the dial to accuracy. Schedule calls with people who've made similar moves. Visit the city. Run the financial model. Confidence target: 85%+.

3. Which JavaScript framework to use for a new internal tool. The team has spent two weeks debating React versus Svelte. Both would work. The internal tool serves fifty people. This is a Type 2 decision — you can migrate later if needed, and the blast radius is small. The cost of delay (two weeks of a team not building) already exceeds any possible quality difference between the frameworks. Set the dial to maximum speed. Pick one in the next meeting, commit, and build.

Three decisions, three different speeds, all rational. The variable isn't your temperament. It's the structure of the decision.

The connection to your epistemic infrastructure

This lesson sits between Kill criteria (L-0456) and Post-decision review (L-0458) for a reason. Kill criteria give you pre-commitment to abandon failing paths — which makes fast decisions safer, because you know you'll catch errors early. Post-decision review gives you the feedback loop that grows your pattern library — which makes future fast decisions more accurate.

Together, the three form a cycle: decide at the right speed, monitor against pre-set kill criteria, review the outcome to improve future calibration. Each pass through this cycle increases your decision throughput without decreasing your decision quality. Over time, you become the person Eisenhardt described — someone who decides fast because they have deep real-time awareness, not despite lacking it.

The primitive bears repeating: sometimes deciding fast is more important than deciding optimally. Not always. Not recklessly. But more often than your instinct for certainty wants to admit.

The next time you catch yourself in extended deliberation, ask the calibrating question: "Is the cost of this delay higher or lower than the expected value of the additional information I'm seeking?" If the delay costs more — and it often does — decide now, act, observe, and correct. That is not impulsiveness. That is speed as a deliberate, rational, calibrated variable in your decision framework.

---

Sources

1. Bezos, J. (2016). Letter to Amazon Shareholders. — Type 1/Type 2 decision framework and the 70% information rule.
2. Boyd, J. (1976, 1986). OODA Loop / "Patterns of Conflict" briefings. — Decision tempo, operating inside the opponent's loop, orientation as the critical phase.
3. Eisenhardt, K. M. (1989). "Making Fast Strategic Decisions in High-Velocity Environments." Academy of Management Journal, 32(3), 543-576. — Empirical evidence that fast strategic decision makers use more information and outperform slow deciders.
4. Klein, G. (1998). Sources of Power: How People Make Decisions. MIT Press. — Recognition-Primed Decision model, naturalistic decision making, expert pattern recognition under time pressure.
5. Eisenhardt, K. M. (1990). "Speed and Strategic Choice: How Managers Accelerate Decision Making." California Management Review, 32(3), 39-54. — Extended analysis of decision acceleration practices.
6. Cowan, N. (2001). "The magical number 4 in short-term memory." Behavioral and Brain Sciences, 24(1), 87-114. — Working memory constraints that make externalization necessary for parallel option evaluation.
7. NVIDIA Technical Blog (2024). "Mastering LLM Techniques: Inference Optimization." — Latency-accuracy tradeoff, streaming inference, Time to First Token as a design variable.