Consensus feels safe because it spreads responsibility. What it spends is founder attention, which at seed stage is scarcer than cash.
Most early-stage teams treat a decision as a social event. The default pattern is familiar: a meeting is scheduled, viewpoints are collected live, objections surface late, and the decision is delayed “until we align.” The cost is not the meeting. The cost is the latency introduced into every downstream action. When decisions stall, work becomes speculative, people start hedging, and execution fragments into half-commits.
What consensus actually costs
Three features of an early-stage environment make consensus expensive, and none of them is about personality.
Information is incomplete. You cannot debate your way into certainty. You can only choose a bet size that matches uncertainty and design for reversibility.
The blast radius is asymmetric. A small decision can block many other tasks. The time you spend aligning on one decision is paid again as a compounding delay across the rest of the roadmap.
Ownership is fragile. If everyone must agree, no one owns the outcome. That creates a predictable failure mode: people participate in discussion but avoid commitment because commitment has costs and discussion feels costless.
A seed-stage founder does not win by perfect decisions. You win by maintaining decision throughput while keeping errors cheap. Consensus is optimised for social comfort and reputational protection. You are optimising for controlled speed.
The opposite failure is just as fatal, and it looks nothing like this one. Under pressure, decisions stop spreading and start collecting on the founder instead, which is a different way for the same system to stop working.
The signals that change the decision
You do not need sophisticated data to diagnose this. You need operational signals that change decisions. These are the ones that matter.
Decision lead time. Track how many days pass between “we need a decision” and “we decided.” Resist setting a fixed threshold, because the right one scales with what the decision costs to unwind. The sharper test is how much information you insisted on first. Bezos put it at roughly 70 per cent of what you wish you had, and a team that routinely waits for ninety is not being careful. It is being slow in a way that will not show up in any meeting.
Revisiting rate. How often do you reopen decisions that were agreed to in meetings? High revisit frequency usually means the decision was never actually committed to. It was socially acknowledged, not operationally owned.
Meeting-to-decision ratio. Count how many meetings occur per real decision shipped. If the ratio drifts upward, meetings are substituting for commitment.
Late objections. If stakeholders raise their real objections after the meeting, you have a psychological safety issue or a process issue. In practice it is usually a process issue: people do not have the time or structure to think in advance, so they react late.
Work in progress on decisions. Look for shadow work created by undecided choices: parallel implementations, speculative design, half-built docs, and multiple versions. This is definition debt and decision debt showing up as rework.
These signals are useful because they point to one conclusion: the organisation is paying for alignment theatre with execution speed.
Consensus is risk avoidance, not quality
The common assumption is that consensus produces better decisions because more brains contribute.
The falsifiable reframe is this: consensus is a risk-avoidance mechanism, not a quality mechanism. Decision quality improves when you separate input collection from decision authority, then make commitment explicit in writing.
This is testable within a week. If you install a written, owner-driven decision protocol, you should see decision lead time drop while revisit rate stays the same or improves. If lead time drops but revisit rate spikes, you are deciding fast without clarity. If lead time stays high, you did not actually change authority or commitment mechanics.
The practical implication is simple: you do not eliminate input. You eliminate the requirement that input must converge into unanimous agreement before action begins.
Three decisions, not three steps
What follows is not a schedule. Each one is a decision you have been making by default.
Who decides, and does everyone know before the discussion starts? Classify every decision currently in motion as reversible, where a wrong choice is recoverable at limited cost, or irreversible, where it is expensive or damaging. Then assign a single DRI, a directly responsible individual. For founder decisions that is usually you. For team decisions it may be a function owner. Define who holds input rights, who holds a veto, and who only needs telling.
Keep vetoes rare and explicit, and never attach one to a reversible decision. Treating reversible decisions as irreversible is how consensus reappears under a different name. A two-way door that has stayed open a week is not a decision being made. It is a decision being avoided.
Where does the argument happen, in a room or on a page? Replace live debate with a one-page memo, and require the objections in writing. The memo states the decision in one sentence, the context in three to five bullets, two or three options including doing nothing, the recommendation and why, whether it is reversible and what makes it so, the kill criteria and review date, and what becomes unblocked once it is decided.
Send it to input-rights stakeholders with a comment window sized to the decision: a day for a reversible one, longer only when it is not. Anyone who disagrees has to say which assumption they believe is false, what evidence would change their mind, and what risk they think is underestimated.
This forces objections to become legible. It also shifts the burden: if someone wants to slow the decision, they must pay the cost of clarity.
What are you willing to measure, knowing it may say the change failed? Once a week, in half an hour, review what was decided, what is still pending past its deadline, and what was reversed and why. Then track two numbers and nothing else: median decision lead time in days, and revisit rate as a percentage of decisions reopened.
Two numbers is the point. A dashboard that measures everything is another way of deciding nothing.
The rule
Separate input from authority, then commit it to writing.
-
Median decision lead time
Days between “we need a decision” and “we decided.” There is no universal number, because the right one scales with what the decision costs to reverse. The test is whether you decide at around 70 per cent of the information you wanted, or hold out for ninety.
-
Revisit rate
The share of decisions reopened after they were agreed. High means the decision was socially acknowledged and never operationally owned.
-
What the pair tells you
Lead time falling while revisit rate holds means the change worked. Lead time falling while revisits spike means you are deciding fast without clarity. Lead time unchanged means you never moved authority in the first place.
None of this removes the discomfort. It relocates it. Under consensus the discomfort is spread thin and arrives late, as a decision nobody quite made and everyone can account for afterwards. Under a named owner it arrives at once, to one person, before the outcome is known. That is the whole trade, and it is why the practice stays rare in companies that can recite it.
So take the decision that has been pending alignment longest, and name its owner before the discussion rather than after it. Set the deadline by what the decision would cost to reverse, not by what feels decisive.
A decision nobody owns is not a decision. It is a meeting that has not finished.
References
- Bezos, J. (2017). 2016 Letter to Shareholders. Amazon.com.
- Dalio, R. (2017). Principles: Life and Work. Simon & Schuster.
- Drucker, P. F. (1967). The Effective Executive. Harper & Row.
- Grove, A. S. (1983). High Output Management. Random House.
- Janis, I. L. (1972). Victims of Groupthink. Houghton Mifflin.
- Kahneman, D., & Tversky, A. (1974). Judgment under uncertainty: Heuristics and biases. Science, 185(4157), 1124–1131.
- Larman, C., & Vodde, B. (2016). Large-Scale Scrum: More with LeSS. Addison-Wesley.
- Resnick, P. (2014). RFC 7282: On Consensus and Humming in the IETF. Internet Engineering Task Force.
- Simon, H. A. (1955). A behavioral model of rational choice. The Quarterly Journal of Economics, 69(1), 99–118.