Consensus is often treated as a costless benefit — getting everyone to agree is good, and the process of getting there is just friction. This is wrong. Consensus has a real cost that is usually invisible because it comes in the form of information destroyed, not resources spent.
When a group reaches consensus, it produces one view from many. The aggregation process systematically destroys the dissenting signal. The person who thought the project was misconceived, who had a specific technical objection, who had seen this pattern fail before — their input gets smoothed into the consensus output. Their signal is lost.
This is not a problem when the dissenting view is noise. It's a catastrophic problem when the dissenting view is right.
The organizational behavior literature documents this extensively under the heading of "groupthink." The mechanism is well-understood: social pressure, desire for cohesion, and the cost of continued dissent cause individuals to update toward the group position faster than the evidence warrants. The consensus that emerges reflects the social dynamics of the group as much as the underlying reality.
Consensus is valuable when execution requires alignment and the decision is reversible. Getting a team to agree on a process, a naming convention, a meeting schedule — the costs of the consensus formation process are low relative to the coordination value, and being wrong is fixable.
Consensus is dangerous for decisions that are irreversible, high-stakes, or that require integrating heterogeneous expertise. These are exactly the conditions where the destroyed dissenting signal is most likely to be load-bearing.
Organizations that have figured this out don't try to eliminate consensus processes — they build parallel structures that preserve minority views. Red teams, pre-mortems, designated devil's advocates, anonymous voting before discussion. The goal is to capture the dissenting signal before social pressure destroys it.
The insight: consensus is a commitment device, not an information aggregation mechanism. Use it for the former, not the latter.