Why Two Approvers Are Sometimes Worse Than One

Something went out wrong, so you added a second approver. It feels like the responsible move. Often it makes the review worse.

Two approvers reduce safety when both are checking the same thing. Each one reads less carefully because someone else is also looking, and the person who signs second tends to treat the first signature as evidence the work is fine. A second approver adds real safety only when they are checking something different, with knowledge the first approver does not have.

Diffusion of responsibility in a review chain

The pattern is old and well known outside software. When responsibility is shared without being split, each person carries less of it than they would alone.

In an approval chain it works like this. The first approver knows someone else will look, so a borderline item gets passed along rather than questioned. The second approver sees an approval already on the record and reads it as a signal. Neither is being lazy. Both are behaving reasonably given what they can see.

The result is two people who each did seventy percent of a review, which is less total scrutiny than one person doing a hundred percent.

It gets worse when the second approver is senior. Junior colleagues rarely reverse a senior name, and senior colleagues often assume the detail was checked below them. The chain looks strongest at exactly the point where it is thinnest.

When a second approver is real safety

There are cases where two is genuinely better, and they share a feature. The second person is checking something the first cannot.

• Different knowledge. The first approver checks the content is accurate. The second checks it is legally safe to send, and only one of them can do that.
• Different interest. The person who wants the work to ship should not be the only person approving it. A second approver who is neutral is a real control.
• Separation of duties. Anything involving money, access or personal data usually needs the person who prepares and the person who releases to be different, and in regulated sectors this may be required rather than optional. Requirements differ by jurisdiction and industry, so check what applies to you.
• Irreversibility. If the action cannot be undone, a second look is cheap insurance even with some diffusion.

What these have in common is that the two approvals answer different questions. If you cannot state the two questions in one sentence each, you probably have one review split across two people.

Signs your second approver is decorative

Some things to look for in your own workflow.

The second approval always follows the first within a few minutes. That is not review time, that is a click, and it is one of the clearest signs that approving has become rubber stamping.

The second approver has never rejected anything. Not rarely. Never.

Both approvers describe their job in the same words. Ask them separately what they are checking, and if the answers match, one of them is redundant.

Nobody can say what happens when they disagree. A chain with no defined disagreement path is not a control, because the practical resolution will be that the more senior or more insistent person wins quietly.

Better options than adding a person

When a mistake gets through, the reflex is to add a step. Usually there are stronger moves.

Make the single approval better. Show the change rather than the reasoning, put the risky field first, and give the approver the source they would need to check a factual claim. Most approval failures are design failures rather than attention failures, and they get rarer when the agent says it does not know instead of guessing.

Split by risk instead of doubling everything. One approver for routine items, two for the small number that cross a threshold you actually define, which is a matter of setting a confidence threshold you can explain. Doubling all approvals to catch the rare bad one teaches everyone that approval is a formality.

Name the second question out loud. If you keep the second approver, write down what they are checking and make sure it is different from the first. It also helps if the second approver cannot see the first approval until they have made their own call, though whether your system supports that varies.

The trade you are actually making

Every additional approver costs time, adds a place for work to stall when someone is away, and reduces the attention of everyone else in the chain. Those costs are certain. The benefit is only real when the second person brings something the first did not.

This is why approval design matters more than approval count in systems where agents draft and humans sign off, which is how SicherOne works: a human approves agent output before it ships, and the value of that approval comes from it being a considered decision by a named person rather than a queue of clicks.

One approver who reads carefully beats two who assume the other one did.