When to add a valve

Something goes wrong, and the reflex is to add an approval step. A second pair of eyes. A sign-off. A gate.

Gates are valves — they stop flow until a condition is met — and valves are legitimate infrastructure. But they're the most over-installed component in business, because adding one feels like taking action and the cost is paid by someone else, later, in small increments that never get attributed back to the decision.

Three tests before you install one.

Test 1: Does the approver have information the doer lacks?

The only good reason for a gate is an information asymmetry. The approver knows something the person doing the work cannot know: broader context, a legal constraint, a commitment made elsewhere.

If the approver is looking at the same information and applying the same judgment, the gate adds nothing but delay. Worse, it degrades: a reviewer who never finds problems stops looking carefully, and within two months you have a rubber stamp that everyone treats as a control. That's worse than no gate, because now the doer relies on review that isn't happening.

The honest check: what's your rejection rate? If a gate rejects nothing over fifty passes, it isn't a control. Either remove it or find out what it was supposed to be catching.

Test 2: Is the failure it prevents worse than the delay it causes?

Every gate has a price: units of work times the delay per unit.

A gate that adds four hours to a process running two hundred times a month costs eight hundred hours of elapsed time a year. Against that, what does it prevent? How often, and how bad?

Sometimes the math is obvious in favor — irreversible actions, regulatory exposure, anything that ends with a customer's money in the wrong place. Often it isn't, and nobody has done the arithmetic because the delay is diffuse and the prevented failure is vivid. Vividness beats arithmetic in most meetings, which is precisely why you should write the arithmetic down.

Test 3: Could a gauge do this job instead?

The best alternative to a gate is usually detection.

A gate stops every unit to prevent a rare failure. A gauge lets everything through and tells you immediately when something looks wrong. If the action is reversible, the gauge is almost always the better trade: you pay a small cost on the rare bad case instead of a small cost on every case.

The decision rule I use:

• Irreversible and high-impact → valve. Stop it before it happens.
• Reversible → gauge. Let it flow, watch closely, fix fast.
• Irreversible and low-impact → gauge, plus a written policy. Not everything irreversible deserves a queue.

Most gates I encounter are protecting against reversible failures. They were added after a bad week, they've never been revisited, and they've become part of how the company describes itself.

The hidden cost nobody counts

Beyond delay, gates do something worse: they move responsibility off the person doing the work.

When your judgment gets reviewed, you stop exercising judgment. You produce something that will pass review, which is a different and smaller objective. Over time the doer becomes an executor and the approver becomes a bottleneck, and the organization loses the thing it most needed — people who own their outcomes.

This is why gates proliferate so easily. Each one is locally reasonable and the aggregate is a company where nobody decides anything alone and everything takes six days.

Valves you should keep

Not an argument against all gates. Keep the ones where:

• The action is genuinely irreversible (money out, data deleted, a public commitment made).
• The approver holds real context the doer can't have.
• The rejection rate is meaningfully above zero — proof it's doing work.
• The gate has an SLA, so it's a valve and not a black hole.

That last one is non-negotiable and almost always missing. A gate without a response-time commitment isn't a control, it's a queue of unknown depth. If you can't say how long approval takes at P90, you don't have an approval step — you have a place where work goes to wait for someone's attention.

Put a gauge on your valves. Otherwise the thing you installed to make the pipe safer is the thing that's clogging it, and you won't be able to prove it either way.