Three Kinds of Keeping

Knowing what a commitment is protected from tells you which way to watch

Two companies each hold a line about credit decisions. One means that a person makes the call and no model does. The other means that the rule itself does not change, whoever is making the call. Those sound like the same commitment, and they will be lost in completely different ways.

 

Nothing below is measured, and it comes from reading scaling companies closely. The three kinds are a way of sorting, and what earns them a place is that each predicts something specific a reader can go and look for.

Three things a company can mean by keeping something

Human. A person decides, and no system decides for them. Any suitably qualified person will do, so this is a claim about the kind of decider rather than about which individual.

Fixed. The arrangement itself does not change, whoever is deciding: a policy, a boundary, an ownership that survives a reorganisation. Constancy is the claim, and who holds it is left open.

Founder-held. One named individual decides, and the decision does not travel. This is the narrowest of the three, and it is the only one that gets weaker as the company grows.

Most real commitments are one of the three cleanly. Where a company cannot pick, the useful move is to ask what would have to happen for the commitment to be broken, because the answer usually names the kind on its own.

Each kind disappears in its own direction

This is why the sorting earns its place. Commitments do not fail in general; they fail in a particular direction, and the direction follows from the kind.

A human commitment gives way to a system, an exception at a time. The first exception covers a segment that genuinely is different, and the argument for it is sound. The second arrives because the volume grew and a person cannot keep up. Nobody proposes removing the person, and after enough exceptions there is nothing left for a person to do.

A fixed commitment gets replaced, and the replacement is called an upgrade. Nobody proposes changing the arrangement. Somebody proposes a better tool, a cleaner structure, a migration, and the arrangement is carried away inside the change like furniture inside a house move. The word decision is never used, so nothing triggers a decision process.

A founder-held commitment goes quietest of the three, because a delegation nobody wrote down looks exactly like a busy founder. The individual stays involved in principle and is unavailable in practice, somebody sensible acts in the meantime, and the acting becomes the arrangement without anybody proposing it.

Three kinds. Three destinations. A company that knows which kind it is holding knows which of those three stories to check for.

The objection is that this is a taxonomy, and taxonomies are usually decoration

That objection is fair and it applies to most frameworks with three boxes in them. A sorting scheme earns nothing by existing. It earns its place only if putting an item in a box tells you something you would otherwise have missed, and plenty of them do not.

 

The test to apply here is a hard one, and this piece invites it. Take a commitment, tag it, and read off the failure route above. Then go and look for that specific route in your own company: the exception that became a batch job, the migration that carried an arrangement away, the acting-in-the-meantime that became permanent.

 

If the route the tag predicts is not the one you find, the tag was wrong or the scheme is. Either way you learn something in about twenty minutes, which is more than most taxonomies offer.

The uncomfortable case is a commitment that is two kinds at once

One honest limit. Some commitments are genuinely two things. A founder who personally approves an exception class is running a founder-held commitment on top of a human one, and both are true.

 

The scheme does not resolve that and it does not need to. A commitment carrying two tags simply has two failure routes open, and both want watching. The scheme fails where a company uses the tag to feel finished rather than to look, which is the risk with every instrument that produces a label.

What to do this week

Take three things your company holds constant and tag each one. Human, fixed, founder-held. Twenty minutes, and argue about the ambiguous one rather than resolving it quickly.

Then, for each tag, go looking for its specific route. Ask whether an exception has been granted lately for the human ones, whether a migration or a tooling change is scheduled anywhere near the fixed ones, and whether the founder-held one has quietly been acted on by somebody else in the last quarter.

The value is in the looking rather than in the tagging, and the tag exists to tell you where to point.

Of the things your company holds constant, which one is being carried away inside a change nobody is calling a decision?