Perspective / 05

An approval step is not a decision model

Adding an approval is easy. Defining the decision behind it, its owner, its evidence and what “no” means is the actual design work.

Perspective

Independently created. Contains no employer or client implementation detail, internal names or figures.

The thesis

Organizations often add an approval step to feel safer. Without a defined decision, one accountable owner, the right evidence and a designed answer to “no”, the step adds delay and signatures, and very little control.

Why approvals multiply

Approval steps tend to appear after something goes wrong. A discount was too generous, a customer turned out to be a credit risk, a contract carried a clause nobody had read. The response is understandable: from now on, someone has to approve it. The step is cheap to add, it signals that the problem has been taken seriously, and it gives everyone involved a name to point to.

Approvals are much harder to remove. Nobody owns the question of whether an approval still protects anything, and removing one feels like accepting the original risk again. So they accumulate. Over time a process can carry several approvals that each made sense once, together add days, and protect less than anyone assumes.

A workflow step is not a decision

A workflow step routes a record to a person and waits for a button. A decision is a judgement with a question, an owner, the evidence needed to answer it and the criteria for answering it well. The two are easily confused, because the step is where the decision is supposed to happen.

The trouble starts when one step carries several decisions. “Approve customer” can mean whether to commit commercially on these terms, whether this legal entity can transact, and whether the credit exposure is acceptable. Each has a different natural owner and needs different evidence. Merged into one step, each approver signs for things they cannot judge. The sales manager approves tax data. The finance reviewer approves terms they do not understand. Everyone has approved and nobody has decided.

That is what fake accountability looks like. The signatures are real, but the step cannot say who was responsible for which part of the outcome. When something goes wrong, the investigation finds approvals everywhere and ownership nowhere, and the usual response is to add another approval.

One decision, one owner, the right evidence

The first design move is to name the decision behind each approval in one sentence. If the sentence needs an “and”, there are probably two decisions. Each decision then gets one accountable owner, even when several people are consulted, and the approver sees the evidence that decision needs and not much more. A credit decision needs exposure, payment history and the legal entity. It does not need the product configuration.

Criteria matter as much as evidence. An approver who does not know what a good answer looks like will default to whatever is least risky for them personally, which is usually delay. Clear criteria also show which decisions do not need a person at all: when the rule is explicit and the consequence is small, the decision can be made by the rule, and the approver’s attention can go where judgement is actually required.

Design the “no” before the “yes”

Most approval designs describe the path to yes in detail and leave no undefined. In practice, “no” has several meanings. Not on these terms. Not yet, because evidence is missing. Not for this legal entity. Yes, with conditions. Each one should send the work somewhere different.

A rejection should carry a reason, return the work to the stage that can fix it, and leave the record in a state that is true. When rejection has no designed path, it moves into email and side conversations, the record stays “pending”, and the process stops being observable exactly where it most needs to be.

Time is part of the decision

Every approval needs an expectation of time: when the clock starts, when it stops, and who owns it while it runs. It also needs an explicit rule for silence. If nobody responds, does the request escalate, expire, or proceed? Approving on timeout is a policy choice. So is escalating to the next level. Both are legitimate. Neither should happen by accident.

Escalation deserves the same precision. An escalation that only sends a reminder to the approver’s manager is a nudge, not a decision path. A useful escalation moves the decision to someone with the authority and the information to make it, carries the evidence with it, and records that the decision moved.

Measuring approvals this way changes the conversation. Ageing per decision, first-pass rates and the reasons for rejection show which approvals protect something and which only add waiting. That evidence is what makes it possible to remove a step without feeling reckless, and to strengthen the ones that matter.