A 20-minute audit of your approval process

To find where your approvals slow down, take your last ten requests and do four things: count the steps, time the waits, find the approvers who never reject anything, and check whether the threshold still makes sense. It takes about twenty minutes, and it regularly changes what you build — or whether you need to build anything at all.

Most businesses skip this and go straight to automating the approval chain as it exists. That chain usually has steps nobody would design today, and automating it makes them permanent.

What you need

Pick one approval process that people complain about. Purchase requests, variations, leave applications, expense claims and new supplier set-ups are all good candidates. Then pull the last ten examples, with whatever history you have: email threads, the timestamps in your finance system, the approval column in a SharePoint list.

Ten is enough to see a pattern without turning it into a project. If you cannot find ten examples from the last few months, the process may not run often enough to be worth automating; see when not to automate.

A simple worksheet keeps it honest: one row per approval step, filled in from the ten examples.

The rows above are an illustrative example of what the sheet tends to look like, not results from any particular business.

Step 1: Count the steps

Take your ten examples. For each, count how many people touched it — not how many the policy says, how many actually did.

Most organisations discover the number is higher than the policy says, because informal check-ins have accumulated. Someone forwards the request to a colleague "just to check". A project manager copies in the commercial lead. Finance asks the requester a question by email before approving. None of these appear in the policy, but all of them add a wait.

Write down every touch, including the informal ones. The informal ones are often where the time goes.

Step 2: Time the waits

For each step, note how long the item sat before it moved. Not how long the decision took — how long it waited.

The waiting time is almost always more than 90% of the total. That is the number worth attacking.

When you lay it out like this, the improvements become obvious. The decisions take minutes. The waiting takes days. Anything that shortens the waits — fewer steps, parallel rather than sequential approvals, escalation when something sits too long — matters far more than making each decision faster.

Step 3: Find the rubber stamps

For each approval step, ask: when did this person last reject one?

If the answer is "never" or "I cannot recall", that step is not a control. It is a delay wearing the costume of a control. Either delete it, or convert it to a notification.

This single question removes more delay than any automation.

Ask it gently, and ask the approver directly. Most rubber-stamp steps exist for a good reason that has since expired — a bad purchase years ago, a manager who has since left, a new starter who needed checking. The person doing the approving is often relieved to be asked.

Notice that the "after" chain has fewer waits but loses no real control. The department head still sees everything. The finance check still happens — it just happens on the form, before the request goes anywhere.

Step 4: Find the real threshold

Ask what value would genuinely warrant a second approver. Compare that to your current threshold, which was frequently set a decade ago and never adjusted for inflation.

A good way to ask it: "If this request came in at this value with no second approver, would anyone be uncomfortable?" If the honest answer for a threshold of a few thousand dollars is no, the threshold is too low. Agree a new one with whoever owns the delegation of authority, write it down, and record the date so it gets reviewed again.

Sequential or parallel?

While you have the chain in front of you, check whether each step genuinely needs the one before it. A finance coding check does not depend on the department head's view, so there is no reason one should wait for the other. Running independent checks side by side, rather than one after another, cuts the total wait to the slowest single step instead of the sum of all of them.

The same goes for information requests. If finance regularly sends requests back for a missing quote or cost code, make those fields mandatory on the form. Every send-back you prevent removes a full round of waiting.

What you will likely find

• Two or three steps that have never rejected anything
• One step where everything queues, usually one person
• A threshold that is far lower than anyone would now defend

Fix those three things and you may not need automation at all. If you still do, you are now automating a process worth keeping.

The single-person queue deserves its own fix. It is usually a senior person who approves in batches when they get a spare hour. Automation helps here — a Teams approval on their phone, with a reminder after a day and a named delegate after two — but so does simply giving them a deputy.

If you then automate

Once the chain is audited, a Power Automate approval is a modest build. The design choices that matter are the ones the audit just answered:

If any of those cannot be answered in one sentence, write the process on a page before anyone builds; write the process before you automate it has the template.

What to do next

Book twenty minutes this week with whoever runs the process and the ten most recent examples. Fill in the worksheet, ask each approver the rejection question, and agree what to remove. Then live with the shorter chain for a fortnight before deciding whether it still needs automating.

If you would rather someone else ran the exercise across your main approval processes, that is part of our free diagnostic audit.

Automating an approval chain nobody has questioned in five years just makes the wrong thing happen faster.