Approval workflows people actually follow

Every business has approval processes. Most of them are bypassed, worked around, or done retrospectively — and here is the uncomfortable part: the bypassing is usually rational.

When a process is too slow, too rigid or too disconnected from how work actually happens, people route around it to get their jobs done. They are not being reckless; they are being practical. And an approval process that everyone dodges is worse than no process, because it gives you the illusion of control — a policy on paper, a documented procedure — while the real decisions happen in the gaps, unrecorded.

So the goal is not more approvals or stricter ones. It is approvals people actually follow, which is a design problem, not a discipline problem.

Why approvals get bypassed

Understand the failure before designing the fix. Approvals get dodged for specific, fixable reasons:

They are too slow. The approver is busy, travelling, or does not see the request for days. The requester has a job to do now, so they proceed and sort the paperwork later — which means the approval becomes a formality after the fact, not a control before it. Speed is the number one killer of approval compliance.

They are disconnected from the work. The request lives in one place, the approval happens in another — an email, a WhatsApp, a signature on a printout — and the record in a third. The seams make it slow and easy to skip, and skipping leaves no trace.

They ask the wrong person, or too many people. Approvals routed to someone who does not have the context, or through three people when one would do, create delay and diffuse responsibility. When everyone approves, no one really does.

They are all-or-nothing. The same heavy approval applies to a RM50 purchase and a RM50,000 one. So people reasonably dodge it for the small stuff, and the habit of dodging carries over to the large stuff. Uniform friction trains people to route around you.

They give the approver no context. The approver gets "approve this?" with no information, so they either rubber-stamp (no real control) or chase for details (delay). Neither is the control you wanted.

The principles of an approval people follow

Invert each failure.

Make it fast

The single most important thing. An approval that happens in minutes gets followed; one that takes days gets bypassed. That means the approver is notified immediately, can approve from wherever they are — their phone, in a queue at lunch — and the request does not sit waiting to be noticed. If approving is faster than working around it, people approve. Get this one right and most compliance problems dissolve.

Match the friction to the stakes

Not everything needs the same approval. Small, routine, low-risk things should be light or automatic. Large, unusual or risky things deserve real scrutiny. Tiering the approval to the consequence means people do not resent friction on trivial things — which is what teaches them to dodge friction on important ones. This is the same principle as procurement control: proportionate, not uniform.

Give the approver what they need to decide

The approver should see, in the request, everything they need to say yes or no in seconds — the amount, the reason, the supporting document, the context. No chasing, no opening three other systems. A decision that takes ten seconds with full context gets made now; one that requires investigation waits, and waiting is where bypass begins.

Keep it in one place

The request, the approval and the record should be the same thing in the same system — not a request here, an approval in email, and a record in a spreadsheet. When they are one connected flow, there is nothing to skip and nothing to reconcile. Every seam is a place the process leaks, exactly as in order-to-cash.

Make the record automatic

The approval should leave a trail without anyone doing extra work — who approved, when, on what basis. If recording the approval is a separate chore, it gets skipped. If it is an automatic consequence of approving, you get a clean audit trail for free, which is exactly what you will want when someone asks why a decision was made.

Where AI helps

Routing intelligently. Sending each request to the right approver based on its type, amount and context — and to a backup when the primary is unavailable, so nothing sits because one person is on leave.

Chasing. If an approval is pending too long, a reliable nudge — the same follow-up discipline that works for collections. Approvals stall because nobody chases; automate the chase.

Summarising for the approver. Presenting the request with the context distilled — the relevant history, the supporting document read and summarised — so the decision is genuinely ten seconds, not a research task.

Flagging the unusual. "This request is much larger than normal for this person" or "this supplier is new" — giving the approver a reason to look harder exactly when it matters, and to wave through the routine with confidence.

Where judgement stays human

The approval decision itself. AI can route, chase, summarise and flag — it should not make the call on anything that matters. The whole point of an approval is a human taking responsibility for a decision, and automating the decision defeats the purpose while removing the accountability, as we argue in what AI still cannot do in ERP. Automate everything around the decision; leave the decision with a person.

The test

There is one honest test of whether your approval process works: do people follow it when no one is watching?

If they do, you have designed it well — it is fast enough, light enough and connected enough that following it is easier than dodging it. If they do not, no amount of policy fixes it. You do not have a discipline problem to solve with memos and enforcement; you have a design problem to solve by making the right thing the easy thing.

Build approvals that are faster to follow than to bypass, and compliance stops being something you police. It becomes the path of least resistance — which is the only kind of process that actually holds.

A workflow that people actually follow proves the process runs. It doesn't, on its own, prove anyone could explain why it was built this way — see the audit trail tells you who, not why.

Common questions

Why do people ignore our approval process?

Usually because bypassing it is rational. Approvals get dodged when they are too slow to fit the work, when the request lives in one place and the approval in another, when they route to someone without the context or through three people when one would do, when the same heavy process applies to a RM50 purchase and a RM50,000 one, and when the approver is given nothing to decide on. Each of those is a design fault, not a discipline fault.

How quickly does an approval need to happen?

Fast enough that approving beats working around it, which means minutes rather than days. Speed is the number one driver of approval compliance: an approval that happens quickly gets followed, one that takes days becomes paperwork filed after the fact. In practice that means the approver is notified immediately, can respond from wherever they are, and the request does not sit waiting to be noticed.

Can AI just approve things for me?

It should not make the call on anything that matters. The whole point of an approval is a human taking responsibility for a decision, and automating the decision defeats the purpose while removing the accountability. What AI should do is everything around it: route each request to the right approver and to a backup when they are away, chase what is pending, summarise the context so the decision takes seconds, and flag the unusual.

How do I tell whether my approval process actually works?

There is one honest test — do people follow it when no one is watching? If they do, it is fast enough, light enough and connected enough that following it is easier than dodging it. If they do not, no amount of policy fixes it, because you have a design problem rather than a discipline problem. An approval process everyone dodges is worse than none, since it gives you the illusion of control.

This is the kind of work SmartB Studio is built for. Get in touch and we will go through it against your actual processes rather than a generic demo.

Related: AI in procurement for businesses and the new shadow IT.

Also worth reading: moving approvals off WhatsApp.