Why Your Approvals Take Three Days

Somebody needs a discount signed off, a purchase authorised, or a variation approved. The decision itself takes about ninety seconds of a manager’s attention.

It takes three days, and in most businesses nobody has ever asked why.

Where the three days go

Almost none of it is thinking. The request is sent by email to somebody who is on site. It sits unread. A reminder goes out the next morning. The manager opens it, does not have the information they need to decide, and asks a question. The answer comes back four hours later. Then it is approved.

Two minutes of decision wrapped in three days of waiting, and every stage of that wait is somebody unavailable rather than somebody deliberating.

The three causes

The request is incomplete. It does not carry what the approver needs, so the first response is always a question. That single round trip usually costs a day on its own.

It arrives somewhere the approver is not. Email works for people at a desk. A manager who is on site, in meetings or between jobs reads email in the evening if at all.

Nothing chases. The request sits until the person who raised it gets frustrated enough to follow up, which means the delay is bounded by somebody else’s patience rather than by any process.

Set thresholds and stop approving everything

The first question is not how to route approvals faster. It is how many of them need to exist.

Most approval processes were built after something went wrong, and they now apply to everything regardless of size. A manager signing off a small purchase is not managing risk, they are adding a queue.

Agree a threshold below which nobody approves anything. It is uncomfortable for about a month and it removes most of the volume, which makes the remaining approvals faster because they are no longer buried.

Where AI removes the waiting

Two jobs, and neither is the decision itself.

Assembling the request. AI gathers what the approver needs before they see it. The history with this client, the budget position, whether similar requests were approved and on what terms, the relevant document. The manager opens something they can decide on immediately rather than something they have to investigate.

Routing and chasing. It goes to the right person based on what it is and how large, reaches them where they actually are, escalates if nothing happens, and never depends on the requester remembering to nudge.

Stanford’s AI Index for 2024 reviewed the research on assistance of this kind and found it both speeds up task completion and improves quality, with the biggest gains among people less experienced at the task. For approvals that matters, because it means a deputy can handle what previously waited for one senior person.

What to measure

• Elapsed time from request to decision, not the time the approver spent.

• Requests needing a clarifying question. Every one is a day. Should trend to near zero.

• Approvals per manager per week. If it is high, the threshold is too low.

• Requests declined. If almost nothing is ever refused, the approval is a formality and should be removed.

Delegation beats speed

Before optimising the route, look at who is on it.

In most businesses approvals cluster on one or two senior people, not because the decision requires their judgement, but because nobody has ever written down who else could make it. So the bottleneck is a person rather than a process, and speeding up the routing simply delivers the queue faster.

Write down the decisions that genuinely need the founder or the managing director, and honestly assess the rest. Most companies find the list is short, and the moment a deputy is named for everything else, the elapsed time falls without any technology involved.

Approvals that block revenue deserve their own path

Not all delay is equal. An approval sitting between a client saying yes and work beginning is a different animal from one on an internal purchase.

Separate them. Anything on the revenue path gets a shorter escalation window and reaches a person on a channel they actually watch. Anything internal can wait, and treating both the same is why the important ones sit behind stationery orders.

Businesses that measure days from client agreement to work starting are almost always surprised by how much of it is internal approval rather than client delay.

The audit trail you will eventually need

Approvals by email or message have a property that only becomes obvious later: they are unfindable.

Six months on, when a client disputes a variation or an auditor asks who authorised something, the evidence sits in an individual inbox, possibly belonging to somebody who has left.

A logged approval records what was asked, what information was attached, who decided, when, and on what basis. That costs nothing when the system is built and cannot be reconstructed afterwards at any price. In regulated work it is the difference between an inconvenient conversation and a serious one.

The pattern to watch for

If approvals are consistently granted without question, the approval is not a control. It is a delay with a signature on it.

Either the threshold is set too low, or the real control happens earlier and this stage is ceremonial. Both are worth fixing, and both are visible the moment you look at how many requests were ever actually refused.

Start by counting

Before changing anything, take last month and count. How many approvals were raised, by whom, to whom, how long each took, and how many were refused.

Most businesses have never assembled that view and it usually settles the argument on its own. The volume is higher than anyone expected, it concentrates on one or two people, and the refusal rate is close to zero.

Sources

• Stanford Institute for Human-Centered Artificial Intelligence, Artificial Intelligence Index Report 2024, 15 April 2024.

AI Optimize builds the routing, the assembly and the escalation so decisions wait on judgement rather than on availability. That work sits under Workflow Automation.