Why Work Waits: The Real Cost of Manual Approval Chains

The expensive part of an approval is not the decision. It is the waiting. A routine purchase request that needs two signatures takes a few minutes of actual reading and several days of sitting in inboxes, and most companies have no idea which of those two they are paying for.

Here is the short version of this whole post: approval delay is a queue problem, not a judgment problem. The fix starts with measuring waiting time separately from decision time, and it usually ends with removing approval steps rather than speeding them up. Automation comes last, and only for the steps that survive.

This is written for the people who live inside these chains: operations leads, finance managers, and owners of businesses in the 50 to 300 employee range, where approval chains grew one incident at a time and nobody ever went back to prune them.

More than most leaders would guess, and the research here is consistent. Asana's Anatomy of Work Index, built on surveys of more than 10,000 knowledge workers, found that around 60% of the working day goes to what it calls work about work: chasing status updates, searching for documents, and talking about tasks instead of doing them. Approvals sit squarely inside that number, because every stalled request generates its own trail of reminder messages and status meetings.

For a 10-person operations and finance team, that 60% is the equivalent of six people's hours spent on coordination rather than output. Not because anyone is slow, but because the process forces everyone to spend their day asking where things stand.

A separate survey by Workfront in 2021 reported that 60% of employees say their work is often delayed by slow approvals. That figure comes from a vendor survey, so treat it as directional rather than precise. The direction, though, matches what the Asana data and most people's own inboxes already say.

A worthwhile question before reading on: how many of your approval chains have a clock on them? If nobody can tell you the average cycle time for a purchase approval in your business, the rest of this post is for you.

Time a real approval end to end and a pattern shows up fast. The judgment, meaning the minutes an approver spends reading the request and deciding, is a tiny slice of the total. The rest disappears into three specific gaps.

Routing is the first gap: the request sits with the wrong person, or with the right person's overloaded inbox, or waits for someone on leave with no fallback approver defined. Context is the second: the approver opens the request, finds a missing quotation or an unclear cost code, and sends it back a full day later for a fix that takes two minutes. Re-entry is the third and quietest: the approval happens in email or chat, and then someone types the approved result into the actual system afterward, which is where transcription errors and lost records come from.

We hit all three of these while designing the approval chains inside our own HRMS product. The design question that turned out to matter most was never who approves. It was what happens while the request waits: who gets notified, what the fallback is after a defined number of hours, and what context travels with the request so the approver never has to send it back for basics.

None of those three gaps is a judgment problem. All three are process problems, which is exactly why throwing a faster tool at them without redesign does so little.

Because automation multiplies whatever it is given. Automate a five-step chain in which two steps add no real control, and you now have two useless steps running faster, with a tool subscription on top. The chain is quicker and still wrong.

The strongest recent evidence for this comes from McKinsey's State of AI 2025 research. Across 25 organizational attributes tested, redesigning workflows had the biggest effect on whether organizations saw measurable bottom-line impact from new tooling. Yet only 21% of organizations had redesigned at least some workflows before deploying. Nearly 80% layered the new tool on top of the existing process and hoped. The finding is about AI tooling specifically, but the mechanism, tooling bolted onto an unexamined process, is exactly what happens with approval software too.

This is the same reason we keep arguing that custom software wins when the workflow is complex: the value is never the software alone, it is the forced clarity about how the process should actually run. Teams that have outgrown their packaged systems usually discover the packaged tool was faithfully automating a process nobody had questioned in years.

Do not buy the tool first. A workflow tool chosen before the workflow is mapped will be configured to mirror the current chain, including its dead steps, because that is the path of least resistance during setup.

Do not add approvers for cover. Every approver added after an incident feels like control and arrives as delay. If a step exists so someone is informed rather than deciding, make it a notification, not an approval. Notifications do not block the queue.

And an honest limit: some approvals should stay slow and human. Exception handling, judgment calls on unusual spend, anything with real risk attached deserves a person's full attention. The target of this post is the routing and waiting around every approval, not the judgment inside the few that need it.

1. Pick one approval chain that runs often. Purchase requests and leave requests are good candidates because they cycle weekly.

2. Time five real instances end to end, and record two numbers for each: total elapsed time, and minutes of actual decision work. The gap between those numbers is your waiting cost, in writing, for the first time.

3. List every step and ask what each one prevents. A step that has never rejected anything in six months is a notification pretending to be a control. Convert it or cut it.

4. Define fallbacks before tooling: who approves when the named approver is out, and how many hours a request may wait before it escalates on its own.

5. Only then look at automation, and point it at the routing, the reminders, and the record keeping. The steps that survived your pruning are now worth speeding up.