Approval Workflow Automation: How to Shorten Decision Cycles

Approval workflow automation is usually the step missing from companies that have already digitised everything else. The documents are PDFs, the data sits in the ERP, requests arrive through a form. Then everything stops, waiting for someone to press a button.

That is not a technology problem. It is a queueing problem.

Approval time is not working time

Ardent Partners has benchmarked finance teams for 19 years. In AP Metrics that Matter in 2025, the average accounts payable organisation takes 9.2 days to process a single invoice.

The spread is the interesting part. Best-in-Class organisations — the top quintile by lowest processing cost and shortest cycle time — close the process in 3.1 days. Everyone else takes 17.4 days. An 82% difference.

It is worth being precise about what those numbers measure. Processing an invoice, in actual work, takes minutes: open it, match it against the purchase order, post it. The 9.2 days measure no work at all. They measure waiting.

So when 41% of the teams surveyed name "invoice and payment approvals take too long" as a top obstacle, they are not saying people work slowly. They are saying the process spends most of its life parked in an inbox.

Why 60 days is not a lot of room

For a European SME, that delay carries a direct legal and financial consequence.

Directive 2011/7/EU on late payment caps business-to-business payment terms at 60 days, unless a longer period is expressly agreed and is not grossly unfair to the creditor. Public authorities must pay within 30 days. Once the deadline passes, the creditor is automatically entitled to late payment interest — at least 8 percentage points above the European Central Bank reference rate — plus a minimum of 40 euros per invoice in recovery costs.

Now apply the Ardent Partners figures. A 17.4-day approval cycle burns 29% of a 60-day window before anyone even decides when to pay. On the receiving side, that is the same delay hitting your cash position.

The approval cycle is not administrative detail. It is working capital.

Where cycles actually stall

Four causes explain almost all of the delay, and none of them is a lack of goodwill.

One approver for everything. In an SME, the managing director often approves a 40-euro expense and a 40,000-euro contract with the same level of attention. When that person is on holiday or travelling, the entire process stops. This is one of the signs a company needs digital transformation that costs the most and gets measured the least.

No context in the request. The approver receives "can you approve this invoice?" without seeing the linked purchase order, the budget already consumed or the supplier history. So they ask. That question opens a second cycle, over email, that nobody counts.

The wrong channel. A request sent by email competes with 200 other emails. A request that lands in Slack or Teams, with two buttons, competes with nothing.

No deadline, no escalation. If there is no defined limit, there is no such thing as late — there is only a request nobody has looked at yet. Which is why many cycles only move when someone remembers to chase.

A worked example

What follows is our own estimate, with the arithmetic shown, not a figure from a study.

A services SME approves 120 supplier invoices a month. Each one passes two approvers: the department head and finance. If each approver takes two working days on average to respond — a conservative assumption, well inside the 9.2-day average Ardent Partners reports for the full process — the invoice spends roughly four working days waiting purely on people.

The real work on that invoice is perhaps 6 minutes. Four working days are 1,920 minutes of business calendar. More than 99% of elapsed time is queue, not work.

That is the angle approval workflow automation attacks. It does not make the decision faster; it makes the gap between decisions close to zero. The request reaches the right channel, with the full context, the moment the document arrives. If nobody responds within 24 hours, it is resent. Within 48, it escalates.

If you want to turn those days into euros before deciding anything, the method is set out in our article on how much manual processes actually cost.

What approval automation does not fix

This is worth stating plainly, because it is where most projects fail.

Automating an approval chain does not improve a bad decision. If the approval criteria are not written down anywhere, automation simply distributes the ambiguity faster.

It does not eliminate exceptions either. Ardent Partners found that exceptions — coding errors, missing information, absent purchase order data — are now the single biggest challenge in accounts payable, cited by 53% of respondents. The average exception rate is 14%, and even the Best-in-Class quintile sits at 9%. Roughly one invoice in ten will still need a human to investigate. The goal is that only that one reaches a person.

And it does not replace spending authority. A flow that approves on its own above any sensible threshold is not efficiency, it is fraud exposure — already named as a serious threat by 29% of teams.

n8n and Power Automate: the difference shows up in month three

Both platforms handle approvals. The practical difference only appears when a request sits pending for a long time.

Power Automate has a documented limit of 30 days per flow run, and the documentation is explicit: the limit includes flows with pending steps such as approvals, and after 30 days any pending step times out. For long holidays, extended sick leave or a contract waiting on legal review, that forces you to split the flow in two and hold the state in Dataverse — which is a prerequisite for using approvals at all, even on an Office 365 licence. In return, you get approval patterns out of the box: everyone must approve, first to respond, custom responses, and sequential approval.

In n8n, waiting works the other way round. When the workflow pauses, the execution data is offloaded to the database and only reloaded once the resume condition is met — waits under 65 seconds never touch the disk at all. There is no documented maximum. The approval request can go out through nine different channels, including Slack, Teams, Telegram, Gmail, Outlook, Google Chat and WhatsApp Business Cloud.

There is also the question of where the data runs. n8n can be self-hosted, which matters when an approval request carries amounts, margins or client data — a point covered in our article on what n8n is.

The four rules that shorten cycles most

From designing these chains, four decisions matter more than the choice of tool.

Set thresholds by value. Below a limit — 250 euros, 500, whatever fits — approve automatically and notify. If 60% of requests are low value, you have just removed 60% of the volume from the queue.

Approve by exception. Invert the logic: the request proceeds unless someone objects within 24 hours, except above the threshold. This works well for recurring spend and badly for new purchases.

Put the context inside the request. Amount, supplier, linked order, remaining budget, history. If the approver has to open another system to decide, the cycle has already failed.

Escalate by default, always. With no deadline and no automatic deputy, the chain depends on human memory. With them, it depends on itself.

Where to start

Do not start with the most critical process. Start with the one where you can measure the "before".

Pick a chain with volume — supplier invoices, holiday requests, expense claims — and measure two things over two weeks: how many days pass between request and decision, and how many of those requests someone would have approved without looking.

Then answer one concrete question, before choosing any tool: how many of your approval requests were actually rejected last month?

If the answer is close to zero, you do not have a decision problem. You have a queueing problem — and that one is solvable.