Invoice approval: why OCR does not halve your lead time

Invoice approval is the administrative process with the most automation projects — and the highest rate of business cases that do not materialise. A worked model shows why.

Note: the figures below come from an example model, not from a client engagement. They are chosen to match typical mid-market magnitudes. Substitute your own — the pattern holds.

The baseline model

A company processes 1,200 incoming invoices a month, roughly 60 on a working day. The process has five steps:

Add up only the processing times and you get roughly 20 to 30 minutes per invoice. The actual lead time from arrival to payment release in such setups regularly runs to eight to twelve working days.

Between 25 minutes and ten days lie about three orders of magnitude. That gap is the real subject of the project — and it consists almost entirely of waiting.

Where the waiting is created

Distribute those ten days across the steps and a picture emerges that appears almost every time:

• Intake to capture: under a day. Capacity is sufficient.
• Capture: one to two days. Utilisation around 80 % — noticeable but manageable.
• Waiting for departmental sign-off: five to eight days. This is where the process is stuck.
• Coding and checking: under a day.
• Waiting for the next payment run: 1.5 days on average.

The reason for the approval delay is structural and has nothing to do with laziness: approval is not a full-time job. It is a side task for 14 people who each have a main job. Every one of them handles approvals when they get around to it — typically once a day, sometimes less often, and not at all while on holiday.

That gives this step two properties which together are fatal:

1. Very high variability. An invoice is approved in ten minutes or in six days. Variability is the main driver of waiting time — it enters the queueing formula squared.
2. No deputisation. If one person is out, their share waits until they return. There is no pool that takes over.

What OCR actually achieves

The usual proposal: introduce document capture, transfer invoice data into the ERP automatically, cut capture time from 10 minutes to 2.

The spreadsheet computes: 8 minutes × 1,200 invoices × 12 months = 1,920 hours a year. At €55/h that is €105,600. The project is approved.

What actually happens:

Capture does get faster — considerably. That is not a marketing promise, it materialises. Accounts payable utilisation falls from 80 % to under 30 %.

Lead time drops by about one day. From ten to nine. Because capture was never the constraint.

And then it gets worse before it gets better. Previously capture passed roughly 60 invoices a day to approval, evenly spread. Now they arrive in bursts — as soon as the scan is through, all at once. Approval, which used to be throttled by capture, now receives the full volume and the full variability. Its waiting time rises.

The constraint has not disappeared. It has migrated — and the time saved now sits as waiting time in front of the business departments.

The saving that is genuinely there

Do not misread this: the 1,920 hours are real. They are simply not what the business case claimed.

• What materialises: capacity in accounts payable. That is valuable — for growth without hiring, for taking early-payment discounts, for quality.
• What does not: the promised halving of lead time. And that is what the project gets judged on, because the departments complain about it.

The gap between the two is why a technically successful project ends as a disappointment.

What actually reduces lead time

Every effective measure acts on approval, and none of them is a software purchase:

1. Value thresholds. Invoices below a threshold do not need individual sign-off, only a sample check afterwards. In typical distributions, half the documents account for a small share of the value — but bind half of all approval events.

2. Automatic deputy rules. Not manual forwarding, but: after X days without a response the invoice goes to the deputy. That cuts off the long tail of the distribution — and the tail is what ruins the average.

3. Batched approval at fixed times. Instead of “when I get around to it”, a fixed daily slot. This reduces variability drastically without anybody working more.

4. Sign-off before capture. The classic sequencing error: careful capture first, and only then does it turn out the invoice is disputed. A short substantive check at the front of the process saves capturing every disputed case.

Measure 3 is, in our experience, the most effective relative to effort — and the one that appears in no proposal, because it cannot be sold.

The sequence that works

1. Measure where the time sits first. Count inventory in front of each step, divide by daily throughput. That costs one morning and answers the core question.
2. Then relieve approval. Thresholds, deputies, fixed slots. No budget required.
3. Then measure again. The constraint now sits elsewhere — probably in capture.
4. Then automate. Now automation hits the real constraint, and the saving appears in lead time too.

Follow that order and you buy the same software — but get a project that delivers what the business case promised.

Why to play this through before buying

The effect described — constraint migrates, saving evaporates — is invisible in a spreadsheet, because a spreadsheet computes each step in isolation. In a simulation model you see it in seconds: reduce capture time, and the queue in front of approval grows in front of your eyes.

That is what FlowVisual is built for. The process above is modelled in about half an hour; after that, the question “what happens if we introduce OCR?” costs one click — and the answer comes with a P10–P90 band instead of a single number nobody can defend.