Most ROI cases for invoice automation are built to reach a conclusion. They compare the full cost of manual entry against the subscription price and stop there, which produces a number so favourable it stops being believable.
This is the version with the awkward parts left in. It's a model rather than a claim — the numbers below are illustrative, and the point is to run it on yours.
The three costs people leave out
Before any arithmetic, the three things that make honest cases look different from vendor ones.
Residual work. Automation doesn't take invoice processing to zero. You still review flagged fields, still handle documents that read badly, still make coding decisions on new suppliers, still approve things. The saving is on the routine majority, not the whole job. A model comparing your current cost against zero is wrong before it starts.
Setup and learning. Connecting the ledger, setting coding defaults, telling suppliers a new address, and a first month where everything is checked twice because nobody trusts it yet. Real, one-off, and usually a few hours rather than a project.
Saved time isn't automatically saved money. This is the big one. Freeing eight hours a month only becomes cash if those hours stop you hiring, let you take on more work, or replace something you're paying for. If they're absorbed into the general pool of things there's never enough time for, the benefit is real but won't appear in your accounts.
Be clear which of those you're claiming, because they have very different strengths as arguments.
The manual side
Start with time per document, end to end — finding it, working out what it is, typing, coding, checking, filing. Not just the typing, which is under a fifth of it.
For a mixed set of documents, three to five minutes each is a fair working range. Time a batch rather than borrowing that figure; it varies more by document mix than by person.
At four minutes each:
| Invoices per week | Hours per month | At £18/hr | At £25/hr |
|---|---|---|---|
| 25 | 7.2 | £130 | £180 |
| 50 | 14.4 | £260 | £360 |
| 100 | 28.9 | £520 | £722 |
| 200 | 57.7 | £1,039 | £1,443 |
Based on 4.33 weeks a month. The hourly figures are fully-loaded cost — salary plus employer's NI and pension, which is roughly salary ÷ 1,500 for an annual figure.
Two things this table doesn't capture, both of which push the real cost up:
Error correction. Miscoded invoices found at reconciliation, duplicate payments, VAT adjustments. Infrequent and expensive when they happen, which makes them hard to average. If you've had a duplicate payment in the last two years, put its value in.
The interruption cost. Entry done in fragments across the week is slower per document than the same volume in one sitting, and it occupies attention in a way the clock doesn't show.
The automation side
Four components:
Subscription. For small-business invoice capture, roughly £15–£40 a month depending on volume and seats. Check both dimensions — per-document pricing looks cheap at demo volumes and less so at 500 a month, and per-seat pricing looks cheap until approvers need logins.
Residual time. The honest part. Assume you still spend 30–40% of the original time in the first month, settling to 15–25% once supplier coding is learned and you've stopped double-checking everything. Anyone quoting you 5% is describing a business with very uniform documents.
Setup. Call it four to six hours, one-off.
Exception handling. The awkward documents don't get less awkward. They just stop being diluted by the easy ones, so the average time per remaining document goes up even as the total falls.
Running it
At 50 invoices a week, 4 minutes each, £20/hr fully loaded:
| Manual | Automated | |
|---|---|---|
| Time per month | 14.4 hrs | ~3.2 hrs (22% residual) |
| Labour cost | £288 | £64 |
| Subscription | — | £30 |
| Monthly total | £288 | £94 |
Monthly saving £194, or about 11 hours. Against a one-off setup of five hours, the time pays back in under a month and the cash cost is positive from month one.
At 15 invoices a week the same model gives a monthly saving nearer £55 — still positive, but small enough that it turns on whether the freed time has somewhere useful to go.
The shape is the point: this scales with volume, and below roughly 20 invoices a week the case gets thin. Any vendor whose case looks equally strong at every volume is not modelling residual work.
Where the case is strongest
Not where the volume is highest. Where the time is most constrained.
A bookkeeping practice. Freed hours convert directly into client capacity, which is revenue. The clearest version of the argument.
A business at a hiring decision. If the alternative is part-time admin help, the comparison is against a salary rather than against notional time.
Anywhere month-end is late. The cost of a slow close isn't in the table above. It's decisions made on stale numbers, and it's the reason finance directors buy this when bookkeepers don't.
High document variety. Counter-intuitively, businesses with many suppliers and formats benefit more, because that's where manual entry is slowest and supplier defaults help least.
Where it isn't
Very low volume. Under 20 a week, run the numbers before assuming.
Few suppliers, identical formats. Already fast, and supplier defaults in your accounting package capture most of the remaining benefit for nothing.
Mostly non-standard documents. Handwritten dockets and unusual paperwork produce exceptions, and exceptions are the expensive part.
No slack in the process. If entry happens at 7pm because there's no other time, the saving is real but won't show up as money.
A sanity check
Two questions worth answering before committing:
What's the payback period? Setup hours divided by monthly hours saved. Under two months is a straightforward decision. Over six, the assumptions need to be right, and they probably aren't.
What happens if the residual is double your estimate? Run it at 45% residual instead of 22%. If the case still holds, it's robust. If it inverts, you're betting on extraction quality you haven't yet verified on your own documents — which is worth testing before paying for.
The number that matters more than any of this
Take twenty of your real documents, including the worst ones, and run them through whatever you're evaluating. Then count how many you could post without opening the original.
That single figure sets the residual, and the residual is what the whole model turns on. Everything above is arithmetic; that number is evidence.
Cribble reads invoices from your inbox and posts them into Xero, or exports CSV for QuickBooks, Sage50 and DATEV. Every field is scored separately so the checking is a short list rather than the whole document. If you want the residual figure for your own paperwork, send us your three most awkward documents.