Quintessentia Network logo QuintessentiaNetwork Inc.
ACCOUNTS PAYABLE

Accounts payable automation: buy the software, or build the invoice pipeline?

Short answer: buy AP software if your bookkeeping platform already captures receipts or you process under roughly 500 invoices a month; build an invoice pipeline when volume is in the thousands, formats are mixed, there is no usable API into your ledger, or you need a residency requirement named in a contract.

Invoices on a glass track splitting toward a sealed box and a set of modular blocks, each route passing a closed gate
Illustration: invoices go one of two ways, off-the-shelf AP software or a custom pipeline, and payment still waits behind a human approval.

Search for accounts payable automation and you will mostly find AP software vendors. That is a fair place to start, and for a lot of firms it is also where the search should end. This page is the other half of the question: when the software is enough, and when it pays to build the invoice pipeline yourself.

Everything below is in Canadian dollars, and every figure comes from work we publish elsewhere on this site. They are our estimates, not your invoice.

What accounts payable automation actually automates

Five stages, and only one of them involves a model. Ingestion, so invoices arrive from email, a scanner or a supplier portal without anyone re-keying them. Extraction, which reads the fields. Business-rule validation, written in deterministic code rather than left to the model: totals reconcile, tax is checked, the PO exists, the supplier is known, the date is in range, duplicates are caught. Then a write into your system of record, and an exception queue for everything the rules would not pass.

That validation layer is where correctness actually lives. A model that reads an invoice perfectly and posts it against a PO that does not exist has not helped you. The same five stages, in detail, are on the intelligent document processing page.

The arithmetic, in Canadian dollars

Take 5,000 invoices a month. At four minutes each of full manual handling — opening, identifying, keying, filing — that is 333 hours a month, about CAD $12,000–13,300 in loaded clerk time, or most of two full-time people. An Ontario document clerk at a $52,000–58,000 base costs roughly $65,000–72,500 a year fully loaded once you add CPP, EI, Employer Health Tax, WSIB, benefits and paid vacation; over about 1,800 working hours that is $36–40 an hour.

Automated, the same workload runs at roughly $1,000–1,700 a month all-in. Model API is $100–400 of that and infrastructure $200–600. The largest line is not the model. It is human review, and its size is a function of accuracy:

99% accuracy leaves about 50 exceptions a month, 3.3 hours, roughly $130. 95% leaves 250 exceptions, 16.7 hours, about $650. 90% leaves 500, 33.3 hours, about $1,300. 80% leaves 1,000 exceptions, 66.7 hours, about $2,600. Moving from 80% to 95% saves roughly $1,950 a month — about five times the entire model API bill.

The full derivation, including what the infrastructure line is made of, is in what AI automation really costs.

When AP software is the better answer

Two cases, and they are common.

Your bookkeeping platform may already do it. Several practice-management and bookkeeping platforms now ship receipt capture and bank-feed categorisation that is genuinely good. If you are paying for one and not using that module, turn it on first. That costs nothing and it may be the whole answer.

Your volume is low. Under roughly 500 invoices a month the manual baseline is around CAD $1,200 a month, and a build does not pay back against that in any reasonable period. The fit check runs that arithmetic for your numbers in your browser.

When a custom invoice pipeline pays back

Volume in the thousands rather than the hundreds. Mixed formats — PDFs, scans, spreadsheets and images — rather than one clean supplier portal. No usable API into your ledger, so what you actually need is a validated import file. Or a residency requirement you need named in a contract rather than described in a marketing page.

Those are the conditions under which building beats buying. Outside them, buy.

Payment stays behind a human

Automate the reading, the checking and the posting. Do not automate the paying. Every automation we build has a defined approval gate before irreversible actions, and payment is the clearest example there is: an error that posts a wrong code is a correction, and an error that pays a wrong supplier is a recovery.

Measure accuracy before you sign

Ask for a held-out labelled set, sampled from your real invoices, and ask for accuracy reported per supplier rather than as one blended number. It is entirely normal to see 99% on your dozen regular suppliers and materially less on the long tail — and that split tells you exactly where the human queue should sit. A vendor that writes its own test and grades itself against it has told you nothing; that is the subject of self-graded accuracy.

Get the pass threshold written into the statement of work before development starts. On our own builds, if the threshold is not met at acceptance you choose: pay 60% and keep the code, or we run one full remediation cycle at no charge.

What it costs to build with us

One document type across two intake sources is CAD $26,000 over four to six weeks. Three document types with multiple sources and a full review queue is $48,000 over seven to ten weeks. Running it afterwards is $2,500 or $5,000 a month depending on tier, or nothing to us if you run it in your own tenancy.

If you want the question answered properly before committing to a build, that is what the Readiness Assessment is for.

← All insights See AI automation services and prices