ContinuServe launched ContinuFlow on July 21 with a reported invoice extraction accuracy above 95%. NetSuite estimates that a midsize company can move from $16.50 per invoice to about $3.25 with accounts payable automation. At 2,000 invoices a month, that difference is worth $318,000 a year.
Those numbers explain why finance automation is moving up the priority list. They also create a dangerous expectation. Reading an invoice correctly is not the same as paying it correctly.
The useful business question is not, "Can AI read our invoices?" It can. The question is whether automation can reduce manual work without weakening approval controls, vendor trust, or visibility over cash.
What accounts payable automation actually does
Accounts payable, usually shortened to AP, is the process between receiving a supplier invoice and paying it. In many companies, that process still involves a shared inbox, spreadsheets, forwarded emails, manual data entry, and somebody asking who approved what.
Modern AP automation connects those steps.
The software reads an invoice using optical character recognition, or OCR. OCR turns the words and numbers in a PDF or image into structured data. A machine learning model then identifies fields such as the vendor, invoice number, tax amount, due date, and purchase order.
The system checks that information against existing records. If the invoice matches the purchase order and receipt, it can assign the correct general ledger code and send the bill through the approval workflow. General ledger coding simply means placing the expense in the correct accounting category.
The best systems do not hide exceptions. They route a duplicate invoice, an unexpected bank-account change, or a price mismatch to a human with the relevant evidence attached.
That distinction matters. Good automation removes routine touches. Bad automation removes scrutiny.
The 80% saving is possible, but it is not automatic
NetSuite's 2026 business case uses a five-person AP team processing 24,000 invoices a year. Once labor, paper, postage, errors, and rework are counted, the example reaches $16.50 per invoice. The estimated automated cost is $3.25.
That is an 80% reduction.
It is also an illustration, not a promise. A company processing 200 invoices a month will not have the same economics as one processing 20,000. Complicated approval rules, poor supplier data, and a heavily customized ERP can push implementation costs up quickly.
We would calculate the current cost before recommending any platform:
-
AP salaries and the share of time spent processing invoices
-
Time that managers spend approving or chasing documents
-
Late fees, duplicate payments, and missed early-payment discounts
-
Software, integration, training, and ongoing support costs
Suppose a company handles 1,000 invoices a month at a measured cost of $12 each. If automation reduces that to $5, the gross saving is $84,000 a year. A $70,000 first-year project barely pays back within 12 months. A $150,000 implementation does not.
That is why "add AI" is a poor buying strategy. Volume and process quality decide the return.
Why the latest finance platforms matter
ContinuFlow is a useful signal because it does more than capture invoice text. The platform connects AP with general accounting, cash-flow reporting, and approval tracking. It works with SAP and Sage, and ContinuServe says standard invoice capture and coding can exceed 95% accuracy.
The important part is the connection to the rest of finance.
A standalone AI tool that copies invoice data into a spreadsheet may save typing time. It can also create another system that employees must reconcile. The stronger pattern is a controlled workflow where supplier records, purchase orders, approvals, payment status, and the ledger stay synchronized.
For a business owner, that produces a better outcome than a faster data-entry screen. It creates a live view of upcoming obligations and exceptions. The finance team can see what needs attention before a missed payment affects a supplier relationship.
Start with the invoices that should be boring
AP projects often begin with the hardest invoice because it looks impressive in a demo. We disagree with that approach.
Start with high-volume invoices from known suppliers where the purchase order, receipt, and bill usually match. These are the transactions most likely to pass through without human correction. They produce a measurable result quickly and expose integration problems without putting unusual payments on autopilot.
Keep manual approval for:
-
New vendors
-
Changes to bank details
-
High-value invoices
-
Payments with no purchase order
-
Tax or quantity mismatches
We would also separate invoice approval from payment release. The person who approves a bill should not be the only person able to change the supplier's bank account and send the money. Automation should enforce that separation, not work around it.
Measure exceptions, not just accuracy
A vendor may claim 95% extraction accuracy. That can still mean errors across 50 fields for every 1,000 fields processed. If the mistakes land on invoice totals or bank details, the average accuracy number is not comforting.
Track the percentage of invoices that complete without a human touch. This is called straight-through processing. Also track exception reasons, approval time, duplicate-payment prevention, on-time payment rate, and cost per invoice.
Review those numbers by supplier. A single vendor sending inconsistent PDFs can create a large share of the manual work. Fixing its submission format may produce more value than changing the AI model.
Do not allow the system to learn silently from every correction. Finance teams need an audit trail showing what changed, who approved it, and which rule or model made the recommendation.
When custom development is worth it
Buying a mature AP product is usually the right answer when the company uses a common ERP and follows standard purchasing processes. Building the entire system from scratch would be expensive and unnecessary.
Custom development makes sense around the edges. A multi-location retailer may need to match invoices against store-level deliveries. A logistics company may need freight documents checked before payment. An Indian business may need GST fields validated against its accounting workflow. Existing software may cover 80% of the process while the missing 20% still consumes most of the team's time.
That is where an integration layer, a focused AI document workflow, or a custom approval portal can earn its cost.
The limitation is straightforward. AI cannot repair an undefined process. If nobody owns vendor data and approval rules change depending on who asks, automation will reproduce the confusion faster.
Make the first decision with your own numbers
Accounts payable automation can cut invoice costs sharply. It can also improve cash visibility and keep a growing company from adding finance headcount just to move PDFs between inboxes.
But the first deliverable should be a baseline, not a bot. Measure invoice volume, processing cost, exception rate, and approval time. Then automate one controlled path and compare the result.
At Axentia, we build the integrations and custom AI workflows that connect finance processes to the software a business already uses. If your AP team is buried in repetitive invoice work and the off-the-shelf options do not quite fit, book a call with us at axentia.in.