Skip to main content
Viithiisys
Back to blog
AI Engineering6 min readFri, Nov 20, 2026

RPA in finance: three workflows worth fixing

RPA in finance suits three jobs: invoice matching, bank reconciliation and month-end close. See how each breaks, what drives cost, and what to leave to people.

Gaurav Saini

Founder, Viithiisys

RPA in finance: three workflows worth fixing

Provenance: Viithiisys sells AI workflow automation, including builds like the ones below, so weigh this post accordingly. The three workflows are designs, not delivered projects, and the benchmark figures come from the published sources linked in the Sources list.

Finance teams reach for RPA when the same screens get re-keyed every day and the month-end close keeps eating the first week of the month. It is a good fit for some of that work and a poor fit for the rest. Here are three finance workflows where RPA earns its keep, how each one runs today, where it breaks, and what to leave to a person.

What is RPA in finance?

RPA in finance is software that repeats rule-based steps a person would otherwise do on screen: reading an invoice, matching it to a purchase order, or moving a bank line into the ledger. It follows fixed rules, so anything outside those rules has to go to a person.

That last point decides whether a build works. The routine path is easy to automate. The exception path is where finance teams lose their time, and it is where a design has to be honest about people.

Which finance workflows are worth automating first?

Pick work that is high volume, follows clear rules, and has a cost you can measure. The three below meet all three tests. For a wider view of where finance automation breaks, see our post on financial process automation.

How does invoice matching run today, and where does it break?

Today an AP clerk opens an emailed PDF, keys the header and lines into NetSuite, QuickBooks or SAP, and checks the result against a purchase order and a goods receipt. Then it goes to an approver. Most of the time sits in the chasing, not the keying.

It breaks in four places:

  • A supplier changes its invoice layout and the extraction rule stops matching.
  • A price or quantity differs from the order, and the invoice sits in a queue nobody owns.
  • The same invoice arrives twice, once by email and once through a portal.
  • The vendor's bank details change on an invoice, which is a fraud signal and not a data problem.

The cost is real and published. The American Productivity and Quality Center reports a median of $6.00 to process one AP invoice across 5,846 organizations, covering people, systems and overhead, in its invoice cost benchmark. APQC data cited by its own CFO also puts duplicate or erroneous payments at 0.8% of disbursements for top performers and 2% for bottom performers, in its CFO.com column.

What to automate: capture, field validation, the two-way or three-way match, and routing. What stays with a person: any mismatch, any change to bank details, and approvals above a set amount. For the capture step in depth, read our post on invoice processing automation.

Where does bank reconciliation break?

Today someone exports the bank statement to a spreadsheet, exports the ledger, and matches lines by eye or with a VLOOKUP. Whatever does not match gets chased by email.

It breaks where the data does not line up:

  • One bank deposit covers several invoices, so there is no one-to-one match.
  • The bank's description text changes between months.
  • Timing differences push a payment into the next period.
  • Fees and foreign exchange differences show up as unexplained small gaps.

A rule-based bot handles the one-to-one matches well. It needs a defined tolerance for the small gaps and a queue for the rest, with each unmatched item assigned to a named person and a date.

What to automate: statement import, exact and rule-based matching, and the unmatched report. What stays with a person: anything the rules cannot explain, and the sign-off on the reconciliation itself.

The bot should finish the easy matches and make the hard remainder impossible to ignore.

Ventana Research found that companies that automate reconciliations are more likely to close within six business days, 69% against 53% for those that do not, in its Office of Finance research.

A reconciliation queue with matched lines, rule-matched lines and a short list of exceptions assigned to named owners

Why does the month-end close slip, and what can a bot do?

Today the controller keeps a checklist in a spreadsheet, chases each owner by email, and waits on accruals, journal entries and approvals. The close slips when one owner is late and nobody sees it until the last day.

RPA does not close the books. It handles the pieces around the edges: pulling the same reports on the same day, posting recurring journals, and reminding owners as a task falls behind. The same Ventana research reports that 61% of participants complete the monthly close within six business days, and that the share of companies applying a substantial amount of automation to the close rose from 11% to 25%.

The close breaks when a recurring journal is posted with last month's amount, when a report is pulled before a feeder system finishes, and when a bot's task has no owner.

What to automate: scheduled report pulls, recurring entries with a review step, and task status with reminders. What stays with a person: accruals that need judgement, estimates, and the final review.

A close checklist with owners, due dates and a status for each task

What does it cost to build, and how do you compare options?

Our price is set per project after we read the workflow, and we publish no rate card. These are the drivers that move it, so you can compare quotes:

DriverWhy it moves the price
Number of systems touchedEach ERP, bank portal or email source needs its own connection and its own tests
Exception volumeEvery exception path needs a rule or a queue
Document varietyMore supplier layouts mean more extraction rules to maintain
Controls and audit trailApproval steps, logs and segregation of duties add design time
Run and supportBots need an owner, monitoring and a fix when a screen changes

A useful test of any quote is whether it names the exception queue and the person who owns it. If it does not, the cheapest-looking option is leaving that work with you.

When should you skip RPA?

RPA is the wrong choice when the process itself is broken, when the volume is small, or when the rules change every quarter. Automating a bad process makes it fail faster. If you cannot write the steps down on one page, fix the process first.

Also check whether your ERP or accounting package already includes the feature, since a built-in invoice matching setting costs less to run than a bot that copies it.

Tell us about the workflow that keeps getting stuck

If one of these workflows is the one slowing your close, send it to us through the broken workflow assessment. Tell us the systems, the volume and where it jams. You will get an honest view of whether it needs a build, a configuration change or a better handoff. You can also read about our AI workflow automation service.

Sources

  • APQC, What is the average cost to process an invoice: apqc.org
  • CFO.com, Metric of the Month, detect and prevent duplicate or erroneous payments (APQC Open Standards Benchmarking): cfo.com
  • ISG Ventana Research, Finally, a faster close: research.isg-one.com

FAQ

What is RPA in finance?
Robotic process automation in finance means software that repeats rule-based steps a person would otherwise do on screen, such as keying invoice data, matching it to a purchase order or moving bank lines into a ledger. It follows fixed rules and hands anything unusual to a person.
Which finance processes should you automate first?
Start with high-volume work that follows clear rules and has a measurable cost, like invoice capture and matching or bank reconciliation. Leave judgement calls, such as disputed invoices and approvals above a set amount, with people until the exception paths are well understood.
Does RPA replace a finance team?
No. RPA takes the routine path and leaves exceptions, approvals and judgement to people. A finance team that adopts it usually spends less time keying and chasing, and more time on exceptions, controls and analysis, with a named owner for every bot.