<img height="1" width="1" style="display:none;" alt="" src="https://px.ads.linkedin.com/collect/?pid=3960532&amp;fmt=gif">
PredictAP Blog

4 Hidden Costs of "Free" AP Automation in Real Estate

I ran Yardi and finance technology at Colony Capital (now DigitalBridge) across a $60 billion-plus platform and 500,000 invoices a year. I am now the CEO and founder of PredictAP, which sells AP automation, so weigh that accordingly. This is an operator's list, with the receipts.

When AP automation shows up inside a platform you already pay for, the decision makes itself. No new contract. No vendor review. No line item to defend at budget time. Someone checks a box, invoices start getting coded, and "automate AP" comes off the list.

Most real estate owners, operators, and property managers already run approvals and payments through a workflow system like Yardi PayScan or Bottomline NexusPayables. If coding comes along for the ride, why pay for it separately?

Because the license fee was never the expensive part of accounts payable. The expensive parts are the exception queue, the invoice paid twice, the vendor whose bank details quietly changed, and the approval that becomes a formality once a machine does the coding. A tool can be free to license and still be expensive if it leaves those costs where they were.

The four costs

Bundled tools handle routine invoices well. What matters is how much of these four costs is left after they've done their part:

  • Exception queue labor. The cost is where a person has to step in. When exceptions land in one commingled bucket, that step gets slower.
  • Near-duplicate payments. Exact-match rules stop identical invoices. They miss the resend with a revised number or a property suffix.
  • Payment fraud. Coding decides property, GL account, and vendor. Whether the invoice should be paid at all is a separate control, and someone has to own it.
  • Rubber-stamp review. Once invoices arrive pre-coded and mostly right, approvers stop looking.

These aren't the costs on every buyer's checklist: implementation and vendor-data cleanup, integration with the accounting system, training and vendor onboarding, rule maintenance, per-transaction fees, and eventually switching. A bundled tool lowers some of those without making them zero. The four costs here come from operating with the tool, and they rarely get measured.

1. The exception queue and its long tail

Exceptions are the invoices automation can't finish on its own: splits across multiple entities and properties, utility bills with dozens of line items, header-level tax or shipping, statements sent as invoices. And outside construction and capital work, most operating invoices in real estate have no purchase order to match against.

The cost is in the step where a person has to take over. Say you run 5,000 invoices a month and the tool handles 85% on its own. That leaves 750 exceptions (illustrative numbers, not a benchmark). If they land in one commingled bucket, somebody has to work out what is wrong with each one, and whose it is, before anyone can fix it. Meanwhile invoices sit, the AP team is back to manual work, and the bill shows up as late fees, lost early payment discounts, and cash flow that is harder to forecast.

The benchmarks point the same way. Ardent Partners' AP Metrics That Matter in 2026 puts the average cost to process an invoice at $9.90, with 19.9% of invoices needing manual exception handling. In the same research, best-in-class teams process an invoice for $2.67 against $12.50 for everyone else, about 79% less, and move 55.2% of invoices straight through against 31.4%. My read, and it is mine, not Ardent's: most of that gap is the labor of solving problems, not the initial data entry.

Compare a commingled bucket to an exception that arrives with a name and a reason. Hey Bill, here are your two items: one is a new vendor, one is a duplicate. Hey Mary, here are your three: two duplicates and one anomaly that might be fraud, go look. That is what we spent years building at PredictAP (my bias is in the byline).

We have clients who ran both a bundled product and PredictAP. Plenty of invoices flowed straight through the bundled product. What surprised us was the long tail. The hardest items for a given client took 17 days to resolve on PredictAP and almost 90 days on the bundled product.

The cost compounds when recoveries are involved. A wrong allocation on CAM or another pass-through can flow into the recovery calculation. If nobody catches it in reconciliation, the owner absorbs a cost that should have been recovered, or a tenant gets overbilled and it comes back as a rebill.

What to measure: Minutes per exception, whether each exception names an owner and a reason, and days to resolve your slowest items. Then put dollars on it: monthly exceptions, times average minutes each, times loaded labor cost.

2. Near-duplicates look different enough to pass

Most AP systems can be set to stop an exact duplicate: same vendor, same invoice number, same amount. In real estate AP, real duplicates rarely look like that. They come back with a difference just small enough to get past the rule. A vendor resends with a revised number. A split multi-property bill gets a suffix on each piece, and a later resend no longer matches. A utility reissues a corrected bill for nearly the same amount. Not every one of these is a duplicate, but each deserves a look, and an exact-match rule never raises its hand.

One multifamily operator found 124 duplicate invoices posted, worth roughly $118,000, before anyone noticed. Another owner's tool let more than 700 duplicates through to review, where people had to catch them by hand. Recovery takes staff time, strains vendor relationships, and some of the money may never come back. And not all of these are accidents. Deliberate double billing is a known fraud pattern, and it looks exactly like an honest resend.

What to measure: Run the test: same vendor, same amount, different invoice number. See what happens.

3. Payment fraud a coding engine was never built to see

Invoice coding decides where an invoice goes: which property, which GL account, which vendor. It doesn't ask whether the invoice should be paid at all. That gap is where fraud risk lives.

The common schemes aren't sophisticated. An email asks to update a vendor's remittance details. An invoice arrives from a vendor whose name is one letter off from a real one. A familiar vendor bills far outside its normal range. Each one can code perfectly. In the 2026 AFP Payments Fraud and Control Survey, 76% of organizations reported attempted or actual payments fraud in 2025. The FBI's 2025 Internet Crime Report put business email compromise losses at about $3.05 billion. One AFP respondent sent three payments totaling $150,000 on a routine-looking vendor request and got $915 back.

There is a quieter kind too. A solicitation invoice looks like a bill for something you really owe, sent by a third party at a marked-up price. Pennsylvania's annual report costs $7 to file with the state. The Pennsylvania Department of State has warned about a company charging $149 or $199 to file it. Mailings like that go out to everyone, hoping one AP clerk says, oh yeah, we do need to pay that.

Three different controls are in play, and no single product owns all of them:

  • Invoice-level flags can sit with coding: a lookalike vendor name, an amount far outside the normal range, fees layered onto a familiar bill, a solicitation invoice.
  • Bank detail changes belong to vendor master governance.
  • Payment release belongs to the payment process.

For third-party property managers the money belongs to clients, owners, or associations, so a misdirected payment is a client-funds conversation, not an internal write-off. The risk never goes to zero, but it shrinks when vendor master changes sit with someone other than the person coding and approving.

What to measure: Who owns each control, and whether a bank detail change holds payment until someone confirms it through a channel other than the email that requested it.

4. Review becomes a rubber stamp

Once invoices arrive pre-coded and mostly right, approvers stop looking closely, and that isn't a training problem. A 2010 review in Human Factors found automation bias in experts as well as novices, and found that training and instructions don't prevent it. Finance leaders at large REITs have told me the same thing in plainer words: the worry is less that the software is wrong and more that staff stop looking.

Some approval workflows advance any invoice above a confidence threshold, so many invoices never reach a person. That can be a sound design. The failure isn't fewer human touches. It is removing human attention without risk rules, escalation, and an audit trail to replace it. Centralization compounds it, because coding moves from site teams who knew the vendors and the buildings to shared-service or offshore teams who don't.

What to measure: Which invoices advance with no human review, under what rules, and who signed off on the rules.

Six questions before "good enough" becomes the plan

These work on any AP tool, included or paid, because they're about the whole operation: coding and matching, the vendor master, approvals, and payments.

  1. What share of invoices still needs a person, and how long does each one take? Measure the exception queue, not the automation rate.
  2. What happens to an invoice with the same vendor, the same amount, and a slightly different number? If the answer is "it gets paid," you have a near-duplicate gap.
  3. What happens when a vendor's bank details change? Something should hold the payment until a person confirms the change outside of email.
  4. Do invoices ever advance without a human reviewing them? If so, under what rules, and who approved those rules?
  5. Who catches what your site teams used to catch? This matters most if AP is centralizing, outsourcing, or turning over.
  6. What did duplicate and erroneous payments cost last year? Include recovery time and late fees, not only the dollars that left.

Sometimes the bundled tool is enough

If you have a small portfolio, mostly recurring vendors, single-property invoices, and site teams who know the buildings doing the approving, the included tool may be all you need. Take the win.

For everyone else, free software is easy to approve. The harder number to find is what AP costs once the errors, the exceptions, and the exposure are counted too. Find that number first. It should decide how much automation is enough, not the license fee.