Why Invoice Coding, Not Data Extraction, Is CRE's Real AP Challenge

A $4,200 landscaping invoice arrives at a mixed-use property. It looks routine. But in commercial real estate, that single bill may need to be split across multiple legal entities, allocated between residential and retail square footage, coded to different general ledger accounts, verified against contract pricing, and tagged for common area maintenance recovery.

Miss one of those decisions and the error cascades—into tenant billings, recoveries, budgets, property-level reports, and ultimately net operating income. That downstream impact is what the industry has overlooked in its rush to automate accounts payable, according to David Stifter, co-founder of PredictAP and former CTO at Digital Bridge.

Why it matters

Most CRE firms have solved invoice reading—optical character recognition can extract vendor names, dates, and amounts reliably. But the harder problem is invoice understanding: determining which entity bears the cost, whether it's recoverable, and how to split it correctly. That logic still lives primarily in the institutional memory of experienced AP professionals, creating a scalability bottleneck and a single point of failure as portfolios grow and staff turns over.

The judgment problem hiding in AP workflows

Stifter spent more than twenty years running technology at a global real estate investment firm before founding PredictAP in 2020. He watched AP professionals make decisions no purchased system could replicate—looking at an invoice and knowing exactly how to handle it based on accumulated context: vendor history at that property, ownership structure, lease terms, and unwritten coding conventions developed over time.

A landscaping invoice doesn't announce that it needs to be split between two entities with part coded to CAM and one line treated as capital because the property is mid-renovation. That decision draws on thousands of similar patterns a veteran AP coder carries mentally.

Optical character recognition can read an invoice. Robotic process automation can execute a fixed workflow. But CRE coding decisions aren't fixed—they shift with ownership changes, lease modifications, renovations, and portfolio restructurings. The intelligence required to code correctly usually doesn't exist on the invoice itself; it exists in the relationship between that invoice and everything that came before it.

From tribal knowledge to system intelligence

When coding logic migrates from individual memory into a learning system, it becomes scalable. Each correction, new vendor relationship, and resolved exception adds to the knowledge base instead of disappearing when someone leaves. The better platforms are now judged not on clean data extraction but on whether they can handle the accounting logic that used to be tribal knowledge: allocations, entity logic, recoverability, and review-ready coding.

Once invoice coding becomes reliable, the invoice stream becomes useful beyond payment execution. It can surface vendor pricing drift—the same service costing 15 percent more than two years ago with no authorized contract change. It can highlight missed accruals before close. It can expose misclassified costs that weaken tenant recoveries.

It also restores granularity that gets lost when an invoice flattens into the ledger. A $50,000 appliance expense may look acceptable in a report until someone asks what it actually bought—two restaurant-grade Wolf ovens that exceeded approval thresholds, or 25 units of the approved GE model. The ledger won't tell you. The invoice will.

The fraud dimension

Generative AI is making fraudulent invoices easier to create and harder to distinguish from legitimate ones. A convincing fake vendor invoice used to require effort; AI collapses that effort to near zero. The threat includes fabricated service invoices, inflated charges from compromised vendor accounts, and duplicate billings designed to disappear in busy AP queues.

A human reviewer examines one invoice at a time. A bad actor using AI can generate thousands of plausible invoices and rotate them across properties, entities, and vendors faster than any team can track manually. Finance teams will need AI working permanently on defense—flagging anomalous billing patterns, catching vendor inconsistencies, and surfacing signals no busy AP team could reasonably catch alone.

The strategic shift

As automation improves, the AP role evolves: less time on repetitive coding, more on oversight, controls, and financial analysis. The people who understand vendor relationships and cost structure are exactly who should spend time on higher-value work.

In an industry where the margin between a good year and a difficult one often comes down to financial data quality, the invoice is one of the most information-dense documents in the building. It appears before the accrual, before the report, before the audit. It's where a property's financial story begins.

The details in this analysis were first reported by Propmodo in a piece by Stifter published in June 2026.