AI Washing in Transaction Coordination: The Five-Signal Test

AI washing is when transaction coordination software wraps ordinary rule-based automation in AI language, if-then logic relabeled as “intelligent,” a template dressed up as a recommendation engine. With most TC platforms now claiming some form of AI, the only way to know what you are actually buying is to test it. Here is a five-signal test that separates real AI from a marketing refresh.

What AI washing looks like in transaction coordination software

Every TC platform now says “AI-powered” somewhere on its homepage. Ava. Smart deadline tracking. Intelligent contract review. The words are everywhere, and that is exactly the problem. A recent industry analysis from Onyx Technologies describes AI washing as “layering the language of artificial intelligence onto software that, underneath, is doing nothing more than what it was always doing.” A chatbot bolted onto an old dashboard. A report that looks cleaner but does not learn anything. Automation that was scripted years ago, just renamed.

The pace makes this worse. According to MRI Software’s 2026 PropTech outlook, the share of corporate real estate firms running AI pilots jumped from 5% to 92% in three years, citing JLL research. That is not 92% of firms running real AI. It is 92% of firms with something on their roadmap labeled AI, real, cosmetic, or somewhere in between. For a transaction coordinator deciding which tool protects a file when it counts, that gap matters more than any feature list.

So before the next demo, run the file through five specific tests designed to catch AI washing. Real AI passes all five. Rule-based automation, however well designed, usually fails at least two.

The Five-Signal Test

These five signals were built specifically for transaction coordination, not lifted from generic proptech buying guides. Each one maps to something a TC does on a file every week: reading a contract, handling a change, learning from repetition, catching what nobody flagged, and fitting into a real workflow. Fail two or more and the rest of the feature list stops mattering.

Signal 1: The Contract Read Test

Ask the vendor this directly: when a contract is uploaded, does the system extract dates, parties, and contingencies from the actual document, or does a human still need to key in the closing date and contingency deadlines by hand? Real contract-reading AI parses the PDF or scanned copy and pulls structured data out of unstructured text. If the “AI” step is really a data-entry form with a friendlier interface, that is automation, not extraction.

Signal 2: The Cascade Test

This is the one that separates the two categories fastest. Take a real scenario: a buyer removes the appraisal contingency 10 days early via a signed amendment on a 45-day escrow. In a genuinely AI-native system, that single event should automatically re-derive every downstream deadline tied to it, financing contingency reminders, the earnest money release trigger, the final walkthrough scheduling window, without a human re-entering a single date. In a rule-based system, the amendment gets logged as a document, but the task list does not know it changed. Someone has to open the file, manually recalculate what shifted, and re-enter each date by hand.

Run the math on what that actually costs. A TC handling 25 active files a month sees a contingency change, amendment, or extension on roughly 40% of them, call it 10 events a month. Manually tracing a cascade and re-entering the affected dates runs 12 to 18 minutes per event once you account for checking every dependent task. That is 2 to 3 hours a month spent doing arithmetic the software should have done on its own, on just one TC’s caseload. At 100 files a month, the same ratio turns into 8 to 12 hours, more than a full working day, burned on manual re-derivation that a cascade-aware system would have handled the instant the amendment was signed.

Signal 3: The Pattern Test

Ask how file 500 is handled differently from file 5. A system with real learning should get faster and more accurate as it processes more of a given TC’s or brokerage’s files, recognizing which lender consistently sends title docs late, which MLS form version a specific brokerage uses, which disclosure a particular county always requires even when the statewide checklist does not list it. If file 500 runs through the exact same static logic as file 5, nothing was learned. It was just executed, correctly, but without improvement.

Signal 4: The Judgment Test

Every TC tool can flag what it was explicitly programmed to check: missing signature, missing date, missing disclosure form. The judgment test is different: does it catch a combination nobody explicitly coded for? An unusual pairing of property type, state, and financing that historically causes problems. A pattern of late responses from one specific lender on this specific deal type. Rule-based systems only catch what someone anticipated and wrote a rule for. Real AI recognizes patterns nobody explicitly asked it to watch for.

Signal 5: The Workflow Test

Where does the intelligence actually live? If it sits inside the same inbox, calendar, and task list a TC already works in, invisible, running in the background, that is AI built into the process. If it requires opening a separate dashboard, a distinct AI module, or toggling into a different tab to get any value from it, the AI was bolted on afterward rather than built into the core product. A tool that makes a TC do extra work to access its own intelligence has failed the most basic test of usefulness.

Score your TC software for AI washing: the Five-Signal scorecard

This is how you catch AI washing before it costs you a missed deadline instead of after. Run any tool you are evaluating, or already paying for, through all five signals and score one point for each it genuinely passes. Do not round up on a vendor’s word; ask for a live demo of each specific behavior.

Score: 4 to 5 means the tool is doing genuine AI-native work. 2 to 3 means a hybrid, some real machine learning layered on a rules engine, which is common and not automatically disqualifying, but worth pricing accordingly. 0 to 1 means you are paying an AI premium for automation that a well-built rules engine from five years ago could have delivered just as well.

The test applies a little differently depending on who is asking. A solo TC evaluating a $30 to $60 a month tool mostly cares about signals 2 and 5, the cascade and workflow tests, since those are what save actual hours on a caseload. A broker or team lead evaluating software for a whole office should weigh signals 3 and 4 more heavily, pattern and judgment, because those are the ones that scale in value as file volume grows, and AI washing on those two signals is the hardest to catch from a sales demo alone.

Why the AI washing distinction matters for a TC’s caseload

This is not just a pricing question. A TC or brokerage that assumes cascade-level intelligence exists when it does not is one missed re-derivation away from a blown deadline, a broker liability exposure, or an earnest money dispute nobody caught in time. States vary widely on how much statutory nuance a TC is expected to track without error; a California TC handling the state’s specific disclosure and timeline requirements is carrying real exposure if the “AI” they were sold turns out to be a checklist with a chatbot attached. Knowing which of the five signals a tool actually passes is the difference between trusting the software to catch what you would catch yourself, and quietly doing that work twice.

It also changes how a TC should think about capacity. If you have ever wondered what makes a transaction coordinator AI-native rather than just automated, the five signals above are the practical version of that definition. And if you are weighing AI software against a virtual TC or an in-house hire, the same test applies before you compare cost, since an AI vs. virtual vs. in-house comparison only holds up if the “AI” option is actually AI.

How ReBillion holds up against its own test

Running ReBillion through its own five signals: contract terms are extracted directly from uploaded documents rather than keyed in by hand, an amendment or contingency removal re-derives every downstream deadline automatically, and the system builds on patterns across a brokerage’s file history rather than treating every deal as a blank slate.

Anomaly detection surfaces combinations that were not individually pre-programmed, and the entire process runs inside the TC’s existing inbox, calendar, and voice-agent outreach rather than a separate module someone has to remember to check. For the fuller picture of how that stack fits together end to end, the full TC tech stack breakdown covers what a 100-deal-a-month coordinator is actually running underneath it. Pricing and a live walkthrough of the cascade behavior described above are on the ReBillion pricing page.

Frequently asked questions

What is “AI washing” in transaction coordination software?

AI washing is marketing a product as AI-powered when the underlying functionality is rule-based automation, if-then logic, static templates, or a chat interface layered on top of unchanged workflows, rather than a system that reads, learns, and adapts.

What is the Five-Signal Test?

It is a five-part check built specifically for TC software: whether the system reads contracts directly, whether it re-derives cascading deadlines after a change, whether it improves with repeated use, whether it catches issues nobody explicitly coded for, and whether it runs inside your existing workflow rather than a separate module.

Can a TC tool use AI for one task and rules for everything else?

Yes, and this is the most common pattern in the market today. A tool might use genuine contract-reading AI but fall back to static rules for deadline cascades. This is exactly how AI washing hides inside otherwise legitimate software. Score each signal independently rather than accepting one strong feature as proof the whole platform is AI-native.

Does AI washing mean a tool is bad?

Not necessarily. A well-built rules engine can be reliable and worth paying for. The problem is paying an AI premium, in price or in trust, for a system that cannot actually do what AI-native tools do, particularly the cascade and judgment tests, which is where missed deadlines tend to originate.

How do I ask a vendor about their AI without getting a marketing answer?

Ask for a live demo of a specific scenario, not a slide. Have them remove a contingency on a test file in front of you and show whether the downstream dates update on their own. Vague answers about how the platform “does AI” usually mean the vendor has not been asked this before.

Is ReBillion’s AI real, or is this just marketing?

Run it through the five signals above during a live walkthrough: contract extraction, cascade re-derivation, pattern recognition across files, anomaly detection, and workflow-native execution. The pricing page includes a request for a live demo built around exactly this test, since ruling out AI washing is the entire point of the framework.

Written by Vikas Malpani, CEO and co-founder of ReBillion and a CAR-certified transaction coordinator. Connect on LinkedIn.

Written by Vikas Malpani

Vikas Malpani is the CEO and Co-Founder of ReBillion and a CAR-Certified Transaction Coordinator. A serial real estate technology entrepreneur with 15+ years across technology and real estate operations, he was named to MIT Technology Review's TR35 list of young innovators. At ReBillion he leads the AI systems that deliver compliant, accurate transaction coordination for brokerages and agents across all 50 US states. Connect with Vikas on LinkedIn: https://www.linkedin.com/in/vikasmalpani/

View Profile