How MCA Lenders Use AI to Verify Merchant Affordability During Live Banking Sessions

The Upstart Debate Exposed a Blind Spot in AI Underwriting for MCA

AI underwriting for merchant cash advance has become the industry's most debated topic in 2026, and for good reason. When David Roitblat published his response to Upstart's CEO on deBanked, the core argument landed hard: consumer lending AI models simply do not translate to merchant cash advance. Upstart's vision treats underwriting as a data science problem solved by feeding more variables into a credit model. Roitblat's rebuttal pointed out that MCA underwriting depends on understanding a merchant's real-time cash flow behavior, something no algorithm can assess from a credit bureau file alone.

That debate, however, stopped short of addressing the most practical question MCA funders face every day. Even if your AI model is purpose-built for merchant cash advance, how do you verify that the merchant can actually afford the advance? Balance checks return a single number. Bank statements can be manipulated. Open banking APIs show transaction data but strip away the visual context that reveals whether a merchant is juggling multiple advances, timing deposits to inflate balances, or hiding obligations across accounts. The affordability gap is not a modeling problem. It is a verification problem.

Why Affordability Verification Requires More Than Static Data

The Limits of Balance Snapshots and Statement PDFs

Most MCA funders still rely on some combination of uploaded bank statements and real-time balance checks to assess affordability. Both methods carry serious limitations. A bank statement PDF captures a fixed window, typically 90 days, and presents transaction data in a format that is trivially easy to edit with modern tools. Even legitimate statements lack context. A $15,000 deposit on the fifteenth of each month could be a healthy recurring revenue stream, or it could be a transfer from another account timed to coincide with verification requests.

Real-time balance checks are even more misleading. As we have explored in our analysis of how real-time balance checks create false confidence in MCA underwriting, a single balance figure tells you nothing about the trajectory of that balance, the obligations draining it, or the pattern of deposits that sustain it. A merchant might show $28,000 in their account at 10 a.m. and have $3,000 by 5 p.m. after scheduled payments clear.

What AI Can Detect in a Live Banking Portal Walkthrough

When an applicant records a walkthrough of their banking portal, the visual data stream contains far richer affordability signals than any static document. AI models trained on screen recordings can identify several critical patterns that point to repayment risk.

First, transaction velocity. By watching the merchant scroll through recent activity, AI can assess how frequently money moves in and out, and whether the cadence is consistent with a healthy business or a merchant juggling cash between accounts. Second, obligation density. Live portal views reveal pending transactions, scheduled payments, and recurring debits that do not always appear on downloaded statements. If a merchant has four separate daily ACH debits labeled as advance payments, that stacking risk is immediately visible in the portal but might be obscured in a reformatted PDF.

Third, account navigation behavior. AI vision systems can detect when a merchant hesitates before opening certain accounts, skips tabs that might reveal additional obligations, or navigates in patterns inconsistent with someone who regularly uses their banking portal. These behavioral signals are subtle, but they correlate strongly with concealment attempts. Fourth, balance trajectory across date ranges. When the AI-guided recording instructs the merchant to show specific date ranges, the system can compare ending balances across periods and flag declining trends that suggest the business is contracting.

How AI-Guided Recordings Close the Affordability Verification Gap

The challenge with live verification calls has always been operational. Scheduling calls across time zones, walking merchants through each screen manually, and relying on an underwriter's memory of what they saw is slow, inconsistent, and unscalable. This is precisely why asynchronous verification has gained traction. When a merchant records their banking session through a platform like Exact Balance, the AI-guided floating coach directs them to show account summaries, specific date ranges, and transaction details in a structured sequence. The merchant completes the recording on their own time. The underwriter reviews it on demand.

What makes this approach powerful for affordability assessment is the combination of structured capture and reviewable evidence. Unlike a live call where the underwriter must make judgments in real time, a recorded session can be paused, replayed, and analyzed frame by frame. AI systems can overlay affordability metrics on the recording: flagging transactions that look like existing MCA payments, calculating average daily balances from the visible data, and highlighting date ranges where deposits drop below typical thresholds.

This is not theoretical. The pattern detection capabilities that Exact Balance applies to screen recordings build on the same AI vision techniques described in our coverage of how MCA lenders use AI to detect fake banking sessions in screen recordings. The same models that spot manipulated portals can also assess whether the financial activity shown is consistent with a merchant who can sustain an additional advance.

Real-World Scenarios Where Session-Level Affordability Checks Matter

Catching Stacking Through Portal Observation

Consider a restaurant owner applying for a $75,000 advance. Their uploaded bank statements show healthy deposits averaging $4,200 per day. The balance check returns $31,000. On paper, this looks fundable. During the recorded banking session, however, the merchant scrolls through recent transactions and the AI identifies five separate recurring ACH debits labeled with abbreviations matching known MCA funder payment processors. The total daily obligation from existing advances exceeds $1,800, meaning the merchant is already sending 43% of daily revenue to existing funders. Adding another advance would push the total remittance rate past sustainable levels.

No uploaded statement flagged this. The balance check certainly did not reveal it. Only the live portal walkthrough, captured on video and analyzed by AI, exposed the stacking risk.

Assessing Seasonal Business Affordability

A landscaping company applies for funding in March, showing strong prior-year revenue. The AI-guided recording asks the merchant to display account activity for November through February. The portal reveals that deposits during winter months dropped to 30% of peak levels, while fixed obligations remained constant. The business survived the slow season, but barely. Funding a new advance that requires daily remittance through another slow season would create severe cash flow stress.

This kind of seasonal affordability analysis is only possible when the underwriter can see the full date range in the banking portal. Uploaded statements might cover only the most recent quarter, which for a seasonal business is always the best quarter. The AI-guided recording ensures the merchant shows the periods that matter most for repayment risk assessment.

Affordability and Concentration Risk in Large Deals

The deBanked reporting on how big MCA deals go catastrophically wrong highlights what happens when affordability is not verified at the session level. In the 1 Global Capital case, a $40 million exposure to a single dealership conglomerate collapsed when the business failed. The underlying question in every large deal is whether the merchant's actual banking activity supports the repayment schedule. For deals above $500,000, the cost of a thorough verification session is negligible compared to the loss from a default. Yet many funders still rely on the same lightweight checks they use for $25,000 advances.

AI-powered session analysis scales to match deal size. For larger advances, underwriters can request extended recordings covering multiple accounts, longer date ranges, and specific transaction categories. The AI processes the same visual data regardless of deal size, applying consistent affordability criteria that do not degrade under time pressure.

Frequently Asked Questions

How does AI assess merchant affordability during MCA bank verification?

AI assesses affordability by analyzing visual data from recorded banking portal sessions. The system identifies recurring obligations, calculates visible average daily balances across date ranges, detects transaction patterns consistent with existing MCA payments, and flags declining deposit trends. Unlike static document analysis, session-level AI review captures the full context of a merchant's financial activity as it appears in their live banking portal, including pending transactions and scheduled payments that may not appear on downloaded statements.

Can AI detect MCA stacking from bank portal recordings?

Yes. AI vision models can identify recurring ACH debits that match known MCA payment patterns, even when transaction descriptions use abbreviations or generic labels. By analyzing the frequency, timing, and amounts of outgoing payments visible during a portal walkthrough, the system estimates total existing obligations and calculates the merchant's effective remittance rate. This allows funders to determine whether adding another advance would push repayment demands past sustainable thresholds.

Why not use open banking APIs instead of screen recordings for affordability checks?

Open banking APIs provide structured transaction data, but they strip away visual context that matters for affordability verification. APIs do not show pending transactions, scheduled future payments, or the merchant's navigation behavior across multiple accounts. They also do not capture what the merchant chooses to show or hide. Screen recordings provide a visual audit trail that includes all of this context, making them a stronger evidence base for affordability decisions. Many funders use APIs for initial screening and screen recordings for final verification, treating them as complementary rather than interchangeable.

How long does an AI-assisted affordability verification take with Exact Balance?

The merchant typically completes their guided recording in under ten minutes. The AI processes the recording and generates affordability flags as part of the review workflow. An underwriter can then watch the recording, review the AI-generated activity log, and make a verification decision in a single session. Because the entire process is asynchronous, there is no scheduling overhead. Most verifications move from request to decision within the same business day.

Conclusion

The debate over AI underwriting for merchant cash advance will continue, but the practical question has a clear answer. Affordability verification works when it is grounded in observable banking behavior, not just modeled predictions. Screen recordings of live banking sessions give underwriters the visual evidence they need to assess whether a merchant can sustain repayment, and AI makes that analysis faster, more consistent, and harder to game.

Exact Balance brings this capability to MCA funders through asynchronous, browser-based recordings guided by AI. No scheduling. No software installs. Just structured, reviewable evidence of a merchant's real financial position. Visit exactbalance.ca to see how async verification with AI-powered affordability analysis fits into your underwriting workflow.