How OnDeck's SMB Growth Confidence Reshapes AI Underwriting for Merchant Cash Advance

SMB Growth Confidence Is Surging and MCA Underwriting Pipelines Are Not Ready

When OnDeck reports that 93% of small business owners feel confident about growth, the natural reaction for MCA funders should not be celebration. It should be concern about throughput. That level of optimism translates directly into loan and advance applications. More merchants calling brokers. More bank statements landing in inboxes. More deals competing for the same underwriting hours. For funders still relying on manual review of PDF statements, spreadsheet-based revenue calculations, and phone-tag document collection, the math breaks quickly. AI underwriting for merchant cash advance is no longer a future aspiration; it is the operational floor required to stay competitive in 2026.

This article examines what the current wave of SMB confidence means for MCA underwriting capacity, how platform lenders are pulling ahead with data-driven decisioning, and what independent funders must build or buy to keep pace. The gap between a merchant expressing interest and a funder making a decision is where deals are won or lost, and that gap is getting narrower every quarter.

Platform Lenders Are Setting the Underwriting Bar

Shopify and Square: The Data Moat in Action

Shopify Capital originated $1.4 billion in small business loans and merchant cash advances in Q2, with CFO Jeff Hoffmeister noting that "capital was a larger driver this quarter" while loss rates remained normalized. Square, meanwhile, continues to post loss rates below 4% across every economic cycle, with CEO Jack Dorsey emphasizing that "Square Loans are underwritten on data banks can't see." These are not abstract technology demos. They are production-grade underwriting engines processing billions in originations with loss rates that most independent MCA funders would envy.

The advantage is structural. Shopify sees every transaction flowing through a merchant's store. Square sees every card swipe, every deposit, every refund. Their underwriting models ingest real-time revenue data, not four-month-old PDF bank statements that a broker emailed on a Tuesday. As we explored in our analysis of how Square's sub-4% loss rates prove the case for AI underwriting, this data depth is the single largest driver of their performance advantage.

Independent funders do not have this luxury. They cannot see a merchant's POS transactions or e-commerce dashboard. But they can close the gap by applying AI to the data they do have: bank statements, application forms, voided cheques, and government IDs.

The Throughput Ceiling in Manual Underwriting

Consider a funder processing 200 applications per week. Each deal requires collecting four months of bank statements, a government ID, a void cheque, and a signed application. A junior underwriter spends 25 to 40 minutes per file extracting average monthly revenue, counting NSFs, calculating daily balances, and flagging anomalies. At that rate, one underwriter handles roughly 12 to 15 files per day. To process 200 applications, a funder needs 13 to 17 underwriters working full days, and that is before accounting for document chase time, re-requests for illegible scans, and the inevitable Friday afternoon backlog.

When SMB confidence is high and application volume spikes 20% or 30%, the math falls apart. Hiring is slow. Training is slower. The alternative is cutting corners on review depth, which is how defaults creep in and how stacking fraud slips through.

AI-powered bank statement analysis changes this equation. Machine learning models trained on thousands of bank statement formats can extract deposits, withdrawals, daily balances, and NSF counts in seconds, not minutes. Document classification models sort uploaded files into the correct categories automatically. And anomaly detection flags patterns that human reviewers miss when fatigued, like round-number deposits that suggest fabricated statements or sudden revenue spikes that may indicate stacking.

Where AI Extraction Outperforms and Where It Does Not

Precision matters here. AI bank statement extraction is not infallible. Scanned PDFs with poor resolution, statements from smaller credit unions with non-standard formatting, and handwritten annotations all present challenges. The best systems combine optical character recognition with large language model parsing, cross-referencing extracted totals against running balances to catch errors before they propagate into underwriting decisions.

The key insight is that AI does not need to be perfect. It needs to be faster and more consistent than a human reviewer working their 14th file of the day. When an AI extraction pipeline flags a discrepancy, a human underwriter reviews it. When the numbers reconcile cleanly, the deal moves forward without manual intervention. This hybrid approach, often called human-in-the-loop automation, is how the most effective MCA operations scale without ballooning headcount.

As the Federal Reserve's most recent Small Business Credit Survey confirms, non-bank lenders now process the majority of small business financing applications in the United States. The volume is not going down. The only question is whether a funder's infrastructure can absorb it.

Closing the Gap Between Lead Contact and Funded Deal

Why Outreach Speed and Document Collection Are the Same Problem

Most funders think of outreach and underwriting as separate functions. The sales team contacts leads and books callbacks. The operations team collects documents and underwrites deals. In practice, the handoff between these two stages is where the most time bleeds out of the pipeline.

A broker contacts a merchant on Monday. The merchant expresses interest. The broker sends a follow-up email requesting bank statements. The merchant does not respond until Wednesday. The broker follows up again Thursday. The merchant finally uploads statements Friday afternoon. By Monday, the underwriting queue is backed up, and the deal sits for another two days. Total elapsed time from first contact to underwriting decision: eight to ten business days. In that window, two or three other funders have already made offers.

This is why platforms like Let's Submit combine AI-powered outreach with asynchronous document collection in a single workflow. When Sabbie, the AI sales rep, texts a merchant and qualifies them for funding, it immediately shares a secure upload link. The merchant can drop bank statements, a government ID, and a void cheque from their phone in two minutes. No email chains. No waiting for a broker to forward files. No lost attachments. By the time a human advisor calls back, the documents are already collected, parsed, and ready for review.

Asynchronous Verification as a Competitive Edge

The concept of async bank verification is simple: let the merchant upload documents on their own time, from any device, without requiring a live call or screen share. The execution is where most tools fall short. Upload links need to work on mobile without app downloads. File formats need to accommodate everything from crisp digital PDFs to photos of paper statements taken under fluorescent lighting. And the entire flow needs to feel trustworthy, because merchants are uploading sensitive financial data to a company they may have heard of for the first time 30 seconds ago.

Let's Submit handles this with bank-level encryption, a clean mobile-first upload interface, and AI extraction that pulls revenue, deposits, and key fields automatically. The funder or broker gets a pre-populated application with average monthly revenue, average daily balance, NSF count, and time in business, all extracted without manual data entry. The underwriter's job shifts from data entry to decision-making, which is where their expertise actually matters.

For a deeper look at how this workflow compares to traditional methods, our piece on OnDeck's growth confidence report and bank verification software breaks down the operational benchmarks that separate high-throughput funders from the rest of the market.

What SMB Growth Confidence Means for Fraud Exposure

High confidence periods are also high fraud periods. When more merchants apply for funding, the absolute number of fraudulent applications increases even if the fraud rate stays constant. Fabricated bank statements, synthetic identities, and stacking schemes all spike during origination surges because fraudsters know that overwhelmed underwriting teams are more likely to miss red flags.

AI fraud detection addresses this by applying consistent scrutiny to every file, regardless of volume. Machine learning models trained on known fraudulent statements can flag pixel-level inconsistencies in PDF metadata, detect round-number deposit patterns that suggest fabrication, and cross-reference business names against databases of known bad actors. These checks happen in parallel with extraction, adding seconds rather than hours to the review process.

The alternative, relying on an experienced underwriter's gut feeling to catch fraud, works when volume is low. It fails spectacularly when that underwriter is reviewing 20 files a day and the 18th one contains a well-crafted fake. The 2026 Inc 5000 list of fastest-growing funders includes companies scaling originations by 500% to nearly 2,000% over three years. At that pace, manual fraud detection is not a bottleneck; it is a liability.

Frequently Asked Questions

What is AI underwriting for merchant cash advance?

AI underwriting for merchant cash advance refers to the use of machine learning models, optical character recognition, and natural language processing to automate the analysis of bank statements, application documents, and cash flow data during the MCA decisioning process. Instead of a human manually reviewing each page of a bank statement to calculate average monthly revenue and flag NSFs, AI systems extract this data in seconds. The technology does not replace human judgment on final funding decisions but eliminates the repetitive data entry and initial review that consume most of an underwriter's day.

How do MCA funders use AI to analyze bank statements?

Funders upload or receive bank statements in PDF, image, or photo format. AI extraction models parse each page, identifying deposits, withdrawals, daily ending balances, and fees like NSFs. The system cross-references extracted totals against the statement's printed running balance to catch extraction errors. Advanced systems also flag anomalies such as perfectly round deposits, sudden revenue spikes, or formatting inconsistencies that suggest document tampering. The output is a structured data summary that an underwriter can review in a fraction of the time required for manual entry.

Can AI catch fabricated bank statements in MCA lending?

Yes, with caveats. AI fraud detection models analyze PDF metadata, font consistency, pixel patterns, and mathematical coherence of transaction histories to identify likely fabrications. Well-crafted fakes can still pass initial AI screening, which is why the most effective systems combine automated detection with human review of flagged files. The advantage of AI is consistency: it applies the same checks to every document without fatigue, reducing the odds that a fraudulent file slips through during high-volume periods.

Why is asynchronous document collection important for MCA lenders?

Asynchronous document collection allows merchants to upload required files, such as bank statements, government IDs, and void cheques, on their own time and from any device. This eliminates the back-and-forth email chains and phone calls that typically add three to five days to the funding timeline. For funders, async collection means documents arrive pre-organized and ready for AI extraction before a human underwriter even opens the file. Platforms like Let's Submit combine async upload links with AI parsing so that by the time a callback is scheduled, the application data is already populated and reviewed.

Conclusion

The convergence of record SMB confidence, platform lender dominance, and rising application volume makes one thing clear: AI underwriting for merchant cash advance is not optional for funders who want to scale. Manual workflows that served a 50-deal-per-week operation cannot survive a 200-deal surge without either ballooning costs or cutting review depth. The funders growing fastest in 2026 are the ones automating document collection, extraction, and fraud detection while keeping human judgment where it belongs, on the final funding decision.

Let's Submit brings this full workflow together. From Sabbie's AI-powered lead outreach to secure async document uploads and automated bank statement extraction, the platform ensures your team only touches deals that are ready to fund. Visit letssubmit.ca to see how it fits into your pipeline.