The Same Skill, a Different Speed
Reading a merchant’s bank statements by hand means scanning for the same handful of signals every time: daily balance patterns, how often NSF events show up, and whether monthly deposits support the revenue a merchant claims on their application. That manual read is a real, learnable skill, and it is also slow, especially across a busy submission pipeline moving several merchants through underwriting at once.
Automated bank-statement-analysis software exists to do the extraction part of that work faster, not to replace the judgment call underneath it.
What the Software Delivers
Plaid’s underwriting-focused product, built around what it calls a Consumer Report, delivers up to 24 months of categorized transaction data to a lender, with real-time cash-flow insight available in as little as 10 seconds once a merchant connects their account. That data draws on a network of more than 12,000 connected financial institutions, and Plaid layers a proprietary score, LendScore, on top of the raw transaction data as a risk signal built specifically for lending decisions.
Plaid also reports up to an 80% conversion rate for account-linking inside lending workflows, a vendor-published figure worth reading as a claim about its own product’s adoption, not an independently audited industry benchmark.
Why the Underlying Signals Haven’t Changed
What underwriting reads for has not changed just because software does the reading faster. Negative or chronically low daily balances still trigger concern, frequent NSF events still signal cash-flow stress, and most funders still cap total funding somewhere between 10% and 25% of a merchant’s annual gross revenue, with a preferred debt-to-income ratio around 36% or lower. Automated software is extracting exactly those same numbers, just without a human manually scrolling through pages of a statement to find them.
That matters for how a broker should think about this tooling: it speeds up data extraction, it does not change what a good underwriting read is looking for.
What Automated Parsing Cannot Judge on Its Own
This is reasoning, not a cited statistic. A parsing tool can flag that a merchant had six NSF events last quarter. It cannot tell a broker, on its own, whether those six events reflect a business in real trouble or a seasonal dip that already reversed by the time a submission goes out. The extraction is automatic. The judgment about what the pattern means for a specific deal still belongs to whoever is reading the output.
A broker who treats a software-generated score as the final word, rather than a faster starting point for the same read they would otherwise do by hand, is skipping the part of the skill that matters.
Where This Fits Against the Manual Read
For a shop moving a handful of submissions a week, the manual skill of reading statements directly may still be fast enough, and it builds a broker’s own underwriting instincts in a way that outsourcing the read entirely does not. For a shop moving high submission volume across multiple closers, automated parsing turns a bottleneck task into a background process, freeing a closer’s time for the actual merchant conversation instead of statement review.
Neither approach is universally right. The volume a shop is moving is the honest variable that decides which one makes sense.
The Conversation Still Has to Happen Either Way
Whether statements get read by hand or parsed automatically, the numbers only tell a broker whether a merchant looks fundable on paper. They say nothing about whether that merchant will take the call, what they are hoping to use the capital for, or whether the timing is right.
Human + AI SDRs deliver that conversation directly, a double-confirmed merchant meeting on the calendar, so a shop’s underwriting tooling and its actual pipeline of live conversations are working on two different, complementary problems instead of one covering for a gap in the other.
Sources
The external data in this article draws on the sources below. Figures described in the text as estimates or industry triangulations are directional and are not attributed to a single dataset.
- Plaid, Consumer Report underwriting product
- MCashAdvance, MCA Underwriting: Process, Criteria, Red Flags