Underwriting is the most misunderstood job in lending. From the outside it looks like a checklist. Minimum revenue, maximum positions, credit score above a threshold, time in business above a floor. That's not underwriting. That's screening. Screening is what happens before the underwriter touches the file. It's the filter that decides whether a human should spend time on this deal at all.
Underwriting is what happens after the filter passes.
The deal cleared the automated checks. The revenue is in range. The credit isn't disqualifying. The business has been operating long enough. Now someone has to decide whether this specific business can service this specific debt alongside everything else they're already carrying. That decision is not a formula. It's a read.
The underwriter opens the bank statements and starts building a picture. Not by reading transactions top to bottom. By scanning for shapes. Revenue consistency across months. Deposit cadence. Whether Tuesday and Friday have heavier inflows than other days. Whether December drops off or whether the dip in March was a one-time event. They're constructing a sense of the business's rhythm from twelve months of raw data.
Average daily balance tells you how much cash the business actually holds between deposits and withdrawals. The range tells you how volatile that is. Negative days tell you whether the business is operating at the edge. NSF count tells you whether payments are bouncing. But NSF amount tells you something different. A handful of $34 bank fees is noise. A string of returned payments in the thousands means the business can't cover its obligations. The distinction changes the entire risk picture.
Then the debt picture. How many active positions. What the daily or weekly pull totals relative to average revenue. The debt service coverage ratio. Whether there's room for another payment on top of what's already going out. Whether any positions recently renewed, which might mean the borrower is cycling to stay liquid. Whether the payment schedules leave enough in the account to actually operate the business between pulls. The stacking analysis is where deals die or get priced, and it depends entirely on classifying every position correctly. Miss one and the ratio is wrong.
Then the soft signals. How the owner moves money. Regular owner draws that suggest the business is the owner's primary income. Large round-number transfers between accounts. Payment processor deposits from Stripe or Square that tell you the revenue is card-based. The industry itself. A restaurant with seasonal foot traffic prices differently than a trucking company with contract revenue. Credit history. Not just the score. The tradeline depth. Whether there are derogatories. Whether the borrower has been through this before and paid it back.
The underwriter is synthesizing all of this into one question. Can this business support this payment. Not in the average month. In the worst month visible in the statements. With the existing positions still pulling. With the owner still drawing. With the seasonality still cycling.
Two experienced underwriters can look at the same deal and price it differently. Both can be right. One sees the revenue trend stabilizing after a dip and prices for the recovery. The other sees three active positions with overlapping pull schedules and prices for the stack risk. The numbers are the same. The judgment is different. This is not a flaw in the process. This is the process. Underwriting is the application of experience to ambiguous data under time pressure.
The checklist people think the job can be automated because they think the job is the checklist. The job is everything that happens after the checklist. It's the part that requires having seen a thousand deals and knowing what this one feels like compared to the ones that paid back and the ones that didn't.
The technology that actually helps underwriters is not the technology that tries to replace this judgment. It's the technology that gives them the full picture faster so they can spend their time on the read instead of on the assembly. The parsing and the classifying and the stacking and the scheduling and the reconciling. That's bandwidth work. The underwriter was always doing it to get to the part that matters. The decision.
The best underwriters have always known what to look for. The constraint was never their judgment. It was how long it took to get the information into a shape where the judgment could be applied.