The Loan-Committee Packet That Drafts Itself

Every community bank in Nebraska has a version of the same Friday afternoon. A credit analyst is pulling together the loan committee packet for Monday’s meeting. The meeting covers eight new credit requests and three renewals. The analyst has been at this since 1 p.m. She’s working from the credit file, the core system, the borrower’s financial statements, the appraisal, and whatever she can find in the shared drive. The packet is due to the loan committee chair by 4 p.m.

The packet isn’t hard to assemble. It’s just slow. Every piece of information is somewhere — in the core, in the file, in the title commitment, in last year’s tax returns. Getting it from “somewhere” to a formatted summary document is mostly a copying-and-organizing job. A skilled credit analyst doing a copying-and-organizing job for three hours every Friday is one of the more expensive ways a community bank spends its Friday afternoons.

The Friday-afternoon packet build

The standard packet format varies by institution, but the core content is consistent: borrower summary, credit history, financial ratios, collateral description, proposed terms, repayment analysis, and a risk assessment section where the analyst flags anything worth the committee’s attention.

For a new commercial real estate credit, assembling that packet involves pulling the last three years of business financials from the file, calculating or verifying the key ratios (debt service coverage, loan-to-value, global cash flow), pulling the credit history from the core, summarizing the appraisal findings, and writing the risk narrative. For a straightforward deal with clean financials and a single property, it takes 90 minutes to two hours. For a complex deal — multiple entities, guarantors with their own financials, a partially completed development — it can run past four hours.

That’s one deal. The packet might have eleven.

What the AI is actually doing (extraction + risk-flag pre-marking, not approval)

This is important enough to say plainly: the AI is not making a credit decision. It is not approving loans, recommending rates, or assessing creditworthiness. What it is doing is faster than it sounds and more useful than it looks.

The pipeline pulls the documents associated with a credit request — the financials, the appraisal, the core export, the existing credit file — and extracts the structured data: revenues, expenses, EBITDA, existing debt obligations, collateral values, the borrower’s credit history with the institution. It calculates the ratios the committee always looks at: debt service coverage, current ratio, loan-to-value against the appraisal.

It pre-marks flags. Not decisions — flags. If the trailing twelve-month DSCR is below 1.10, that’s a flag. If the LTV exceeds policy threshold, that’s a flag. If there’s a year-over-year revenue decline that changes the repayment analysis from the prior renewal, that’s a flag. The committee officer decides what to do with each flag. The analyst sees the flags before the committee does and can add context or override a mechanical flag that doesn’t apply in context.

The output is a formatted draft that the analyst reviews. Not a form. A readable summary in the packet format the committee actually uses.

Why this is a perfect fit for the BSA-conscious culture of community banks

Community banks take compliance seriously in a way that larger institutions sometimes don’t, partly because the regulatory consequences of a compliance failure land harder on a $300M bank than on a $30B one, and partly because the culture of community banking — knowing your borrowers, knowing your market, being personally accountable for decisions — is already aligned with a human-review model.

That culture makes a human-review-required AI pipeline easier to adopt, not harder. The loan committee isn’t being asked to accept AI recommendations. They’re being asked to work from a better-prepared packet. The analyst who assembled the packet manually is still accountable for it. The change is that she spent 30 minutes reviewing and correcting the AI draft instead of 90 minutes building it from scratch.

Community banks also have strong documentation habits. The credit file exists, the core exports cleanly, the appraisal is in the file, the financial statements have been collected as part of underwriting. All the inputs the pipeline needs are already there — they’re just scattered. Bringing them together is exactly the work the pipeline does well.

Where compliance still has to live

The AI is not in the credit decision. Full stop.

The risk narrative in the packet is drafted by the pipeline, but it must be reviewed and signed off by a human officer before it goes to committee. The flags the pipeline pre-marks are starting points for analysis, not conclusions. The credit decision — approval, denial, modification, terms — is made by the loan committee on the basis of their review of the packet, not on the basis of anything the AI recommended.

Audit trails matter. Every draft the pipeline produces should carry a timestamp and a version log so examiners can see what the analyst received, what she changed, and when she approved the final packet. That’s not compliance theater — it’s the kind of documentation that protects the institution and the officer when an examiner asks how a particular decision was made.

BSA-specific: nothing in this pipeline touches transaction monitoring, suspicious activity review, or 314(a) matching. Those workflows have their own compliance requirements and their own human-review standards. Loan packet assembly is a credit workflow, not a compliance workflow, and it should stay in that lane.

What it looks like in practice

A community bank in central Nebraska — the kind of bank that has been in the same market for 60 years, knows its agricultural borrowers by name, and has a loan committee that meets every other Monday — runs a 90-day build starting in early fall.

The first two weeks are interviews with the credit team and the committee chair. The pipeline is built against the bank’s actual core export format and the specific packet template the committee has used for years. By week six, it’s handling the packet build for every new commercial credit request and renewal.

By the end of the 90 days, the Friday-afternoon packet build takes the credit analyst about 45 minutes instead of three hours for a typical deal. She’s reviewing, correcting, and adding judgment — not pulling numbers from PDFs. The loan committee chair notices that the packets are more consistent. The flagging is more systematic. He still reads every one carefully. He just reads them faster because the structure is cleaner.

The analyst, for her part, leaves by 3 p.m. on Fridays for the first time in two years.

See how Blue Sage approaches the broader range of credit and compliance workflows in community banking.