Commercial loan portfolio analysis that reads every credit file, and shows you the page behind every row
A board or examiner question answered in an hour, with proof pages.
See one case, screen by screen ↓Software that lets a bank look across its whole loan book — exposure by industry, rating and maturity, concentrations and stress. Commercial Loan Portfolio Analysis adds agents that read the credit files themselves, so it can answer questions the loan system’s data cannot, with the page that proves each row.
How to choose AI solutions for lending →
Updated
The board risk committee asks something nobody planned for: which of our commercial borrowers depend on imported inputs, how much do we have out to them, and who would fall below covenant if those costs rise? The loan system knows the industry code, the commitment, the rating and the maturity. It does not know that a borrower buys nearly half its cost of goods sold from two mills overseas.
That answer sits in the credit memos, the annual reviews, the notes to the financial statements, the supplier lists and the RM call memos. So the portfolio team screens by NAICS code, sends a spreadsheet to the relationship managers, and re-reads files by hand for weeks — and still misses the solar installer or the IT reseller whose code never looked exposed.
Estimated split for one new question across the whole commercial book, by hand and with the solution. In the working solution the read, the totals and the stress took 38 minutes, and the Chief Credit Officer approved the summary within the hour.
Six specialist agents read every credit file against the question and return a table, each row with its proof; the Chief Credit Officer approves what goes to the board.
Monday 12 October at Ashgrove Commerce Bank. The board risk committee meets Thursday, and its chair has asked about tariffs. Here is the morning, screen by screen, in the working solution.
Owen opens Ask the book: 1,184 relationships, 4,912 credit files, book as of 30 Sep 2026. Under the box is the last saved answer — 41 borrowers that are themselves lenders, $512.0M committed, 14 of them in ordinary C&I codes that the call report would miss.
“New on Sunday’s re-run: Marlowe Equipment Finance, now funding 40 dealers.”
Owen types it in plain words: hotel loans over $5M whose guarantor has liquidity under 10% of the loan. The agents read 29 personal financial statements and 34 guaranty agreements in 3 min 20 s. Eleven loans come back — $132.1M committed, $124.0M outstanding — each a red dot on a chart of all 34 hotel loans over $5M.
“Liquidity” = cash and marketable securities on the guarantor’s latest PFS; retirement accounts and real estate left out, as guaranty policy §6.2 says.
The proof chip on Riverside Lodging Partners opens Daniel Voss’s personal financial statement of 31 Mar 2026 at page 2, with cash and marketable securities highlighted and retirement accounts marked “no · policy §6.2”. The division is written out under it.
“Liquid assets $620K ÷ commitment $14.8M = 4.2% — under the 10% the question set.”
Compare with 30 Jun shows 8 loans then and 11 now: Old Mill and Hocking Hills because liquidity fell, Short North because it is a new loan. Owen saves the question to re-run each quarter — next 2 Jan 2027 — and whenever a guarantor files a new statement.
Six agents read all 4,912 files and finish in 38 minutes. 63 borrowers depend on imported inputs — imports at least 20% of cost of goods sold, or one named foreign supplier — with $611.4M committed, 9.5% of the book. 57 came through the industry-code screen; 6 did not. 351 relationships in trade-sensitive codes were cleared with the passage that shows domestic sourcing.
“Not in file for 4 borrowers … The files do not say where their inputs come from, so they are not counted; their RMs have been asked.”
The evidence table lists all 63 with RM, committed, outstanding, rating, imported share of costs, countries, coverage now and stressed, maturity and how it was found. Filters cut it to the 17 below 1.20x, the 9 maturing within 12 months, the 6 outside the codes and the 4 not in file.
The annual review of 14 Mar 2026, p. 6, says about 46% of cost of goods sold is imported carbon steel wire rod from two mills in Vietnam and Türkiye; the July field exam puts those two mills at 31.2% and 14.9% of purchases. On the FY2025 spread, a 15% rise with half passed through takes debt service coverage from 1.38x to 1.06x, below the 1.20x covenant. The $12.0M revolver matures 31 Mar 2027.
The rise applies only to the imported share the files state; ratings are not changed by a stress.
At +15%, half passed through, 17 of the 63 fall below 1.20x with $142.6M outstanding, 5 of them also maturing within 12 months. Owen can move the input-cost rise, the share passed through and the coverage line, or pick the board, severe or mild case, and add the case to the board summary.
The answer writer drafts the summary for Thursday’s committee: the question, the answer by industry, the stress, what the bank is doing and how it knows — every number footnoted to its source. Owen sends it to Ruth at 09:24. She approves at 09:31 and it goes into the committee pack.
“Good — put the five maturing names first on Thursday.”
Saved questions re-run whenever a new memo, review or supplier list arrives, and every Sunday at 06:00. This week’s new hits came from 14 files added since Sunday: a going-concern paragraph at Fenwick Hospitality Group, one retail customer now 38% of Kinsale Logistics’ revenue.
The same board question, seen by the four people who carry it — what it used to take, and what it takes now.
Six specialist agents read every credit file against the question and return a table, each row with its proof; the Chief Credit Officer approves what goes to the board.
Turns a plain-words question about the book into a search plan: what counts, which files, which part of the book and which numbers — for imports, at least 20% of cost of goods sold or one named foreign supplier, unless Settings say otherwise.
Reads the loan and exposure data — codes, commitments, outstanding, ratings, maturities — ties relationship totals to the book and lists first candidates, then passes every relationship to the file reader.
Reads memos, reviews, statement notes, supplier lists, field exams and call memos for the passage that answers the question, quoted with document, date and page. Prefers audited figures over memos, and newer over older.
Totals committed and outstanding by industry, rating and maturity, and ties them to the book.
Applies a shock to each borrower’s latest audited spread — only to the share the files support, net of the pass-through — and reports coverage before and after against the covenant line.
Writes the cited answer and the board or examiner summary: leads with the number, names what the codes missed and what the files cannot answer, and footnotes every figure.
Approves what goes to the board; the portfolio manager owns the answer. The agents propose; a named person decides.
Ask about any borrower, industry or stress in the answer — or ask for a draft. Every answer cites the file behind it, then opens the screen it is about.
Why is Kessling Fastener on the list?
Its annual review of 14 March says about 46% of cost of goods sold is imported carbon steel wire rod, from two mills in Vietnam and Türkiye on 90-day orders. The July field exam agrees: the two mills are 31.2% and 14.9% of purchases. On FY2025 figures — EBITDA $4.0M, debt service $2.9M — a 15% rise with half passed on takes coverage from 1.38x to 1.06x, below the 1.20x covenant. Its $12.0M revolver matures 31 Mar 2027.
Which borrowers did the industry codes miss?
Six borrowers depend on imports but sit outside trade-sensitive codes — $39.6M committed. Northfield Solar Installers is the one below the line: 1.44x to 1.01x. I have filtered the evidence to these six.
What if costs rise 20% and only a quarter is passed on?
Then 44 of the 63 borrowers fall below 1.20x, with $308.6M outstanding — against 17 and $142.6M in the board case. It is the same method on each borrower’s FY2025 spread, with the rise on the imported share the files state.
Draft the board summary for Thursday
Drafted for Ruth Adebayo to approve: the question, 63 borrowers and $611.4M committed, the 17 below 1.20x at +15% ($142.6M), the 5 that also mature within a year, what the RMs are doing, and the 4 borrowers the files cannot answer. Every number carries a footnote to its source. Nothing goes to the board until Ruth approves.
13 screens from the working solution, on its sample data. Pick one to see it large.
What counts, what is searched and who sees what are settings — versioned, and on the audit trail.
| Setting | Default | Choose from |
|---|---|---|
| Imported share of cost of goods sold | 20% | Up or down in steps of 5% · one named foreign supplier also counts on its own (on · off) |
| Matures soon | 12 months | 6 months · 12 months · 18 months |
| Input cost rise | +15% | +10% · +15% · +25% |
| Share passed through | 50% | None · 25% · 50% · replaced by the borrower’s own figure where the file states one |
| Coverage line | 1.20x | 1.15x · 1.20x · 1.25x |
| Oldest annual review accepted | 15 months | 12 months · 15 months · 24 months · older files are used but flagged |
| Board and examiner summaries approved by | Ruth Adebayo, Chief Credit Officer | Ruth Adebayo · Ruth Adebayo and the CRO |
| Sweeps re-run | New files + Sunday 06:00 | New files + Sunday 06:00 · new files only · daily 06:00 |
Every figure is labelled: a target the solution is built to, an estimate, a typical published result, or a proven one.
“demo” = seen in the working solution, on its sample bank data · “target” = the design goal, measured in the live solution · “estimated” = our estimate, from the working solution’s sample case · Context: Federal Reserve SLOOS, October 2025 (banks less likely to approve C&I loans to firms with high trade exposure); Fed H.8 (C&I loans ≈ $2.8T); Interagency Guidance on Credit Risk Review Systems (SR 20-13). People, banks and borrowers named on this page are fictional — characters and sample data in the working solution.
Software that lets a bank look across its whole loan book — exposure by industry, rating and maturity, concentrations and stress. Commercial Loan Portfolio Analysis adds agents that read the credit files themselves, so it can answer questions the loan system’s data cannot, with the page that proves each row.
An industry code does not show that a solar installer buys 52% of its job cost from Vietnam and Malaysia. In the working solution, 6 of the 63 import-dependent borrowers sat outside trade industry codes, and 351 code-screened relationships were cleared with the passage that shows domestic sourcing. The screener never clears a borrower on its code alone.
Credit memos and annual reviews, financial statements and their notes, customer and supplier lists, borrowing-base certificates and field exams, RM call memos, and guarantors’ personal financial statements — alongside the loan and exposure data. Which sources are searched is a setting.
The borrower is marked “not in file”, left out of the count and named on the answer, so its RM can be asked. On the board’s import question that was 4 borrowers. Annual reviews older than 15 months are still read, and flagged next to the borrower.
Yes. The stress tester applies an input-cost rise to each borrower’s latest spread, only on the share the files support and net of what is passed through, and shows debt service coverage before and after against your covenant line. It never changes a risk rating.
Yes. Save it as a sweep and it re-runs whenever a new memo, review, statement or supplier list arrives, and on a schedule — Sunday 06:00 in the working solution — with new hits marked.
Yes. The portfolio manager owns the answer, relationship managers confirm with borrowers, and the Chief Credit Officer approves every board and examiner summary before it leaves the credit team. Agents never send to the board, contact a borrower or change a rating, and answers show only the relationships the asker may see.
The Agentic Solution Engine builds and deploys it from your requirements — your credit policy, your loan system extract and a sample of your credit files — and it goes live once every quality gate has passed. We will walk you through it on your own book first.
Every related entity spread in hours, each number linked to its page.
1-min demoCredit Memo AutomationA cited first draft the same day, not after days of pasting.
1-min demoCommercial Loan OriginationFiles ready for underwriting in days, not weeks of chasing documents.
1-min demoWe’ll run Commercial Loan Portfolio Analysis on a sample of your own credit files and a question your board has asked.