Skip to content

Compress a 15-page SME credit memo into a 3-bullet summary

← Use cases Risk

Compress a 15-page SME credit memo into a 3-bullet summary

A relationship manager hands you a fifteen-page credit memo and the committee meets in an hour. Claude reads it and produces a three-bullet exec summary plus a risk-flag list — the structured starting point your senior credit officer actually wants, not a paraphrase of the whole memo. Audience: beginner · ~20 min · 7 steps

You'll end up with

A one-page markdown brief — 3-bullet exec summary, top 5 risk flags with severity (HIGH/MED/LOW), and an "open questions for the RM" list.

Before you start, you need
  • Nothing — Claude will invent a realistic 15-page synthetic credit memo for a fictional Greek SME. (Once you trust the loop, you can swap in a real memo your line manager has cleared.)
  • Claude Code installed and a terminal open (see Day 1)

Credit committees run on density. The deck wants a one-pager, the memo from the relationship manager is fifteen. Junior risk analysts spend a Tuesday morning condensing it; senior officers re-read the whole thing because they don’t trust the condensed version.

This use case is the version where the condensed version is trustworthy because the itself names what “trustworthy” means.

Compliance check before you start. Credit memos contain client-confidential financials, security details, and sometimes director-level information. Confirm the classification with your line manager before putting a real memo into a Claude-readable folder. For your first run, use a sanitised or training-deck memo — same shape, no real client data. The point is to learn the loop; you can graduate to real memos once you trust the output.

Step 01

Build the workspace

Open the .

Press ⌘+Space, type “Terminal”, and press Enter.

Open the Start menu (press the Windows key), type “Ubuntu”, and press Enter. If you don’t see Ubuntu listed, install WSL first.

In Ubuntu, ~/Desktop is a folder inside ’s Linux home (/home/<your-Linux-username>/Desktop) — not the Windows desktop you see in File Explorer at C:\Users\...\Desktop. That’s fine: the files are real and Claude can read and write them. Anywhere this use case says “open in Finder / File Explorer”, run explorer.exe . from your Ubuntu terminal — Windows opens that exact WSL folder in Explorer.

Then:

mkdir -p ~/Desktop/claude-lab/credit-memo-review
cd ~/Desktop/claude-lab/credit-memo-review
claude --dangerously-skip-permissions
  • mkdir -p ~/Desktop/claude-lab/credit-memo-review — make a fresh folder inside claude-lab on your Desktop (-p creates claude-lab too if it’s not there yet — it’s the one home for all hub use cases).
  • cd ~/Desktop/claude-lab/credit-memo-review — move into it.
  • claude --dangerously-skip-permissions — start here. The flag stops Claude you for permission on every file write — safe in a fresh, dedicated folder like this one. (If you’d rather see every prompt for your first run, just type claude — same thing, more interruptions.)
Step 02

Ask Claude to invent a realistic credit memo

You don’t have a real memo to hand and you don’t need one. Tell Claude:

Create a file called source.md in this folder. Write a realistic synthetic credit memo for a fictional Greek SME — long enough to feel like a real fifteen-pager, but what matters is the structure and the planted tensions below, not the word count. Use page-break markers like <!-- page 2 --> every few hundred words so we can cite “page X” later.

Borrower: Aegean Solar A.E. — a 12-year-old Greek SME installing residential and commercial solar PV systems, based in Thessaloniki, ~85 employees, annual revenue ~€18M.

The ask: €4M, 5-year amortising loan, secured against installation contracts and a second-ranking charge over the warehouse.

Structure with numbered sections:

  • §1 Executive summary (RM’s view: recommend approve)
  • §2 Borrower profile, history, management team
  • §3 Industry context — Greek solar market, regulatory tailwinds
  • §4 Financial analysis — 3 years of P&L, balance sheet, cash flow, with realistic numbers. Include EBITDA margin trending from 11% (2023) → 9% (2024) → 7% (2025) — a real concern.
  • §5 Customer concentration — single customer “Hellenic Public Properties” is 32% of 2025 revenue.
  • §6 Security — list collateral with realistic LTV math.
  • §7 Covenants — proposed: net debt / EBITDA < 3.0x, minimum DSCR 1.25x, no dividends if covenants breached.
  • §8 Risk factors — RM lists 4 risks but downplays them.
  • §9 Recommendation.

Make sure the memo includes these tensions a senior risk officer would catch: (a) declining EBITDA margin, (b) the customer concentration, (c) covenant headroom under the new loan is tight if 2026 EBITDA doesn’t recover, (d) the RM’s projections assume 18% revenue growth in 2026 — well above industry consensus of 8–10%, but the memo doesn’t justify why.

Tone: confident RM advocacy. Use specific numbers, not ranges. Add a header on page 1 with borrower / amount / RM name / date.

Claude writes the file straight away.

Renaming the file to source.md keeps the in Step 4 identical for the next memo. Change one file, reuse the prompt.

Step 03

Let Claude create the context file

Risk summaries are worthless without judgement about what kind of risk matters to whom. Tell Claude:

Create a file called context.md in this folder. Put these 5 lines inside it:

Reader: senior credit officer at NBG
Portfolio: Greek SME secured + unsecured lending
Reader's seniority: knows the sector, doesn't want the memo paraphrased — wants a triage
Goal: a one-page brief before the credit committee in an hour
What counts as a "risk flag": concentration > 20% of revenue, customer churn rising, covenant headroom < 15%, refinancing risk inside 18 months, declining gross margin > 200bp YoY, owner-manager succession unclear

Claude writes the file straight away.

The “what counts as a risk flag” line is the load-bearing one. Without it Claude flags everything (“interest rates may rise”) and useful signal drowns. Spend two minutes tailoring this line to what your committee actually argues about.

Step 04

Ask Claude for the brief

Send this to Claude:

Read source.md and context.md.

Produce brief.md with these four sections, exactly in this order:

1. Exec summary — three bullets. Lead with the headline (size + sector + ask). Bullet two: the single strongest reason to approve. Bullet three: the single strongest reason to push back. Nothing else.

2. Risk flags — table with columns: Flag | Severity (HIGH/MED/LOW) | Evidence (page + quote) | What this means. Only use the flag categories defined in context.md — don’t invent new ones. Order by severity, highest first.

3. Numbers I would double-check — bullets. Any figure in the memo where the source (audited financials / management accounts / projections) is unclear or the year-over-year delta is suspicious. Cite the page.

4. Open questions for the RM — bullets. The things you’d ask the relationship manager before voting. Phrased as actual questions, not statements.

Hard rule: every claim must cite a page or section of source.md. If you can’t cite it, don’t write it.

Press Enter. A 15-page memo takes 60–120 seconds.

Step 05

Verify three citations before you trust the brief

Ask Claude to show you the brief:

Show me brief.md.

Pick three risk flags or numbers at random. For each one:

  1. Ask Claude to show you the cited page: “Show me the <!-- page 7 --> block of source.”.
  2. Confirm the evidence quote actually appears there.

If any citation doesn’t match, tell Claude:

Risk flag 2’s evidence quote doesn’t appear on page 7 of source.md. Re-extract the verbatim text, or downgrade the severity if you can’t substantiate it.

Iterate until all three spot-checks pass. The brief is now a starting point your senior officer can trust enough to argue with.

Then make Claude audit every citation, not just your three

Your random checks calibrate trust; the full sweep is Claude’s job. Paste:

Audit your own brief. For every row of the risk-flags table and every “numbers” bullet: locate the cited page block in source.md, verify the evidence quote appears there verbatim (use grep -F and show me the output), and mark the row ✓ or ✗. Every ✗ gets fixed — re-extract the quote, correct the page, or downgrade the severity — then re-run the audit until the table is all ✓.

A risk flag that can’t produce its evidence quote on demand isn’t a flag, it’s a hunch. This audit is what makes the difference between “the AI said so” and “page 7, second paragraph, here’s the line” when the committee pushes back.

Step 06

Add a one-line recommendation and ship

The brief is structured triage, not a recommendation. The recommendation is your call — but Claude can shape it once you’ve decided.

Tell Claude:

Based on brief.md and source.md, draft three one-line recommendations:

  1. Approve, with conditions
  2. Decline
  3. Defer pending [specific information]

Each one in the form: “Recommendation: [verdict]. Conditions/reasons: [one sentence].”

Pick the one that matches your read. Paste it at the top of brief.md.

In a real Tuesday morning you’d send brief.md to your senior officer — we’re pretending here, so the file on your Desktop is the deliverable.

You’ve turned a Tuesday morning of condensing into twenty minutes of triage. The senior officer (in your head, for now) reads three bullets, scans the risk flags, asks you the open questions — that’s exactly the conversation you wanted.

Lock the risk-flag rules in CLAUDE.md

The most valuable file you produced today is context.md — specifically the “what counts as a risk flag” line. That definition shouldn’t be re-typed every memo. Rename it:

mv context.md CLAUDE.md

CLAUDE.md is the magic filename reads automatically every time you start claude in a folder containing it. Next memo:

  1. New folder, drop in source.md
  2. cp ~/Desktop/claude-lab/credit-memo-review/CLAUDE.md . — your risk-flag thresholds, reader profile, and brief structure all travel with you
  3. claude --dangerously-skip-permissions and one line: “produce the brief from source.”

Claude already knows what counts as HIGH/MED/LOW severity from CLAUDE.md. The brief lands consistent across memos — which means your senior credit officer learns to trust the format, and you stop arguing about whether something is “really” a flag.

Step 07

Level up — a committee card with traffic lights

Optional: the same brief as a single screen the committee can absorb in ten seconds. One :

Read brief.md. Build a single self-contained file committee-card.html that I can open by double-clicking — no server, no internet, no external libraries. One screen: borrower, ask, and my recommendation across the top; the three exec-summary bullets; then the risk flags as traffic-light cards — red for HIGH, amber for MED, green for LOW — where clicking a card reveals the evidence quote and page citation. Open questions for the RM as a checklist at the bottom so I can tick them off live in the meeting. Include a print stylesheet so it prints cleanly on one A4 page. Accent colour #007a8a.

Screen-share it at committee, click a red card when challenged, and the verbatim evidence is right there with its page number. The checklist at the bottom turns “did we ask the RM about the projections?” into a box that’s either ticked or not.

The pattern to remember — any output Claude can produce as text, it can also produce as a small interactive page. The upgrade prompt is always the same shape: “turn this into a single self-contained interactive HTML file I can open by double-clicking.” It works on every use case in this hub.

Next use case →

Turn a business question into a SQL query you can actually run

A manager asks "how many cards did we to under-25s last quarter, split by ?" You don't write SQL. You describe the tables to Claude in plain English, ask for the query, run it against the data warehouse. Twenty minutes from question to chart.

Open it → Back to all 15 use cases