Day 4 of 14 · AI in Lending & Credit Control

Application & Affordability Analysis in Minutes

⏱ 6 min 📊 Beginner

Days 1–3 gave you the map, the toolkit, and the theory. Today is the payoff lesson — the workflow you'll use more than any other in this course: AI-assisted application review.

A typical consumer or small-business file means an application form, months of bank statements, payslips or accounts, and supporting documents. Read properly, that's a solid chunk of your morning — per file. Today you learn to turn that reading into a structured first-pass summary in minutes, so your time goes where it belongs: verifying and deciding.

Step one: summarize the pack

Every application review starts the same way: what am I actually looking at? Instead of building that picture by flipping between PDFs, give an AI (approved tool, anonymized data — Day 2 rules always apply) the pack's contents and ask for a structured snapshot:

"Summarize this loan application pack into: applicant profile, loan request and purpose, stated income vs documented income, existing commitments, and anything incomplete or inconsistent across documents."

That last clause is quietly the most valuable. AI is good at noticing that the stated salary doesn't match the payslips, that an address differs between two documents, or that one bank statement month is simply missing. Those cross-document checks are exactly what tired human eyes skip at 4pm on a Friday.

Step two: the affordability workup

Now the heart of the review: can this applicant afford this credit? From anonymized statement data, AI can build the workup fast if you tell it exactly what to look for:

Income stability — How regular are the credits? Same employer, same rhythm, same amounts? Is there secondary income, and does it persist across months or appear once?

Expense patterns — Fixed commitments (rent, utilities, insurance, existing repayments) versus discretionary spend. What's left over each month, and how much does that figure swing?

Debt stress markers — The quiet warnings: overdraft fees, returned payments, balances that hit zero before payday, minimum-only card payments, new credit taken mid-period, gambling patterns. Any one of them is a question; a cluster is a conversation.

The point isn't that AI sees things you couldn't. It's that AI reads every line of every month with equal attention and hands you an organized picture — while you'd still be on statement two.

Knowledge Check

Which of these is a debt stress marker to look for in statement data?

A

A single utility bill that increases in winter

B

Overdraft fees and balances that hit zero before payday

C

Fixed rent payments to the same landlord

D

A salary credited on the same date each month

Stable salary credits and consistent fixed commitments are signs of a healthy, predictable pattern. Overdraft fees and running out of money before payday signal that current commitments are already straining income — exactly the kind of marker an affordability review must surface and a human must weigh.

Step three: a prompt that makes every review consistent

The real professional win isn't speed — it's consistency. The same file reviewed by two analysts (or by you on a good day vs a bad one) should surface the same picture. A structured prompt is how you get that. Save this one:

"You are assisting a credit analyst. From the anonymized statement data below, produce a review with exactly these sections: 1) Income — sources, amounts, stability rating with evidence. 2) Fixed commitments — itemized monthly total. 3) Discretionary spending — pattern and monthly range. 4) Debt stress markers — each with the specific transaction evidence. 5) Affordability summary — net monthly surplus, best and worst month. 6) Questions a human reviewer should resolve before any decision. Flag every figure that should be verified against source documents."

Same sections, every file, every time. Your memos get easier to write, your reviews get easier to compare, and nothing gets forgotten because the template doesn't allow forgetting.

🤖

ChatGPT

online

Run an affordability workup on this synthetic 2-month extract. Applicant B, requesting a loan with repayments of 350/month: Month 1: salary +2,800; rent -950; utilities -180; car finance -310; card payment -140; groceries/discretionary -820; overdraft fee -25; balance at month end: 375 Month 2: salary +2,800; rent -950; utilities -195; car finance -310; card payment -140 (minimum only); groceries/discretionary -1,050; returned payment fee -30; balance at month end: 125

**Affordability workup — Applicant B (synthetic)** **1. Income:** 2,800/month, single source, stable across both months. ✅ **2. Fixed commitments:** ~1,590/month (rent 950 + utilities ~188 + car 310 + card ~140). **3. Discretionary:** 820 → 1,050 — rising, and the swing matters more than the average. **4. Stress markers:** overdraft fee (M1), returned payment fee (M2), card at minimum-only (M2), month-end balance falling 375 → 125. Individually minor; together they show a budget tightening month over month. **5. Surplus vs request:** nominal surplus ~360–390 before discretionary swing; the requested 350/month would consume nearly all of it. Worst month suggests the buffer is already gone. **6. For the human reviewer:** obtain a third month, verify every figure against source statements, ask about the returned payment, and confirm whether discretionary spend in M2 was exceptional. This is a first-pass summary of synthetic data — not a recommendation. The affordability judgment is yours.

↻ Replay conversation

Final Check

In this workflow, what remains the human analyst's job after AI produces the affordability workup?

A

Forwarding the AI summary directly to the applicant

B

Verifying every figure against source documents, resolving the open questions, and making the credit decision

C

Nothing — the workup is the decision

D

Formatting the AI output into the house template

AI's output is a first-pass summary — a well-organized draft of the evidence. The analyst verifies each number against the actual statements, chases the open questions (like the returned payment), and then makes the affordability judgment. The 40 minutes of reading becomes a 10-minute verification pass, but the verification and the decision are never skipped.

Verify like it's your signature — because it is

One discipline makes this whole workflow safe: every number gets checked against the source documents before it influences anything. AI occasionally misreads a figure, merges two transactions, or — importantly — sounds most confident exactly when it's wrong. Treat the workup as a brilliant junior analyst's draft: it structures your thinking and saves you an hour, and you'd never sign it unread.

That's the shape of the daily win. The reading compresses; the verification and judgment stay human-sized. Multiply that across every file you touch this week and you'll understand why this is the lesson people come back to.

Tomorrow: the same cross-checking skill, pointed at a darker problem — spotting fraud red flags at origination.

Affordability dashboard showing an income stability strip, expense category bars, debt stress markers, and a verify-against-source stamp

The workup organizes the evidence — income, expenses, stress markers — but nothing counts until it carries the verify-against-source stamp. That stamp is yours.

⚡

Day 4 Complete

"AI turns a 40-minute file review into a 10-minute verification pass."

Tomorrow — Day 5

Fraud Red Flags at Origination

Tomorrow you'll use AI as a second pair of eyes to cross-check applications for fraud patterns — income inflation, doc tampering, and synthetic identities.

Continue to Day 5 →