AI Credit Scorecard Template — Google Sheets + Notion
A working scorecard with AI-assist columns, weights, risk grades, and decision rules. Open the file, paste a borrower, get a grade. No model training required.
What's inside
- Google Sheets scorecard with 7 scoring dimensions, AI-assist columns pre-wired to a prompt, computed weights, sub-scores, and an overall A/B/C/D grade.
- Notion database version for teams who live in Notion (same logic, Notion formulas instead of Sheets formulas).
- A 5-minute walkthrough video (script included) + a 2-page setup doc.
- The source prompts the AI-assist columns use — so you can see, modify, or replace them.
- A decision-rules appendix: grade-to-action mapping, deviation thresholds, and when to override.
Prompts are the surface layer. Underneath sits the scorecard — the thing your credit committee actually signs. This is that scorecard, built to be read, tuned, and defended, with AI doing the qualitative heavy lifting in exactly the right places and staying out of the quantitative ones.
What the scorecard covers
Seven scoring dimensions, each a sub-score out of 10, each weighted:
- Identity & KYC integrity — is the applicant who they claim to be, and do the documents tell one consistent story?
- Employment & income stability — tenure, volatility, seasonality, sector risk.
- Affordability & leverage — DTI, FOIR, stress-tested EMI headroom.
- Credit history — bureau score, utilisation, past-defaults, enquiry pattern.
- Behavioural signals — application completeness, response speed, consistency across stages.
- Narrative risk — the qualitative one; AI-assist writes a one-paragraph risk narrative the underwriter can accept, edit, or reject.
- Collateral / co-applicant uplift — optional, applied only where relevant.
Weights are yours to change. The defaults are defensible starting points, not gospel.
Where the AI helps
Exactly two places. Cell Q4 asks the model to summarise employment stability from pasted payslip data. Cell U4 asks the model to generate a narrative-risk paragraph from the rest of the row. Everything else is arithmetic. That deliberate scoping is the difference between a scorecard your credit committee trusts and one they reject.
What you get in the Sheets file
One master sheet, one dashboard sheet, one decision-rules sheet, one “scoring log” for audit history, and one README tab that explains every column, weight, and threshold. Conditional formatting is already wired: A grade is green, D is red, cells needing AI-assist output are highlighted until filled.
The Notion version
Same scorecard, Notion formulas, so teams already operating in Notion don’t have to maintain two systems. Same prompts, same logic, same grade boundaries.
The honest caveat
This is a heuristic scorecard with AI-augmented qualitative fields. It is not a statistical scorecard trained on your default history. If you have 10,000 loans of history and a data scientist, you should build the latter. If you have fewer than that, this scorecard will beat spreadsheet-by-vibes underwriting every time, and it’s defensible in front of a credit committee or auditor who asks how the grade was derived.
Where it sits in the ladder
If you want the prompts that feed this scorecard, get the Prompt Library. If you want the scorecard plus the library plus the fraud playbook, that’s the Starter Kit and it’s the best-value package on the site.
This is for you if…
- Small lenders, P2P operators, or HNI investors who underwrite by spreadsheet today and want AI to carry the tedious parts.
- Founders who need something credibility-worthy to show a credit committee or investor without hiring a data science team.
- Teams who want a starting scorecard they can fork and tune, not a black-box model.
Skip this if…
- You have a full statistical scorecard built on historical default data. This is a heuristic + LLM-augmented scorecard, not a GLM.
- You need scoring at API latency for a large application volume. This is for human-in-the-loop workflows.
After buying this you can…
- Score a borrower from raw application data in under 10 minutes with a repeatable grade.
- Give your credit committee a scorecard document that's defensible, transparent, and not a mystery model.
- Onboard a new underwriter in a day because the scorecard is self-documenting.
Frequently asked questions
Is this a predictive model or a rules-based scorecard?
Rules-based, with LLM assistance on qualitative fields (employment stability, narrative risk, document consistency). It's deterministic where it needs to be and AI-augmented where AI adds real value.
Do I need to connect an API?
No. The AI-assist columns are prompts you paste into ChatGPT / Claude and paste back the result. You can upgrade to an Apps Script integration — instructions included — if you want automation, but it's optional.
Will this work for my loan product (personal, business, home, MSME)?
The base scorecard is for unsecured personal and small-business loans. Home and MSME need 2–3 additional dimensions which are noted in the appendix. If you're underwriting very specialised products, this will be a starting point, not the finished article.
Can I tune the weights?
Yes — that's the point. Every weight, every threshold, every grade boundary is a single cell you change. The sheet is structured to survive you tweaking it.
Refunds?
14-day, same as everything else on this site.
Ready?
One-time payment. Instant delivery. 14-day refund if it doesn't deliver what this page promises.
Buy AI Credit Scorecard Template — Google Sheets + Notion · $49 →