# Lending rules card Fill in one card for the first finance process you automate, most likely your lending pipeline. Fill in what you know; every blank line gets an owner and a date. Any blank line means: not ready to build. | Question | Answer | Owner / by when | | --- | --- | --- | | The process: where do we already have the volume, the history and the written rules? | | | | Cost per loan today, fully loaded (who measured it, over what period?) | | | | Volume: cases processed a year | | | | Baseline: turnaround today, and fraud and default rates | | | | The model: what it scores, and on which years of our own history it is trained | | | | The rules: where the written policy rules live, and who owns the list | | | | When a rule fires against a confident model score, what happens? | | | | Share of volume the system must handle, and why that number | | | | Experts: who gets questioned until their instincts come out as rules, by whom, until when | | | | Rules captured in software so far: how many, signed off by whom | | | | The flagged few: who works the cases the system sends to a person | | | | Reconciliation next: where does matching still cause trouble? | | | | Automation budget, priced against the fully loaded cost per loan | | | | Decision and date | ☐ BUILD ☐ NOT YET ☐ RE-CHECK ON ______ Date ______ Signed ______ | | ## Example (fictional): auto loans at a credit union Fictional example. | Question | Answer | Owner / by when | | --- | --- | --- | | The process: where do we already have the volume, the history and the written rules? | Used-auto loans at a regional credit union: high volume, 9 years of history, a written credit policy | n/a | | Cost per loan today, fully loaded (who measured it, over what period?) | **Blank: not ready** | Finance business partner, by 30 Nov | | Volume: cases processed a year | About 40,000 loans | n/a | | Baseline: turnaround today, and fraud and default rates | Turnaround 3 days; first-year default rate 2.8%, watched monthly | n/a | | The model: what it scores, and on which years of our own history it is trained | Likelihood of full repayment, trained on 9 years of our own disbursed loans | n/a | | The rules: where the written policy rules live, and who owns the list | The credit policy manual, moving into the decision system; owner: head of credit policy | n/a | | When a rule fires against a confident model score, what happens? | The rule wins; checked on 200 historical cases where both fired | n/a | | Share of volume the system must handle, and why that number | 70%, because below that the team still staffs the whole queue; the pilot measured 66% | n/a | | Experts: who gets questioned until their instincts come out as rules, by whom, until when | 4 senior underwriters, questioned weekly by the credit data lead, to 15 Dec | n/a | | Rules captured in software so far: how many, signed off by whom | 18 of an estimated 60 written into the system → **Fails the bar** (target: all 60 by 15 Dec) | Head of credit policy, by 15 Dec | | The flagged few: who works the cases the system sends to a person | A 3-person review desk; each case decided within one business day | n/a | | Reconciliation next: where does matching still cause trouble? | Supplier statements against the general ledger, about 300 unmatched lines a month | n/a | | Automation budget, priced against the fully loaded cost per loan | $480,000 for year one, build plus two data engineers; not signed until the cost per loan is measured | CFO, after 30 Nov | | Decision and date | ☐ BUILD ☒ NOT YET ☐ RE-CHECK ON 15 Jan · Date 20 Nov · "Rules shell a third done, cost per loan unmeasured." | Head of lending | From the course AI in banking and financial services, Dr. Tony Hoang · v1 · Oct 2026