Model scores, rules decide. Train a predictive model on your own lending history, then wrap it in written policy rules that override any score it gives.
Write your experts' instincts down as rules. Ask the same question enough times and experienced decision-makers can state their intuitions as explicit rules.
Let the model take the volume. One lender cut turnaround from 24 hours to nine seconds, with the system handling roughly 75% of volume.
Automate reconciliation next. Matching line items between receipts and purchase orders is a messy problem for finance and accounting teams in large companies.
Know your cost per loan. Price it fully loaded before you set the automation budget.
A free lesson from AI in banking and financial services, one of its 9 lessons. Read another free lesson
What a mortgage costs
Depository lenders averaged a fully loaded $16,320 to originate each retail loan in 2025, on MBA's benchmark. In finance, start where you already have the volume, the history and the written rules: the lending pipeline. That is where AI pays first.
Inside one lender's pipeline
A consultancy that builds AI systems describes one client, a lender processing about $30 billion a year. The lender combined a model trained on its historical trades with a shell of human-written policy rules. Turnaround fell from 24 hours to nine seconds, and fraud and defaults dropped. You can copy the shape in three steps.
The model, trained on your historical trades, estimates how likely each case is to be repaid.
Rules your people wrote override it. When a policy rule fires, it wins, whatever the model's confidence.
As of early 2025, the system handled roughly 75% of the volume.
Where a rule belongs, inside the model or around it, is settled in an earlier course; so is the advice to try cheap logic before any model. Lesson 2 takes the gate between the system and the person: how wide it opens. Lesson 4 takes what a decline must explain.
The rules were already in the building
Experienced decision-makers often cannot explain their calls at first. With repeated questioning, they can articulate their intuitions as rules. One such rule: refuse a client's invoices for 60 days after a certain client behavior. Capture those rules explicitly in software, where the pipeline can enforce them. That is where the shell around the model comes from.
Reconciliation comes next
Once the lending pipeline runs, the next finance target is reconciliation: lining up the line items in receipts against the matching lines in purchase orders. It is a messy problem for finance and accounting teams in large companies, and a strong target for high-accuracy AI automation. An earlier lesson maps which back-office processes agents take first. Ask your finance team where matching still causes trouble. The split holds beyond lending too: predictive models on all the volume, language models on the flagged few.
The size of the prize
A 2025 estimate put mortgage originations at about 4 trillion dollars a year, with most of the cost of making one being people moving digital information through processing, underwriting and lender matching. Price your own cost per loan, fully loaded, before you set the automation budget.
Exercise (20 minutes): fill in the lending rules card
Pick the one finance process where you already have the volume, the history and the written rules; for most lenders that is the lending pipeline. Fill in the card for that process. You won't know every line yourself; some sit with the process owner and some with finance. Every line you can't fill gets an owner and a date before you close the card. The card download includes a filled-in example of a pipeline that isn't ready yet, and why.
Lending rules card (PDF)
A blank card and a filled-in example, one page each.
Show the three prompts: fill the card, test it, turn it into a plan
Use AI to fill in and test your card
These three prompts go into your company's approved AI assistant. Use the first to fill in the card one row at a time; it won't guess a number for you. Use the second when the card is done, to hear the case against building yet. The third asks about your industry and your process and turns the lesson into a 30-day plan, with a checklist, rubric or brief for your boss if you want one. Use placeholders for company, system and people names; the help doesn't depend on them.
Prompt 1: Fill in my card with me
Paste this, then add your process in one or two sentences and the card below.
Prompt
Before you start: use your company's approved AI assistant, not a personal account. Replace company, system, supplier and people names with placeholders (Company A, System X) and round any figures. The help you get does not depend on them.
You are helping me fill in a lending rules card for one finance process. The card is pasted at the end.
Rules:
- Ask me about one row at a time, in order, and wait for my answer before moving on.
- Write down only what I tell you. Never suggest, estimate or invent a number, a name, a yes/no answer or a decision for me, even if I ask you to guess. If I ask, say that this row needs a real answer and ask who could give it.
- If I don't know an answer, write "Not known" and ask who should find out and by when. Put that in the Owner / by when column.
- After the last row, show the whole card as a table with three columns: Question, Answer, Owner / by when. Then list every row marked "Not known" with its owner, and remind me that any unanswered row means this process is not ready to build.
My process in one or two sentences: [DESCRIBE YOUR PROCESS]
The card:
[PASTE THE CARD]
Prompt from The AI Decision Course by Dr. Tony Hoang, host of The Artificial Intelligence Podcast (Spotify, Apple Podcasts): tonyphoang.com/learn
Prompt 2: Stress-test my card
Paste this, then your filled card.
Prompt
Before you start: use your company's approved AI assistant, not a personal account. Replace company, system, supplier and people names with placeholders (Company A, System X) and round any figures. The review does not depend on them.
Below is a lending rules card I filled in for one finance process. Act as a skeptical reviewer at the decision meeting and argue the case for NOT YET.
For each row:
- Say whether the answer is specific enough to act on (a number, a named role, a date) or too vague.
- Point out where answers contradict each other, for example the share of volume the system handles against the rules captured so far and who works the flagged few.
- Say what evidence you would ask for before accepting it.
Do not rewrite my answers and do not invent facts about my process. If a row is blank or says "Not known", treat it as not ready.
Finish with: your recommendation (BUILD, NOT YET or RE-CHECK ON a date), the three rows that most need work, and one question I should ask my team on Monday.
My card:
[PASTE YOUR FILLED CARD]
Prompt from The AI Decision Course by Dr. Tony Hoang, host of The Artificial Intelligence Podcast (Spotify, Apple Podcasts): tonyphoang.com/learn
To paste the card into prompt 1, copy its rows from the PDF.
Prompt 3: Make it yours
Paste this and answer its four questions.
Prompt
Before you start: use your company's approved AI assistant, not a personal account. Replace company, system, supplier and people names with placeholders (Company A, System X, Supplier B).
You are helping a leader apply one lesson to their own situation. The lesson's five decisions are below. Use only them and what I tell you. Never invent facts, numbers, vendors, laws or results about my organization or my industry. Where a number is needed and I have not given one, say what to measure and who should measure it, instead of guessing.
First, ask me these four questions, one at a time, and wait for each answer:
1. My industry, and roughly how many people work in my organization.
2. The finance process I have in mind, in one or two sentences.
3. Where it stands today: an idea, a pilot running, or about to launch.
4. What worries me most about it.
Then give me the following. Keep it short and practical:
A. My plan. For each of the five decisions: one action written for my process, the role that should own it (a role, not a name), and what should be done within 30 days. If my industry makes a decision harder, for example regulated lending, model-risk review or audited credit decisions, say which decision and why in one sentence.
B. Three questions to ask my team on Monday, specific to my process.
C. Ask me which one I want: a checklist for my process, a rubric to score the system's decisions, or a one-page brief for my boss. Then write it, and finish with a prompt I can paste later to update it.
Mark anything that rests on an assumption with "(check)".
The five decisions:
1. Model scores, rules decide. Train a predictive model on your own lending history, then wrap it in written policy rules that override any score it gives.
2. Write your experts' instincts down as rules. Ask the same question enough times and experienced decision-makers can state their intuitions as explicit rules.
3. Let the model take the volume. One lender cut turnaround from 24 hours to nine seconds, with the system handling roughly 75% of volume.
4. Automate reconciliation next. Matching line items between receipts and purchase orders is a messy problem for finance and accounting teams in large companies.
5. Know your cost per loan. Price it fully loaded before you set the automation budget.
Prompt from The AI Decision Course by Dr. Tony Hoang, host of The Artificial Intelligence Podcast (Spotify, Apple Podcasts): tonyphoang.com/learn
Your company's assistant may answer differently, so read what it writes before you use it.
Check yourself
A vendor pitches a lending model with no policy shell. What do you add before it touches a live application?
Show the answer
The written rules that override the score. Without them the model is making policy alone.
Your best underwriter cannot explain a decline. What is your next move?
Show the answer
Keep asking until the instinct comes out as a rule, then write it into the software.
The model is confident on an application that breaks a policy rule. What happens?
Show the answer
The rule wins, whatever the model's confidence.
You have one automation budget line for finance. Where does it go first?
Show the answer
The lending pipeline. Reconciliation comes once it runs.
Sources
The advice draws on interviews with the co-founder of an AI engineering consultancy, the chief technology officer of an enterprise AI-agent startup and a managing partner of a venture firm focused on real estate.
Mortgage Bankers Association, "Chart of the Week: Retail Production Channel-Cost to Originate a Loan", MBA Newslink, 22 June 2026. (https://newslink.mba.org/mba-newslinks/2026/june/mba-newslink-monday-june-22-2026/chart-of-the-week-retail-production-channel-cost-to-originate-a-loan/) What it counts: the retail production channel only, fully loaded, from the MBA and STRATMOR Peer Group Roundtables benchmark run in spring 2026, an industry-body benchmark of participating lenders rather than a census. Independent mortgage companies averaged $12,209 per loan in 2025, so the depository figure is not an industry average. The round $16,000 heard in the interviews matches the depository retail figure only.
Ask about this lesson, or how it applies to your team. Every answer cites the lessons it draws on.
Try asking
Free preview: 3 questions a day, answered from this lesson. Follow-ups remember your earlier questions on this visit. Don't paste confidential details.