Refinance opportunity scoring — what actually makes a client likely to refinance

Every broker knows that "old loans refinance more often", and most stop there. The problem is that on a 200–500 loan book, you can't ring every client whose settlement was three years ago. You need a ranking — something that tells you which loans to call first this week.

This is the working model I'd build for a junior broker who has the data in front of them but no idea where to start. It's a heuristic, not magic. But it beats "I'll call the ones I remember."

What you're actually trying to predict

The honest framing: you're predicting which clients are most likely to act on their loan in the next 3–6 months. That includes refinancing away (bad — clawback risk if recent, and lost trail either way), refinancing internally with you (good), or restructuring to release equity (great).

You're not predicting who deserves to refinance. You're predicting who's about to.

The signals that genuinely matter

In order of weight:

1. Loan age (single strongest signal)

Australian loans behave predictably here. Three windows:

• 0–18 months: refinance probability is low. The client just settled, the comparison shop is fresh in their head, and any move triggers an upfront clawback for the broker. Most clients don't move in this window unless rates jump materially.
• 18–36 months: probability climbs. The honeymoon discount has rolled off, the new rate looks worse than a freshly-advertised one, and the client starts seeing competitor ads with intent.
• 36+ months: the population of "I should look at this" clients dominates. By month 48, comparison-shopping behaviour is the modal customer state.

If you score nothing else, score age. A loan settled in 2022 is a candidate; a loan settled three months ago is not.

2. Repayment trend

Pull the last six months of balances from your aggregator file. If the balance is dropping faster than scheduled, the client is paying ahead — usually a sign of financial breathing room and lower refinance intent (they're happy with the product).

If the balance drop is slowing — repayments below schedule, or at the minimum — that's pressure. Pressured clients shop. Pressured clients refinance.

The math is unfussy: take a least-squares slope of the last six monthly balances, normalise by current balance, and classify into accelerating, steady, or slowing. The "slowing" bucket is your call list.

3. LVR (when you have it)

Sub-70% LVR clients have leverage. They can refinance into competitive rates, and lenders compete for them. Sub-60% gives them their pick of cashback offers.

Above 80% LVR, the universe of lenders willing to refinance shrinks, and the client typically stays put. They're not necessarily happy — they're stuck.

If your aggregator file doesn't have LVR, you can proxy with current_balance / original_loan_amount, but it's a weak proxy because property values move independently of the loan.

4. Lender (the stickiness prior)

The Big 4 (CBA, ANZ, NAB, Westpac) all have above-average refinance rates compared to specialty lenders. Two reasons: their standard variable rates tend to drift higher than market-leading rates over time, and their existing customers are precisely the demographic that competitor lenders target with cashback campaigns.

A 36-month-old ANZ loan is statistically more likely to move than a 36-month-old Macquarie loan with the same LVR. Score for it.

5. Cashback campaign timing

Less a signal in the data, more a contextual overlay. When a major lender runs a $4,000 cashback campaign for refinancers, the population of "candidates" temporarily expands. Watch lender announcements; if Westpac runs a 60-day cashback, every 24+ month CBA loan in your book moves up the list.

Signals that don't matter as much as people think

A few things brokers overweight:

• Total loan size. A $400k loan refinances at roughly the same rate as a $1.4M loan. The dollar value at stake is bigger, but the probability isn't.
• Postcode / state. Anecdotes about Sydney metro vs. regional refinance behaviour don't survive contact with the data once you control for LVR and age.
• First-home-buyer vs. investor. Some difference, but it's swamped by the four signals above.
• Repayment type (P&I vs. IO). IO loans roll back to P&I and that often triggers a shop — but it's a one-time event you can flag separately rather than a continuous signal.

Don't waste model weight on these. Save it for age and trend.

A scoring function that works

If you wanted to build this in a spreadsheet:

base = 50
+ 20 if loan_age_months > 36
+ 10 if loan_age_months between 18 and 36
+ 10 if LVR < 70%
+ 10 if repayment_trend = 'slowing'
+ 10 if lender in (CBA, ANZ, NAB, Westpac)
+ 5  if loan_age_months > 48 and lender is Big-4
- 30 if loan_age_months < 18  (clawback territory — don't proactively cannibalise)
- 20 if status != 'active'

That's a 0–100 scale that ranks every loan in the book by refinance probability. Top 25 by score is your call list for the month. Re-rank monthly as ages tick over.

It's blunt. It works. You can layer AI-generated reasoning on top to remember why each loan scored highly ("ANZ standard variable, settled Nov 2022, balance plateauing since Jan") — that's what Trail AI's refinance scorer outputs alongside the score — but the ranking itself doesn't need ML to be useful.

Calling cadence

A 300-loan book typically has 40–80 loans in the "candidate" zone at any time. You can't call them all. The discipline:

• Pick the top 10 by score each Monday.
• Call them that week.
• Re-rank the following Monday.
• Repeat.

That's roughly one productive refinance conversation every two days from your existing book — without buying leads, without paid ads, and without making cold calls.

What this looks like end-to-end

A well-run refinance pipeline on a 300-loan book might produce:

• 40–80 candidates flagged per month
• 10–20 contacted
• 4–8 substantive conversations
• 2–4 internal refis closed
• 1–2 lost to competitors (which is fine — they were going to move anyway)

The clients you don't call are the ones who quietly walk in 18 months when a friend mentions a better rate. The scoring exists to make sure that conversation is with you, not with them.

Key takeaways

• Score every active loan: age (heaviest), repayment trend, LVR, lender, with a sub-18-month penalty.
• The top-25 ranked list is your weekly call queue. Re-rank monthly.
• Lots of "obvious" signals (loan size, postcode, FHB status) don't move the needle. Don't overweight them.
• The math is straightforward — the discipline is doing it consistently. Most brokers don't, which is why the few who do compound their book defensively.
• For an automated version, see how Trail AI's refinance scorer ranks every loan with AI-generated reasoning per row.