The mortgage is the highest-value, longest-duration relationship in retail banking. A customer who holds their primary mortgage with a bank is also, in the vast majority of cases, the bank’s current account holder, savings customer, and the most likely candidate for subsequent lending and investment products. The mortgage anchors the relationship in a way that almost no other product does. Losing it to a refinancing competitor is not a single-product event. It is, in most cases, the beginning of a full relationship migration.
Most banks identify refinance intent after it has matured into a competitive shopping process. The customer has already compared rates, has already received one or more competing quotes, or has already submitted an application elsewhere. At that point, the bank is not managing a retention decision. It is managing a defensive response to a process that has already begun, and the cost of that response — in offer discount, advisor time, and relationship goodwill — is materially higher than the cost of a proactive approach made 60 to 90 days earlier.
Where the decision breaks down
The behavioral signals that precede a refinancing decision are visible in the data weeks or months before the customer acts. Rate sensitivity is the most direct: the spread between the customer’s existing mortgage rate and current market rates is observable and, as that spread widens, the customer’s incentive to refinance increases. A model that tracks this spread against each individual mortgage in the portfolio and alerts on widening can identify rate-sensitive customers without waiting for any behavioral change from the customer themselves.
Behavioral signals that complement rate sensitivity include changes in payment patterns. A customer who begins making additional voluntary repayments is demonstrating an interest in their mortgage balance trajectory and may be positioning for an early redemption or refinance. Increased engagement with the bank’s own mortgage rates pages or contact centre calls about refinancing are direct signals of active intent that are often available in digital channel and CRM data but rarely connected to the mortgage servicing record in a way that produces an automated alert.
Credit inquiry patterns provide a later-stage signal. When a customer formally applies for a mortgage at another institution, a hard credit inquiry appears on their bureau record. By this point in the process, the customer has completed their rate research, has received a competing offer, and is in the final stages of switching. This signal is real but arrives late. The 60-90 day window that makes proactive retention economically preferable to defensive matching requires identifying the earlier behavioral signals rather than waiting for the hard inquiry.
The two economic leakages
The first is direct refinancing runoff. When a mortgage is refinanced to a competitor, the bank loses the net interest margin on the remaining loan term, the redemption fee income is consumed by early settlement, and the customer transitions their primary banking relationship to the new lender at a rate that is higher than most banks formally acknowledge in their mortgage runoff analysis. The lifetime value loss is systematically underestimated when it is measured only as the foregone mortgage margin rather than the full relationship value including current account, savings, and future lending.
The second is the cost of reactive retention. A bank that identifies refinance intent after the customer has received a competing offer is negotiating against that offer. The discount required to match or beat a competitor’s rate, combined with the advisor time to manage the conversation and the relationship goodwill cost of having made the customer feel they needed to shop elsewhere to get a competitive rate, produces a per-retention cost that is substantially higher than the cost of a proactive offer made before the competitive process began. Proactive retention is not just more likely to succeed. It is cheaper per retained customer.
The asymmetry between proactive and reactive retention cost means the economic return on early detection is higher than a simple comparison of retention rates would suggest. A model that retains 60% of at-risk customers proactively at a lower offer cost produces better portfolio economics than a model that retains 70% reactively at a higher one.
The competitive dimension
The insight that makes this strategically urgent rather than just operationally important is that competitors with better early detection models are already prospecting the same customers. Digital mortgage brokers, challenger banks, and large-scale direct lenders have built outbound prospecting models that identify rate-sensitive mortgage holders and target them with refinancing offers before those customers have begun actively shopping.
A bank that waits for behavioral signals of active shopping to trigger its retention programme is not competing with those prospecting models. It is responding to them. The customers most likely to be targeted by competitor prospecting — those with the widest rate sensitivity spread, the longest time since origination, and the strongest credit performance — are exactly the customers the bank most wants to retain, because they are the lowest-risk, highest-value segment of the mortgage portfolio.
The first-mover advantage in the retention conversation is significant. A customer who receives a proactive rate review from their existing bank before they have received a competing quote is in a different psychological and commercial position to one who has already been offered a better rate elsewhere. The former is being looked after by their bank. The latter is being matched by their bank after they had to do their own shopping. Both may result in retention, but only one of them strengthens the relationship.
Identifying the right intervention
Not every mortgage customer showing rate sensitivity is worth a proactive retention investment. The model output needs to be combined with a relationship value assessment to prioritise the outreach. A customer with a high rate sensitivity score, strong behavioral signals of active intent, and a high lifetime relationship value is a clear priority. A customer with moderate rate sensitivity but low relationship value and a history of seeking the best rate at every renewal is a different proposition — retaining them may require an offer that produces no margin, or may not be worth pursuing at the prevailing offer cost.
The intervention itself should be designed to feel like financial advice rather than defensive retention. A customer who receives an outreach framed as “we have reviewed your mortgage and identified that current market conditions may create an opportunity for you to reduce your rate” is receiving a service. A customer who receives an outreach that feels transactional, or who perceives that the bank is responding to their shopping activity rather than anticipating their needs, will be less receptive and more likely to have already moved forward with a competitor.
The technology dimension
Mortgage servicing data — payment history, balance trajectory, origination rate, remaining term — combined with the customer’s full account behavioral history, sits primarily in the core banking and mortgage servicing systems that run on IBM Z at most large retail banks. A propensity model deployed on IBM Z via IBM Machine Learning for z/OS can score the entire active mortgage portfolio daily against a combination of rate sensitivity triggers and behavioral signals, drawing on the full account data estate without extraction. The scored output, ranked by estimated refinance probability and relationship value, feeds the advisor workflow or outbound campaign management system that manages the proactive outreach. The model runs on the data where it lives, the scoring is continuous rather than periodic, and the intervention reaches the customer through the channel and the relationship manager best positioned to manage the conversation.
What success looks like
The metrics for a mortgage retention programme are refinance retention rate, cost per retained mortgage, and runoff rate as a proportion of the maturing and at-risk portfolio. The programme should establish a baseline on all three before deployment, set targets for each, and track the difference between proactive and reactive retention cost as a measure of model value rather than just model accuracy. Prepayment rate and runoff rate are the upstream metrics that determine the portfolio’s sensitivity to rate environment changes. Retention rate and retention cost are the operational metrics that determine whether the programme is responding to that sensitivity effectively.
The customers retained through an early-identification programme are a different population from those retained through reactive matching — generally longer-tenured, more behaviorally engaged with the bank, and more likely to respond to relationship-based outreach. That difference is visible in subsequent attrition data and is one of the reasons why the long-term value of proactive retention exceeds what the immediate retention rate comparison would suggest.
