Scaling a loan book without losing credit quality

A lending business that has found product-market fit faces a particular temptation. Demand is there, funding is available, and the quickest way to grow is to approve more of the applications that are already coming in, or to open new channels that bring in more of them. Both work. Both also change the portfolio in ways that will not show up in the default rate for some time.

The challenge in scaling a loan book is not growth itself. It is keeping enough visibility to know what kind of growth you are buying, and keeping enough control to slow down the parts that are not working without stopping the parts that are.

Read vintages, not the portfolio average

When a book grows quickly, the portfolio-wide default rate improves almost automatically, because new loans have not had time to default yet. That makes the average one of the least useful numbers during a growth phase. Vintage analysis, which follows each cohort of loans from the month it was issued, shows whether recent loans behave like older ones at the same age. It is the most reliable early signal that something in origination has changed.

Watch the approval rate as closely as the loss rate

Approval rates drift for many reasons: a new channel, a marketing campaign that attracts a different audience, a rule adjusted to hit a volume target, a data source that quietly stopped returning results. Each of these changes who is being lent to. Tracking approval rates by channel, product and segment, and asking why whenever they move, catches most of these shifts before they reach the loss line.

Treat each channel as its own portfolio

Customers acquired through a partner, through paid search and through an existing customer base are rarely the same customers, even when they pass the same rules. Channel mix is one of the most common hidden drivers of portfolio change during growth. Reporting risk and economics by channel, rather than in aggregate, makes it possible to grow the channels that perform and to fix or close the ones that do not.

Make early arrears a daily metric

Defaults are a lagging indicator. Missed first payments and early arrears are not. A rise in customers who never make their first payment often points to fraud or to a broken step in onboarding. A rise in early arrears among otherwise good customers can point to affordability problems that the scorecard does not capture. Both deserve a daily or weekly view during growth, with a named owner and a defined response.

Scale operations at the same speed as volume

A loan book that doubles also doubles the number of customers who need servicing, the number of payment exceptions and the number of accounts in collections. If those teams and systems grow more slowly than the book, service quality drops and recoveries fall, which shows up later as credit loss even though the cause was operational. Capacity planning for operations belongs in the same plan as the volume targets.

Keep the funder close

Rapid growth usually means drawing more on funding lines, and funders notice portfolio changes as quickly as lenders do. Sharing vintage and channel data proactively, explaining changes before they appear in covenant reports, and agreeing in advance how new products or channels fit within the facility all make it easier to keep funding aligned with growth rather than chasing it.

Signals worth acting on

• Recent vintages performing worse than older ones at the same age.
• Approval rates moving without a decision anyone can point to.
• One channel growing faster than the others with weaker early performance.
• Rising first-payment defaults or early arrears.
• Servicing or collections backlogs growing faster than the book.

Growth you can explain

The aim is not to grow slowly. It is to grow in a way that the business can explain at every stage: which customers it is adding, through which channels, at what expected loss and at what cost. A portfolio that can be explained can be funded, managed and corrected. One that cannot tends to surprise everyone at the same time.