A cash advance loss rate is the share of advances that never get paid back, and the honest way to model it is by how long each user has been on the app. New users default far more often than users of a year or more, so one falling rate per year makes a forecast look better than it is.
The company is a US start-up that lets workers draw on wages they've already earned, before payday comes. It was getting ready for a seed round. It had a 12-slide investor deck and a five-year forecast that runs month by month.
The founder asked us to review the model before investors saw it. We checked where each number came from, then held the key ones up against what's normal for the industry.
The structure was good
The model flowed cleanly from one assumptions tab through to the statements and a dashboard. Gross margin sat inside the normal industry range. We said so in the review, because a founder needs to know what to keep as well as what to fix.
Four fintech model assumptions looked too good
We set eight of the model's year-five numbers beside industry ranges. Three passed. Four needed a closer look, and one was missing completely.
The four that needed a look all pointed the same way. Customer lifetime value was far above the usual 3x to 5x of what it cost to win the customer. Monthly churn was well under the usual 5% to 10%. The time to earn back that cost was much shorter than the usual 12 to 18 months. And the share of advances that never got paid back ended up below the normal 5% to 15% range.
That last one mattered most. The model used one default rate for each year, and it dropped fast. A single rate doesn't explain why people would stop defaulting.
People who are new to the app are the riskiest, because the company has no history with them yet. People who've used it for a year are much safer. So the honest way to model defaults is by how long each user has been around. We built a worked example with four tenure bands, from new users at 20% to 25% down to users of more than a year at 2% to 4%. With a sample mix of users, the blended rate came out at 12.2%. A growing app keeps adding new users, so its blended rate won't fall as fast as a single yearly number suggests.
The missing costs
The fifth number wasn't in the model at all. The forecast carried no cost for compliance. A company that advances money to US workers answers to federal consumer rules and to state laws that change from one state to the next. Most states also want a money transmitter licence.
We listed seven costs the model left out: a compliance officer, legal counsel, state licence fees, an annual audit, a legal reserve, data privacy tools and fraud checks. Together we put them at about 3% to 5% of revenue by year five.
The model also didn't show what it costs to borrow the money the company advances. The deck told investors the product wasn't a loan. Even so, an investor will ask what happens when advances go unpaid, and where the cash to fund them comes from.
What we handed back
- A five-tab review workbook the founder can work through line by line
- 14 recommendations ranked by priority, each scored for impact and effort
- Six high-priority fixes, led by cohort-based defaults and the missing compliance costs
- A plan for best-case and worst-case scenarios, with sensitivity tables
Why it matters for your raise
Strong numbers help a raise right up to the point where they look too strong. An investor who sees a loss rate below the industry floor will assume the founder hasn't thought hard about losses. Then they'll read the rest of the model with that doubt in mind.
A model that shows losses by user group, and pays for its own licences, gives up a little on paper. It wins back a lot of trust in the room.
If you advance money, model the advances that don't come back by how long each user has been with you, and put the cost of staying licensed in the plan before investors see it.
Questions founders ask
How do you calculate a cash advance loss rate?
Divide the advances that are never repaid by the total advanced over the same period. A single blended rate hides a lot, though. Split users into groups by how long they've been on the app, give each group its own default rate, and weight them by how many users sit in each. In our worked example, new users defaulted at 20% to 25% and users of over a year at 2% to 4%, for a blended 12.2%.
What is vintage loss analysis?
Vintage loss analysis groups loans or advances by when they were made, then follows how much of each group goes unpaid over time. It shows whether newer groups behave better or worse than older ones. For a fast-growing app it matters a lot, because new users keep arriving and they're the riskiest. That's why a growing company's blended loss rate won't fall as fast as one yearly number suggests.
What assumptions go in a fintech financial model?
Start with the ones investors test first: the cost to win a customer, lifetime value, monthly churn, payback time and losses on money you advance. Hold each one up against industry ranges, like lifetime value of 3x to 5x the cost to win a customer, or 12 to 18 months to earn that cost back. Then add what founders often miss: compliance staff, legal counsel, licence fees, audits, fraud checks and the cost of borrowing what you lend.