Why algo routing beats waterfalls — measured in basis points
Most lead networks ping lenders top-down until someone says yes. That model was never optimal — and at scale, it leaves real money on the table.
The default lead-network architecture is what we call a waterfall: a prioritized list of buyers, sorted by CPL or contract date, that gets pinged in order. Top of the waterfall sees every lead first. If they decline, the lead drops to the next buyer.
This worked when networks had three buyers. With a hundred lenders across seven countries, it's a structural underperformer.
The math the waterfall hides
Take a synthetic example. You have an applicant: 35 years old, requesting DKK 200k for a debt consolidation, top-quartile credit signals, applying at 14:00 on a Tuesday in Aarhus.
A waterfall sends this to Lender A first because Lender A pays the highest CPL. But Lender A's actual funding rate for this profile is 22%. Lender B, two slots down, funds this profile at 51% — but they only see the lead if A and the buyer between them both decline.
Three things go wrong:
- Lower fund rate for the network — the applicant is twice as likely to walk away with no loan.
- Lower revenue for the broker — they only get paid on funded loans (or sold leads, depending on contract), and the waterfall path produces fewer of both.
- A worse experience for Lender A — they waste underwriting capacity on leads they won't fund, which makes their unit economics look bad.
What we do instead
We model each lender's funding probability for the specific applicant, not the segment. The features look something like:
- Risk-band fit (their last 90 days of accept/reject signal)
- Requested-amount fit (their funding distribution by amount)
- Time-of-day fit (some lenders process applications faster on weekday mornings — and that affects fund rate)
- Geographic fit (regional product preferences)
- Capacity (have they hit a daily volume cap?)
The probability model retrains hourly. The applicant goes to the lender with the highest expected funding probability — not the highest CPL.
What this means for everyone
For brokers: higher EPL. We measure meaningful lift across our broker book versus the waterfall they were on before.
For lenders: better-fit traffic. They underwrite leads that match their actual book, and their accept rate goes up.
For applicants: a lender that can actually fund them. The single highest- impact UX improvement in our network was switching to algo routing — applicants are more likely to walk away with the loan they came for.
The waterfall worked when networks were small and unsophisticated. They aren't anymore. Routing intelligence is now the table stakes.