There's a pattern I've seen repeat itself across every lending product I've built: the funnel looks great on paper and falls apart in practice — almost always at the very first step.
When I launched a secured lending product on a large consumer fintech app, we had significant daily top-of-funnel traffic from a very large eligible user base. That's not a reach problem. That's about as strong a top-of-funnel as a lending product could ask for. And yet a large majority of those users were dropping off before they'd done anything meaningful. Overall conversion sat at a very low single-digit percentage.
The instinctive response to a number like that is to look at the UX. Maybe the form is too long. Maybe the copy isn't clear. Maybe the load time is bad. Those are real problems, and they're worth fixing — but they're not the root cause of majority drop-off. The root cause is almost always something more fundamental: the wrong journey being served to the wrong user.
The single-funnel trap
Most lending products are built with one onboarding flow. That flow is typically optimized for the "average" user — someone who is moderately aware of the product, has a reasonable credit profile, and is somewhere in the middle of the intent spectrum. The problem is that this average user barely exists in practice.
What you actually have is a population of users spread across a wide spectrum of awareness, intent, and creditworthiness. A first-time borrower who has never used this product before needs a completely different first experience than a repeat borrower with a strong credit history who just needs a fast path to disbursal. Serving them the same journey doesn't meet either user where they are.
This is the single-funnel trap — and it's the reason so many lending products spend months A/B testing button colors and copy while the real problem goes untouched.
Where AI/ML actually helps (and where it doesn't)
There's a lot of noise in the industry about AI/ML in lending products, and most of it focuses on underwriting — using machine learning to improve credit decisioning, fraud detection, and risk scoring. That's valuable work. But the underwriting layer is downstream of the problem I'm describing.
If the majority of your users are dropping off before they've submitted a loan application, a better underwriting model doesn't help you. The user never got that far.
Where AI/ML genuinely moves the needle on top-of-funnel conversion is in entry-point decisioning — using signals available before the user enters the active funnel to route them to the journey most likely to convert them. Not after drop-off. Before the journey has even started.
The signals that matter here are different from underwriting signals. They're behavioral (how has this user engaged with the app? have they visited this product screen before?), financial (what does their credit profile suggest about their familiarity with lending products?), and contextual (where are they in the intent journey?). These signals can be structured and fed into a model that makes a routing decision at the entry point — and that decision compounds through every subsequent step.
When I applied this approach to a secured lending funnel, overall conversion improved by ~3x, early drop-off reduced by ~35%, and monthly disbursals lifted ~50% within the first month of the new flows going live. The UX didn't change dramatically. The routing logic did.
The lesson for product managers building lending products
If your top-of-funnel conversion is poor, resist the urge to immediately optimize the existing journey. Ask a harder question first: is one journey the right answer for the population of users you're serving?
Segmenting your users before they enter the funnel — grouping them by awareness, intent, and creditworthiness signals — and building distinct journeys for distinct personas is more work upfront. But it changes the ceiling on what's achievable. A single optimized journey might move the needle modestly. A persona-based, AI/ML-routed set of journeys can move it by 3x.
The difference isn't the technology. It's the framing. Treat the funnel entry as a decisioning problem, not a design problem, and the right tools — including AI/ML — become obvious.