A loan marketplace never lends anyone a dollar. It takes a borrower's request, decides who on its panel is likely to say yes, sends the introduction, and gets paid for making the match happen. That is the entire business, and it is a genuinely different engineering problem from building a lender, one that lives or dies on how good the routing is rather than how good the underwriting is.
Here is the plain version. A borrower fills in a short form. Before anything touches their credit file for real, a soft check runs against the panel's criteria and comes back with likely offers, no score impact. Only when the borrower picks one does a hard application start, at which point the marketplace has already done its job and gets paid whether or not the loan ultimately funds, depending on the model. The loan-app guide covers where a marketplace sits among the other three ways to build a lending product; this is what happens inside the one that just matches.
Soft search first, or you lose everyone
The single feature every serious player in this space treats as non-negotiable is a pre-qualification step that runs before any hard credit inquiry. Comparison sites in the UK run an eligibility check against the borrower's file and tell them plainly it will not affect their score, then show a likelihood, sometimes as a plain percentage, sometimes as a rating band. US platforms do the same thing under different names: a pre-qualification flow that checks fit against lender criteria using a soft pull, so a borrower sees real offers before committing to anything that shows up on their credit report.
The reason this is table stakes rather than a nice-to-have is conversion math, not generosity. A borrower who has to accept a hard inquiry before seeing a single offer abandons at a much higher rate than one who sees likely outcomes first and only takes the hard step once they have chosen. Every platform that has been in this space for years converged on the same answer independently, which is a strong signal that it is the right one.
The two ways to match a borrower to a lender
Underneath the soft search, there are really only two matching models, and most serious platforms end up running some blend of both.
The first is rules and bidding, sometimes called a ping-tree: the platform ranks the lenders on its panel by their live bid and their acceptance criteria for that borrower's profile, and routes the lead to the winner in real time. It is fast, well understood, and it is where most marketplaces start, because it needs a criteria engine and a payout ledger more than it needs a model.
The second is data-driven ranking: instead of a single winner, the platform returns a rate table of several matched offers built from the borrower's profile plus whatever signal the platform has, credit data, bank-linked cash flow, prior behavior on the platform. This is closer to a recommendation engine than an auction, and it tends to show up once a platform has enough volume and enough first-party data to make ranking worth the investment over routing.
We have built the first kind at real scale: a real-time lead-routing platform for a UK instalment-loan network that processed tens of thousands of loan applications a day, matching each one against a live panel of lenders' criteria and routing the introduction in real time, with a full affiliate-facing analytics layer so partners could see their own conversion and earnings as it happened. The engineering center of gravity was never the matching logic itself, which is genuinely a modest rules engine. It was the plumbing around it: real-time criteria updates from a panel of lenders who all change their appetite constantly, a payout ledger that has to be right down to the individual lead, and partner-facing dashboards good enough that affiliates trusted the numbers without asking.
Who actually gets paid, and when
The monetization sounds simple from the outside and is not, once you look at how it actually reconciles. The cleanest public example: one large US marketplace discloses that it earns a match or transmit fee at the moment a qualified request is sent to a lender, not when the loan funds, and states plainly that a single request form can generate multiple match fees if it is transmitted to more than one lender. That is a materially different revenue model from a platform that only earns when a loan actually closes, and the two require different reconciliation logic entirely: one has to prove a transmission event happened per lender per lead, the other has to prove a funded-loan event happened and match it back to the original introduction, sometimes weeks later.
Layered on top of either model are the affiliate-side economics: a bonus for higher-quality leads, a tiered rate that increases with volume, sometimes a small residual on a booked loan's ongoing revenue. None of this is exotic individually. What is genuinely hard is that every one of these payout types needs the same discipline as any other money-moving system: an event for every transmission, every acceptance, every funded loan, and an audit trail that can answer, months later, exactly why a specific affiliate was paid a specific amount for a specific lead.
The broker disclosure is not optional, and neither is the licensing
Nearly every serious platform in this space states, in almost identical language, that it is a credit broker and not a lender, because that distinction is the one that determines which regulatory regime applies. In the UK that means holding, or operating as an appointed representative under, FCA credit-broking permission. In the US it means state-by-state mapping of mortgage-broker, consumer-loan-broker, or lead-generation licensing wherever the platform has traffic or lender partners, which for a platform operating nationally can mean tracking dozens of separate regimes at once, each with its own renewal cycle and its own definition of what counts as brokering.
There is a second, quieter compliance surface that matters as much as the licensing: how the matching itself is designed. A ranking or routing system built purely around who pays the platform the most, with no consumer-benefit factor in the criteria, is exactly the shape regulators have flagged as a potential abusive practice. The fix is not complicated, but it has to be intentional: document the ranking criteria, keep some consumer-facing signal like acceptance likelihood in the mix, and be able to show, if asked, that the matching is not pure highest-bidder.
What actually gets built
Strip it down to the pieces that recur across every platform that has been doing this for years, and the build is a soft-eligibility pipeline that runs before any hard pull, a criteria engine that ingests and updates lender appetite in something close to real time, a routing or ranking layer that can run either a rules-based auction or a scored rate table, an embeddable widget or API for partners who want the flow inside their own product rather than sending traffic to yours, and a webhook and postback fabric so partners can see conversion and get paid without asking. None of these individually is exotic engineering. The platforms that last are the ones that got the plumbing between them right: the event trail that proves who was paid what and why, and the criteria engine that can absorb a lender changing its appetite without a deploy.
What 2muchcoffee covers
We build the routing, matching, and payout infrastructure behind lending marketplaces, including the real-time criteria engine and the partner-facing analytics a high-volume affiliate network needs before its partners will trust the numbers. If your matching logic is simple but the plumbing around it, the event trail, the payout reconciliation, the partner dashboards, is where the real risk lives, that is the build we do. The plain way in is the AI and engineering work we do.
FAQ
How does a loan marketplace make money if it never lends anything? It gets paid for the introduction, either a fee when a qualified request is sent to a lender, a fee only when the loan actually funds, or both layered with volume bonuses. One large US marketplace discloses that a single request form can generate multiple match fees if it goes to more than one lender.
What is soft search in a loan marketplace? A pre-qualification check that runs against lender criteria without touching the borrower's real credit file, so the borrower sees likely offers before any hard inquiry. Every serious platform in this space treats it as the entry point rather than an optional feature, because it converts far better than asking for a hard pull upfront.
What is a ping-tree in lending? A rules-and-bidding matching model where the platform ranks lenders on its panel by their live bid and acceptance criteria for a specific borrower profile, then routes the lead to the winner in real time. It is the faster, simpler alternative to a data-driven rate-table model.
Is a loan marketplace regulated the same way as a lender? No, but it is still regulated. Nearly every platform discloses that it is a credit broker, not a lender, which in the UK means FCA credit-broking permission and in the US means state-by-state broker or lead-generation licensing wherever it has traffic or lender partners.
Can a loan marketplace's matching logic get it in regulatory trouble? Yes, if the ranking is driven purely by which lender pays the most with no consumer-benefit factor. Regulators have flagged that shape of matching as a potential abusive practice, so the criteria need to be documented and defensible, not just profitable.
What is the hardest part of building a loan marketplace? Not the matching logic, which for a rules-based model is genuinely modest. It is the plumbing around it: ingesting real-time criteria changes from every lender on the panel, and keeping a payout ledger and audit trail precise enough to defend exactly why a specific lead earned a specific fee.
One concrete action
Take one lead your platform routed last month and try to reconstruct, from your own data alone, every event that led to its payout: the criteria it matched, the lender it was sent to, whether it converted, and what was paid because of it. If that chain has a gap anywhere, you have found the part of the build that matters more than the matching algorithm, because a marketplace's entire product is trust in a number, and the number is only as good as the trail behind it.
