Global banking net income hit USD 1.3 trillion in 2025, the industry's best year on record, according to McKinsey's Global Banking Annual Review 2026. But the same report points to a very different kind of pressure behind that growth. Fintechs now account for 17% of industry revenue. Transaction banking and distribution are generating more of banking's profits than lending relationships. And customers are adopting new technology faster than banks have historically been able to keep up.
For lenders, that pressure is showing up in the way they build and run credit products. Underwriting needs to work with more data, including for borrowers with little or no bureau history. Risk teams need to understand and defend AI-driven decisions. And the same credit policy needs to work whether the borrower comes through a bank's own app, a DSA, or an embedded-finance partner. At the same time, the rules around lending keep changing across markets.
A lot of that pressure lands on the loan origination system. The LOS sits between the customer-facing experience and the core banking system, and it is the layer that turns a lending product from an idea into something a lender can actually launch, underwrite, and manage.
So what should lenders expect from a modern LOS now?
Five capabilities keep coming up when growth, product, and risk teams evaluate one. And they are more connected than they first appear.
1. Modular architecture, not a core rebuild
Almost nobody is arguing for replacing core banking systems this year. Most institutions couldn't afford the bill even if they wanted to.
Global spending on outdated payment technology specifically is projected to climb from USD 36.7 billion in 2022 to USD 57.1 billion by 2028, according to IDC, with maintenance costs on that legacy paytech rising around 7.8% a year. That's a specific, tracked figure about payments infrastructure rather than banking technology in general, but it captures the same underlying problem: money spent standing still rather than building forward.
What most institutions are actually doing instead is layering a modular, API‑first origination system on top of what's already there, with pre‑built connectors into KYC providers, account aggregators, payment rails, bureaus, and LMS platforms, rather than custom integrations someone has to keep patched. It's part of why the loan origination software market, sitting near USD 5.94 billion in 2024, is still expected to more than double by 2030 even while core modernisation spend stays flat.
This part of the stack rarely comes up in a first vendor call, but it's the one that ends up mattering most a year in. It shows up as how many integrations a team redoes every time it launches a new product or doesn't. FinBox LOS handles this by connecting 30 or more integrations across KYC, bureau, account aggregator, and repayment once, then reusing them across every product a lender launches afterward, rather than rebuilding per product or per channel.
2. Launch speed a release calendar can't give you
US embedded fintech products are set to carry over USD 7 trillion in transaction volume by the end of 2026, up from USD 2.6 trillion, per Bain Capital's estimate. Embedded lending overall is growing north of 20% a year through the early 2030s under most estimates, and roughly seven in ten SMEs say they'd rather apply for a loan digitally than visit a branch.
What “fast” means has changed. It used to mean beating your own last launch. Now it means keeping pace with a fintech that has no branch network or balance sheet, but can launch a similar credit product inside someone else’s app in a week.
On FinBox LOS, that translates into building a new lending program from a template or a plain‑language prompt, live the same day across products like personal loans, BNPL, MSME, secured, education, or merchant credit, with full programs typically live in about two weeks and edits made in minutes instead of a sprint. The point isn't the number itself so much as what it changes: launching a product stops being a milestone the roadmap revisits once a quarter and becomes something growth can just do, whenever a partner or a segment shows up.
3. AI decisioning a risk team can still explain
McKinsey estimates that banks which don't reposition their business models for AI‑driven shifts in consumer and competitor behavior could see global profit pools decline by roughly USD 170 billion, or about 9%, over the next decade, with individual lines like credit card lending exposed to steeper declines.
But underwriting is one of the slower parts of the bank to actually adopt AI. Industry surveys of banking professionals consistently find the vast majority of banks already use AI for fraud detection, while only a minority have deployed it for credit scoring and underwriting itself. Deloitte's 2026 banking outlook puts this down to fragmented data, legacy systems, and governance that hasn't kept up, leaving a lot of AI work stuck in pilots nobody has scaled. Fraud detection remains the top AI use case, with credit scoring and underwriting adoption lagging.
The problem isn't really the models. It's being able to understand why they made a particular decision. What most risk teams want isn't faster decisioning on its own, it's decisioning where every score traces back to the data point and rule that produced it, and every model can be opened up and inspected rather than taken on faith. FinBox LOS runs this through a single rules engine, so real‑time underwriting and predictive risk signals speed up approvals without the credit team losing the ability to walk a regulator through any individual decision. The AI runs the decision. Someone still has to be able to stand behind it.
4. Underwriting for borrowers the bureau never saw
Financial inclusion is genuinely improving. Global account ownership reached 79% of adults in 2024, up from 74% in 2021, according to the World Bank’s Global Findex Database 2025. But approximately 1.3 billion adults still do not have a financial account, with a significant share concentrated in a small number of countries, including India and Indonesia.
But having a bank account is not the same as being visible to a credit model.
The SME Finance Forum estimates the global MSME financing gap at approximately USD 5.2 trillion annually. Many small businesses may have accounts, digital payment histories, and real economic activity—but still lack the formal bureau history required by traditional underwriting models. That is the challenge.
A bureau-only model is often well suited to a borrower with a stable salary, a long credit history, and consistent bureau data. It is far less effective when assessing an informal worker, a microbusiness owner, or a new-to-credit customer.
The answer is not to replace bureau data. It is to supplement it with the right, consented signals:
- Cash-flow and transaction behaviour.
- Account Aggregator data.
- GST and business information.
- Utility and payment history.
- Device and behavioural signals.
- Application-journey indicators.
- Fraud and identity signals.
This is underwriting for the borrower in front of you, not simply for the borrower your historical model was trained on.
One FinBox LOS deployment with a leading NBFC illustrates what this can look like. Alternative-data underwriting helped new-to-credit customers reach 30% of approved customers, doubled approval rates, reduced delinquency by 75%, and lowered fraud incidence by 45% in the same portfolio.
5. Consistent risk rules across every channel
Retail and e-commerce already account for the largest share of embedded finance activity, and Asia Pacific has become the fastest-growing region for it, helped by mobile-first ecosystems and clearer regulation.
Borrowers rarely go looking for credit anymore. It shows up inside a checkout flow, a DSA's app, a marketplace seller dashboard, a partner's onboarding screen, plenty of places a lender's own risk team doesn't have direct visibility into. That's an easy gap to miss until an audit finds it: a lender can look perfectly clean on its direct channel and still be exposed through a partner running a slightly different version of the credit policy.
A modern LOS needs one rules engine enforcing the same underwriting logic, eligibility checks, and fraud controls no matter which channel an application came through. On FinBox LOS, a journey is built once and launched unchanged across mobile, web, DSAs, and partner channels, so the policy risk signed off on is the policy actually running everywhere. For risk, that turns a review-per-channel problem into one control point. For growth, it means the next DSA or embedded partner doesn't need a policy rebuilt from scratch.
What all five point to
A modular core is what makes fast launches possible. Fast launches only work if the decisioning behind them can be trusted. Trusted decisioning is what makes it viable to underwrite borrowers a bureau-only model would reject outright. Once you can serve that borrower well, you want to reach them wherever they are, which means every channel needs the same risk logic.
Most lenders aren't choosing between these five. They're running into whichever one their current stack happens to be weakest on, usually without realising the other four are downstream of it.
