How Transaction-Level Data Helps Lenders See More and Risk Less

AI in lending is so often misrepresented as a silver bullet that either speeds up existing underwriting processes or as a black box-functioning digital layer that, in reality, would never pass compliance checks. These depictions couldn’t be further from the truth of what is reshaping credit risk assessment at the leading edge of the industry.

The AI that is actually changing the future of underwriting and lending stems from a powerful combination of deterministic code, machine learning and transaction-level data. This distinction matters from a practical and legal front; it means a system where every decision is explainable and where credit access is widened responsibly instead of regurgitating existing biases.

It’s about technology that both helps a creditworthy borrower’s profile escape the rejection pile and slashes risk in favor of more opportunities for financial institutions. Here’s why this way forward is worth pursuing, and how to do so.

An Overhaul is Urgently Needed

Traditional underwriting models are based on a relatively straightforward basis: a borrower’s history, including their past behavior, captured in bureau scores and statements, is a sufficiently reliable measure of their creditworthiness. Over the last century, this approach, defined by lagging indicators, was good enough, mainly because the data wasn’t easily available.

That has all changed now. Traditional underwriting and credit scoring models are retrospective, leaning heavily on a singular moment that doesn’t necessarily depict the full picture of a prospective borrower’s financial health. It’s now been more than two decades since digital data management overtook analog, where instantly-accessible information has long been the norm. A credit file that takes months to hit an underwriter’s inbox is a sign that the current model is a woefully antiquated way of processing information.

And, crucially, this outdated, often inaccurate data infrastructure only raises the threat of the invisible borrower, one that the finance industry must not ignore. Markets have changed drastically in the past half-century. Now, early-stage SMBs make up a huge portion of the private sector in the

US

, and the lion’s share in

Europe

. They’re increasingly seen as the backbone of the future economy, but infrastructural failures in credit scoring leave them at the wayside.

And people at an individual level face the fallout, too. In the UK, one study found that banks might be rejecting

80%

of loan applicants with thin credit profiles—who are actually low risk. That’s $4 billion (£3 billion) in lost low-risk lending. Experts nod to traditional bureau data as a core culprit.

This is precisely what transactional-level AI addresses. Instead of relying on lagging indicators plucked from historical records that could be months out of date, these systems analyze the here and now. Underwriters and lenders are given actual insight into the behaviors and financial health of a would-be borrower right now, not weeks or months back. How? Transactional-level AI, which extracts insights such as cash flows, payment patterns, spending dynamics, and income directly from the bank account.

That gives underwriters a continuous stream of reliable, relevant data to fully gauge a borrower’s financial behavior and health. And there are significant results tied to this approach. Lenders’ approval rates are increasing between 10 and 35%, without taking on extra risk. They’re also accelerating loan decisions by up to five times, and reducing portfolio losses by 15-40%.

Worth noting is that this is not a situation where lenders are suddenly taking a chance on high-risk borrowers. More borrowers do not necessarily mean more risk. What happens is that more borrowers are made visible—SMEs and individuals—in tandem with a methodology that is explainable from A to Z.

Of course, AI can not supplement human judgment, nor should it. Credit analysts and underwriters are still needed in the process moving forward. The role of AI should be to expand confidence, capacity, explainability and accuracy without sacrificing responsibility and human input. After all, finances are a highly sensitive area in terms of consumer trust: people want to know that there is a human involved.

What Transaction-Level Analysis Unearths

Understanding what AI analyzes is just as important as why it’s worth bringing into the underwriting process in the first place.

A bureau score tells an underwriter that a person or company has historically paid its debts. But this doesn’t empower underwriters and lenders to read between the lines on a case-by-case basis. Since it is ultimately just a snapshot, leaning exclusively on indicators such as bureau scores leaves analysts blind to wider patterns that tell the full story of a borrower.

Take the case of seasonal revenue patterns. For many SMEs, revenue is not a flat line but one that has peaks and dips that recover in cycles. These are perfectly normal in various sectors but sound off alarm bells when taken out of context—as is too often the case in a document that captures the moment but nothing beyond that.

Transaction data fed directly from the bank account provides a lens into a company’s overall resilience: how they recuperate revenue and manage the gaps in between peaks and dips. It provides much more insight into a business’s resilience, as well as allowing for proactiveness rather than a retrospective take on a borrower’s financial prospects. Traditional data doesn’t necessarily distinguish between a firm adeptly and responsibly navigating a slow quarter versus one that is steadily losing money with no end in sight. Transaction-level data does.

It can also connect the dots around supplier concentration that may otherwise be missed until much further down the line in a traditional underwriting model. Businesses with narrow supplier or client bases have a completely different risk profile than those with a wider base. A credit file doesn’t fully encapsulate the vulnerabilities attached to that, which can make or break a lender’s decision.

And cash conversion cycles, the time that it takes businesses to turn inventory into cash, are brought to light with transactional analysis, not just a balance sheet. An SME with a far more modest revenue but well-managed, tighter cycle could be much more resilient in the long-run versus a larger counterpart with strong reported revenue but slow conversion cycle.

The same logic applies to variable income patterns, which are becoming more common amid a rising gig economy and a surge in

freelancing

and startups. It’s more difficult for borrowers and loan applicants from these pools to avoid rejection where that income variability is a red flag at first glance. Transaction analysis, though, identifies the underlying pattern and wider trends around their spending habits, payment frequency, incomings versus outgoings, and general financial habits.

Here’s precisely where the real-time nature of digital, transactional analysis is critical. The indicators needed to get the full reading of risk and repayment viability are not generally static. A singular snapshot is less accurate because it’s not continuous. The combination of an indicator that was missed early on and delayed action is where most portfolio losses and added risk are found.

Making Data Democratic and Ensuring Responsible AI Use

Followed correctly, the transaction-level analysis approach also facilitates a fairer, more transparent underwriting process. Long-used parameters such as balance sheets and bureau data showcase a borrower’s history with the formal credit system. This functions somewhat adequately for those who have a longer history with this system. For those who don’t, including freelancers and young SMEs, it makes them invisible.

Transaction-level data is all about visibility, though. It sees what traditional metrics overlook because it sees how money is managed and behaviors fluctuate moment to moment. Underwriters access a much deeper layer of intrinsic understanding of a borrower’s creditworthiness. Because of this, the data is structurally designed to be more democratic.

A vital caveat: none of this means that AI automatically leads to a fairer system. Algorithms are trained on historical data and lending decisions. Biases that exist within the data used to train AI therefore makes it susceptible to bias. If, historically, certain groups were excluded from lending decisions, a poorly designed model will only serve to reproduce that bias and discrimination.

Unfortunately, there have been recorded cases of perpetuated discrimination because of poor transparency and a lack of auditing of AI systems—one recent study found that women were

consistently awarded lower credit scores

than men.

Design choices surrounding data infrastructure and workflow management in connection to AI determine whether fairness becomes a guarantee. AI must be implemented responsibly, and that includes ingraining transparency, auditability, accountability, and security at every stage. Which data is used, and its sources and destinations dictate what transaction-level AI sees and generates insights on. Consistent standards of data accuracy are directly connected with these tools’ reliability. Finally, traceability and explainability are the channels needed to identify where a model goes wrong or needs to improve on, and what it is doing right.

These traits are no longer optional nice-to-haves. Regulators are actually demanding auditable, transparent architectures that expel the risk of bias and discrimination. The EU’s Consumer Credit Directive 2 (CCD2), for instance, sets explicit standards for automated credit decisioning, including borrower rights to explanation and human review. In fact, this regulation has just

undergone revisions

that will make it significantly stricter later this year.

While speed is an important factor, it’s not the purely defining one. Accuracy, fairness, transparency, and visibility are equally important for financial institutions that want to keep up with evolving markets and regulations. Transaction-level AI, built within architectural frameworks that have auditability and accountability enshrined at their core, allows them to do so.