The future of credit underwriting
The process of underwriting a consumer or a small business has always been fascinating for me. I’ve always found it stimulating to determine someone’s creditworthiness using a more or less complicated formula.
That’s why, over the last decade, I’ve consistently gravitated toward lending and credit—across different companies, geographies, and markets.
In this post, I’d like to draw a few conclusions from those experiences and reflect on where credit underwriting might be heading. The goal here is to explore the future of the underwriting process and analyze the forces driving its transformation.
Future Outlook
Automation is already deeply embedded in credit underwriting—even traditional banks now offer fully automated processes for small-ticket loans and partially automated ones for larger cases. But the Age of AI will push automation even further.
I believe the underwriting process of the future will be almost entirely automated, not just for smaller loans but across the board. Financial institutions may choose to retain some form of manual oversight—but that will be a human choice, not a technical limitation.
In short, every credit decision will be automatable, and humans will intervene only when they choose to.
The result? Application and approval times will shrink dramatically for nearly all types of credit. Credit will become even more ubiquitous—embedded as a seamless payment option across more product categories and more customer touchpoints.
Data sources
One of the key enablers of this near-instantaneous underwriting is the rapidly improving quality of programmatically accessible data.
Traditional credit bureaus still sit at the center of most underwriting processes, especially for consumer credit. Most incumbent lenders continue to evaluate applications based on a mix of form data and traditional bureau reports.
But this data has serious limitations: it’s not real-time, it’s difficult to integrate, and it often requires heavy ETL work to be production-ready.
Over the past decade, a first wave of innovation came from providers like Tink, Plaid, and TrueLayer. They introduced API-first products that made it far easier for lenders to access and use banking data, accelerating product development and raising the bar for data integration.
Now, a second wave of financial data providers is emerging. These players are not only API-first but also real-time and vertically specialized.
A very interesting project in this new wave is Infact, a real-time credit bureau. It basically offers a single API to lenders that can search, report and manage information on their borrowers, all based on real-time data. Right now the company is active only in the UK, but with promise to further expand in the near future.
A different angle has been approached by Qlarifi. Qlarifi is an API-first, real-time provider that focuses specifically on BNPL data. It basically offers lenders access to a database of BNPL consumers and their transactions, a vertical that is often underserved by traditional bureaus.
Teal is taking a similar path. Teal offers real-time and API-first salary data, allowing lenders to verify with more efficiency and a stronger degree of trust, the income attestation of their borrowers.
Infact, Qlarifi and Teal are only a subset of the startups emerging in the financial infrastructure space, but they reflect a broader trend: the unbundling and replatforming of credit bureaus around real-time, API-first, and highly specialized data services.
Through providers like these, as they offer a more standardized way of accessing data, lenders are able to iterate on models faster and to make decisions with more precision in a fraction of the time, thus improving their profitability and user accessibility to credit.
Decisions & Models
A second key piece of innovation is presented by modern decision engine SaaS.
Right now, the concept of a decision engine is generally accepted as a standard infrastructural piece in basically every lender architecture - as I also pointed out in my definition of the lending stack a few years ago.
Anyway, considering that lenders see underwriting as one of their main competitive advantages, they tended to use in-house decision engines. These systems often required significant technical resources to build, maintain, and update.
The situation is now drastically changing and decision engines too are becoming an off-the-shelf piece of software that lenders can source externally. Companies like Taktile and Noble epitomise this evolution.
Both started as decision engine platforms that allowed faster creation and, more importantly, iteration of credit risk models. While maintaining a strong decision engine connotation, they now evolved towards being risk decisioning platforms that go beyond credit risk and include other risk dimensions, like compliance, identity and fincrime.
The obvious insight here is that what was once a complex and expensive piece of technology that incumbents used to build and maintain internally (the decision engine) is now a software that lenders can source externally and that requires minimal dev support to be configured and run.
What’s even more transformative is how these modern platforms unlock the full potential of AI within the underwriting process—something proprietary decision engines often struggle with.
Legacy, in-house decision engines tend to be rigid, tightly coupled to internal systems, and costly to update—making it hard to experiment with new models or integrate advanced machine learning. In contrast, modern platforms are designed from the ground up to be modular, API-friendly, and data-agnostic. This makes them ideal environments for AI-driven workflows: underwriting agents that can autonomously test, learn, and optimize decision strategies based on evolving data patterns.
In this new setup, credit models can not only evolve faster but also proactively identify and incorporate new data sources—like alternative credit data, real-time cash flow, or behavioral signals—to enhance prediction accuracy. Underwriting, in other words, becomes a dynamic, adaptive system rather than a static rule set—driven not just by better data, but by a constant feedback loop of learning and improvement.
Conclusions
Credit underwriting is an ancient human activity that goes back thousands of years. Humans have been exchanging IOUs for millennia, and the art of assessing creditworthiness has been central to the evolution of commerce and trust in society.
Today, the Age of AI is revolutionizing how credit underwriting is assessed. The rise of real-time, API-based credit bureaus and the emergence of SaaS decision engine platforms are fundamentally reshaping the way lenders operate.
Building a lending product has never been easier: lenders can now develop better models, faster, and with fewer resources. This means more accurately priced credit products, offered to more people, in more contexts.
This new underwriting stack is no longer a future vision—it’s already taking shape. And incumbents that fail to adapt risk being outpaced by leaner, smarter, AI-native challengers.
P.S. I'll be attending Money 20/20 in Amsterdam (3rd-5th of June), if you are around and want to meet please drop me a message