January 27, 2026
As debate grows over the limits of credit scores like FICO, it’s clear that lenders can make more informed, risk-adjusted decisions with better data.
Lenders are increasingly looking at alternative data sources, such as cash flow analytics, rental and utility payment history, and proprietary transaction data to gain a complete financial picture, and importantly, drive more approvals.
However, this shift begs the question: is more data always better?
The Winning Strategy: Quality over Quantity
When it comes to the data that fuels models for automated decisioning, more inputs can often hurt lenders more than they help. Models that look to hundreds or thousands of inputs introduce significant regulatory and compliance burdens, often without any real performance gains.
For example, a fintech lender might use 100 alternative data points, only to find it provides the same predictive power as a simple 10-variable credit report. Instead of better results, they inherit a massive compliance burden because they must now explain how each of those obscure variables—many of which may be unintentional proxies for race or age—impacted the final decision.
When discussing data for automated lending decisions, it’s helpful to think in terms of rows and columns.
More rows means more loans, ideally with recent performance information, are always valuable because they create a wider diversity of situations and outcomes for models to learn from.
More columns offer more information about a particular borrower or loan, but they often correlate with other information already contained in the data set. To incorporate more columns, a model must become more complex and less transparent.
Introducing a range of data sources typically increases the number of columns, not rows. Lenders must ask themselves if the regulatory and compliance burdens of complex models are worth the minimal, often hypothetical, performance gains.
If you choose to use a specific data set that makes your model 1% more accurate but 100% harder to explain to a regulator, is it worth it?
The ability to transform your decision-making and unlock new growth is likely hidden within the data you already have. The true power lies not in the volume of data, but in the selection of the right inputs and an understanding of their relationships.
Key to Success: Optimize the Data You Have
For most banks—especially community and regional institutions—”big data” often feels more like a concept than a readily available resource. This presents a significant question: how do you harness the power of AI/ML to create smarter decisioning strategies and unlock new opportunities when you’re not swimming in data?
AI doesn’t demand perfect, infinite data. Instead, data demands smart, transparent AI that can maximize its value.
Instead of embarking on a multi-year “big data” initiative to leverage AI decisioning effectively, financial institutions should:
- Prioritize Data Quality: Models are only as good as the data they’re trained on. Inaccurate or biased historical data can be amplified by AI, leading to unfair or discriminatory outcomes. What are your current data sources, and how can they be centralized or simplified? Prioritizing high data quality is essential, and even minor improvements in data quality can significantly boost AI performance. Standardize your data, utilize integration tools to unify repositories, and establish robust data governance for both quality assurance and compliance (see our previous blog for more on data management).
- Extract More Value from the Data You Already Have (ie, Declines): Your historical data is your most valuable asset. But are you using all of it? Many institutions ignore rejected applications, creating a significant blind spot. Champion the use of reject inference, which allows you to glean insights from your declines. This doesn’t just improve risk assessment; it can help identify underserved communities that your institution can responsibly lend to, directly addressing fair lending concerns. By using this overlooked data, you’re not only being more compliant but also unlocking a new source of profitable growth.
- Try Less Data-Hungry Decisioning Methods: Conventional ML models often rely on brute force: feed them millions of examples, and they’ll eventually find patterns. This approach struggles with smaller, more nuanced datasets. It can overfit (learn too much from the few examples it has, making it unreliable for new situations) or simply fail to find robust patterns at all. But if you don’t have mountains of data for conventional machine learning methods, the answer isn’t always more data. Some AI-driven decisioning solutions (like the ones we build at Stratyfy) excel with small datasets.
We’re Helping Lenders Maximize their Data, Responsibly
At Stratyfy, we specialize in helping financial institutions harness the power of their data. We enable lenders to create robust and explainable decisioning strategies from their existing datasets, driving new opportunities, enhancing accuracy, and ensuring responsible growth.
Get a complimentary data quality assessment from Stratyfy: reach out to our team today.