April 30, 2026
Meet the new engine behind Plaid Income
Income verification has a fundamental problem: it’s expensive, high-friction, and often incomplete. Legacy methods like payroll connections and document uploads can be slow and don’t work for everyone—especially those with non-traditional or variable income.
More recently, bank account connections have emerged as a more effective way to verify income, offering broader coverage and lower user friction. But figuring out how much someone earns from bank transactions introduces a different challenge: deposit activity is messy, irregular and hard to interpret at scale. Independent contractors are paid on different schedules than traditional payroll, while one-time lump-sum deposits can be easily misclassified. To accurately derive income from bank data, you need to go beyond surface-level labels and understand the underlying patterns behind each deposit.
That’s why we’ve rebuilt the engine behind Plaid Income, our CRA-backed income verification solution, from the ground up. Our latest release delivers a more stable, explainable, and configurable system for turning noisy deposit activity into clear, reliable income signals—and we believe it’s one of the most accurate income categorization systems available today. That’s not just because of the model improvements. It’s because of the data we train on, and the layered system we’ve built on top of it.
A smarter approach to income categorization
The same way an experienced underwriter develops intuition from reviewing thousands of files, Plaid Income draws on patterns learned from millions of financial data points. We’ve made system updates designed to understand income in context—not just by reading transaction labels, but by looking at how income behaves over time and what those patterns reveal.
At the center is a new income categorization model that uses a transformer-based LLM trained on large-scale financial transaction data across the Plaid network to better interpret noisy or inconsistent transaction data. Deposits that appear to come from the same underlying source of income are grouped into income streams, allowing Plaid to evaluate how income flows over time, how frequently it appears, how amounts shift, and what those patterns reveal about a borrower’s earning profile.
The result is stronger performance where it matters most. Income classification has improved by 48%, making Plaid significantly more reliable at identifying income from transaction history. In the ‘Earned Income’ category specifically—critical for making any lending decision—our new model delivers 84% precision, without sacrificing recall. In practice, that means fewer misclassified deposits and a more stable and explainable view of income across a wider range of borrowers.
A richer taxonomy for broader coverage
Better categorization is only part of the story. Our model update also introduces a richer income taxonomy, providing a more detailed and explainable view of how income is categorized. With more granular categorization, teams can see not just how much a person earns, but where that income comes from—and explain it more clearly across underwriting, verification, and compliance workflows.
The standardized taxonomy starts with six top-level categories—Earned Income, Transfers, Loans, Benefits, Retirement, and Other—and then drills down into specific types of income. For example, Earned Income can include Salary, Gig Economy, and Self-Employed earnings.
That added specificity is especially useful for identifying Self-Employed income—one of the most requested new subcategories from our customers. Historically, Self-Employed income has been harder to verify and easier to miscategorize because it often does not follow a traditional payroll pattern. By recognizing those earnings more accurately, Plaid Income gives teams a clearer view of income that may previously have been harder to evaluate with confidence.
Apply income calculation rules that reflect your policies
A richer taxonomy is only useful if teams can apply it to how they actually make decisions. Not every team calculates income the same way, which is why flexibility matters.
The latest version of Plaid Income is designed to make it easier to control what qualifies as income. For lenders, that can mean including or excluding certain types of income based on internal underwriting policies, instead of relying on a one-size-fits-all default. The same flexibility can support other workflows that require more tailored income calculations. Rules are enforced automatically and version-controlled, so decisions are consistent, traceable, and audit-ready.
Income data that stays current, without adding friction
This update also brings two improvements that make income data more reliable and easier to act on. Behind the scenes, Plaid uses a combination of classified transaction data and payroll and document benchmarks to validate income quality and consistency across different applicant profiles. And with embedded refresh, income data stays current automatically after an initial account link—no re-verification, no repeat uploads, no added friction mid-process.
These improvements are part of Plaid’s ongoing investment in income—because we believe our customers, from lenders to property managers and beyond, are best positioned to grow when they can make confident decisions with data they can trust.
Bringing more clarity to income verification
How people earn isn’t getting simpler, but the tools used to understand it can get smarter. Plaid Income helps teams make sense of that complexity with clearer categorization, more control over what counts as income, and data that stays current over time.