Banks Bet on AI Credit Scores — FICO's Turf Is Shrinking

This shift is quieter than a startup pitch deck and less glamorous — but it will matter a lot to everyday borrowers. Over the past year, banks and legacy credit firms have been quietly running and scaling AI models that score people on bank‑account cash flow, payroll feeds and transaction patterns, rather than relying only on traditional credit‑bureau history.

Why it matters now

• Consumer strain and better tech are meeting in the middle. With costs up and incomes more intermittent, many Americans no longer match the steady‑job profile FICO assumed. At the same time, models that read real‑time deposits and spending behavior have gotten markedly better.
• After Upstart’s stumble, incumbents didn’t give up — they shifted tactics. Instead of surrendering the field to fintechs, big banks are adopting or partnering around similar underwriting tech, but with tighter controls and compliance playbooks.

A quick history, in one paragraph

Credit scoring began as a blunt instrument: static scores, reporting lags and thin signals. Over the last decade, alternative data and machine learning promised more nuance — higher approvals for thin‑file applicants and fewer false negatives. That promise has always run up against bias, overfitting and regulatory scrutiny. What’s different now is scale: banks have the data pipelines and the risk‑management appetite to run these systems in production.

What’s changing in practice

• Lenders are ingesting bank‑account and payroll streams to assess repayment capacity almost in real time. That matters especially for gig workers and small business owners whose cash flow is lumpy.
• Credit bureaus and data brokers are scrambling to enrich their products — buying machine‑learning capabilities or bundling new signals to stay relevant.
• Card networks and processors are quietly folding these signals into approval and fraud flows, which speeds decisions at the point of sale and changes who gets instant acceptance.

The good and the bad

• The upside: more people with thin or damaged files could get access to credit; pricing can better reflect current ability to pay; small businesses may see faster decisions.
• The downside: transaction‑based models can reproduce disparate outcomes. Different spending patterns correlate with race and neighborhood, and regulators are already watching — enforcement or litigation is a real possibility.

What to watch next

• Will the CFPB and state regulators set clear rules for algorithmic underwriting? I’d expect guidance within about 12–18 months, though it may start vague and tighten over time.
• Can credit bureaus monetize this shift, or will banks cut out the middleman by keeping data and models in‑house?
• How pricing will evolve. Early signals: stable inflows are rewarded with cheaper offers; volatility is penalized — which can be unfair in many cases.

Practical advice

• Consumers: pay attention to the permissions you grant. Some apps request deep access to your transactions and payroll; know what they can see and who they share it with.
• Investors: watch Upstart (UPST) as a bellwether, and incumbents such as TransUnion (TRU) and Equifax (EFX) as they chase new revenue lines. Visa (V) and Mastercard (MA) could benefit if underwriting becomes embedded into payments.

The reality

AI underwriting isn’t a cure for credit inequality. It’s a sharper tool that can widen access for some and cut off others. Expect a stretch of real‑world stress tests: better access in some corners, surprising denials in others, and a regulatory debate that will shape whether these models broaden the pie or redraw its slices.

If you want to know whether an app or lender uses cash‑flow scoring, ask: Do you use bank‑account transaction data to underwrite loans, and can I opt out? It’s a small question today — but it will matter a lot in the next credit cycle.