Banks are beginning to use artificial intelligence to handle time-consuming parts of the loan-loss process, but they are not letting machines decide how much money to set aside for bad loans, according to accounting advisers who spoke with Thomson Reuters on July 9, 2026.
“It’s important to distinguish between AI assisting the process and AI actually estimating loan losses,” Gabe Nachand, a principal in Baker Tilly’s financial services practice, said.
Nachand said banks are using AI mainly for the routine legwork tied to a loan file — refreshing a borrower’s credit score, updating the value of a home or piece of commercial property used as collateral, checking whether that property is currently listed for sale, or matching terms within loan documents. What AI is not typically doing, he said, is deciding how large a bank’s allowance for credit losses should be for loans that might not get repaid — a figure required under an accounting rule known as CECL, or Current Expected Credit Losses under FASB ASC 326.
CECL applies beyond loans, including to certain securities and other financial assets. But for banks, Nachand said, the central concern is often the allowance for credit losses on loan portfolios.
“You have a portfolio of loans for which you measure the allowance related to those loans,” Nachand said.
Why banks won’t hand over the final decision
The obstacle isn’t whether AI can produce a number. It’s whether a bank can explain that number afterward.
Some AI tools function like a “black box” — they produce an answer without clearly showing how they arrived at it. For a bank, that’s a problem: regulators and auditors need to know what factors the AI considered, how those factors translated into the reserve estimate, and why the outcome differs from one loan to the next, Nachand said, and the entire process has to be documented in a way that is verifiable and auditable.
That requirement matters to anyone with a loan, because it’s the reason a human loan officer or credit committee — not a machine — is still the one signing off on how a bank accounts for risk tied to unpaid mortgages, auto loans and business credit.
AI systems can also shift over time as they take in new information, which adds another layer of caution. Nachand said AI-generated estimates would likely need to be tested side by side against a bank’s existing methods for an extended stretch before regulators would fully trust them. The most cautious banks, he said, keep new AI tools out of financial reporting until they’ve gone through “a rigid process of testing before releasing into a production environment.”
Despite the attention AI is getting across the financial industry, Nachand said the core math behind loan-loss estimates hasn’t changed much — the underlying models banks and their vendors rely on are largely the same ones used before AI entered the picture. He also said he has not seen a bank have to correct its financial statements because of bad data tied to an AI model estimating loan losses; when data problems do occur, he said, they tend to be data-quality issues that could just as easily happen with manual data entry.
The groundwork banks can’t skip
Jordan Anderson, who advises banks and credit unions on AI strategy and regulatory readiness at Baker Tilly, said many institutions are excited about what AI could do but may be underestimating the work required to use it safely.
“We all understand the promise of AI in financial services — but realizing that promise requires the right data foundation, governance framework, and change management discipline,” Anderson said by email.
In practical terms: a bank needs clean data, clear internal rules and strong oversight before AI can be trusted with anything sensitive. Skip that groundwork, Anderson warned, and an AI investment can turn into a liability instead of a benefit — the opposite of what a bank was hoping to gain.
Faster-moving data, same human judgment
Joshua Chananie, a partner and consumer products leader at SAX Advisors, said this cautious, narrow use of AI matches what he’s seeing across the industry. Using AI for specific tasks — like refreshing a credit score rather than generating the final reserve figure — “is consistent with what most banks are doing,” he said.
Banks are also using AI to pull in more up-to-date financial information when judging credit risk: recent spending patterns, current transaction history and fresh market data, rather than relying only on older credit histories. Chananie said that shift is partly a response to a string of economic shocks in recent years. “Covid disrupted materials and supply chains, then came interest rate spikes, tariffs, and ongoing military conflict,” he said — conditions that can move faster than older rating models can track.
Even so, Chananie said, AI isn’t replacing the judgment of the people who work at a bank. “The use of AI isn’t perfect and still very much relies on the banker’s input and expertise,” he said. He compared it to the way online retailers adjust prices in real time based on recent sales — banks can use fresh data in a similar way to update risk ratings, but a person with experience still has to make the final call, not a machine.
Take your tax and accounting research to the next level with Checkpoint Edge and CoCounsel. Get instant access to AI-assisted research, expert-approved answers, and cutting-edge tools like Advisory Maps and State Charts. Try it today and transform the way you work! Subscribe now and discover a smarter way to find answers.
