One point. That is the entire lead your institution still holds over a chatbot.

Financial institutions have spent two decades defending the transaction and assuming the advice was safe. Payments got unbundled. Lending got unbundled. Brokerage fees went to zero. Through all of it, the relationship held, because the one thing a bank, credit union or advisory firm could still claim was that when a customer had a real money question, they came to you.

New data says that claim is now a coin flip. J.D. Power's June 2026 financial health survey of 4,000 U.S. consumers found that 35% agreed their bank helps them make better financial decisions during affordability challenges — and 34% said the same about AI. Not AI someday. AI now, in a category where being nearly as trusted as a regulated institution is an extraordinary result for a product that has existed at consumer scale for three years.

The Parity ProblemAdoption isn't the story. Action is.

Usage numbers alone would be easy to discount as curiosity. Forty percent of U.S. consumers asked AI for help with their money in a single quarter, and the rate runs highest among two groups: consumers under 40, and consumers J.D. Power classifies as overextended — slightly over budget, carrying debt, and least able to afford a wrong answer. The most common uses are mundane and high-frequency: comparing prices before a purchase (24%), hunting coupons and deals (22%), and finding ways to stretch income or save (21%).

What converts curiosity into a competitive threat is what happens next. In J.D. Power's parallel Canadian study, 64% of Canadians used AI tools in the past year and 41% used AI to ask about their personal finances — and 73% of those who received AI financial advice acted on it, a rate the firm notes is similar to how often customers act on advice from their own bank. That is not browsing. That is a decision being made, executed, and never attributed to any institution's channel.

The context makes it sharper. Sixty-six percent of U.S. consumers were financially unhealthy in June, and 52% of Canadians were financially vulnerable or stressed. The people most likely to consult a model are the people carrying the most risk.

The Blind SpotThe advice is imperfect — and that's your exposure, not your defense

There is a comforting version of this story in which the models are simply wrong and the market corrects itself. The evidence does not support it. A 2026 working paper co-authored by MIT Sloan's Taha Choukhmane had 1,000 adults write their own prompts to an AI model for financial advice, then simulated the lifetime consequences of following it. The finding: the advice was directionally sound and, on balance, left people with more savings. "It tends to push people toward saving more, participating more in the stock market, de-risking as they get older," Choukhmane told NPR. "It gets a lot of things right."

The failures were narrower and more revealing. On complex, emotionally loaded situations — job loss, portfolio rebalancing — the model prescribed spending cuts the researchers considered too harsh and neglected to tell users to draw on emergency savings. The researchers also found it recommended riskier financial moves for men than for women. Practitioners describe the same pattern: a confident first answer that reverses completely once a professional supplies context the consumer would never think to provide.

Read that as an institution and the exposure inverts. Every one of those gaps — the missing context, the confident wrong answer, the emotional dimension of a job loss — describes precisely the value a financial brand exists to provide. And every one of them is currently being resolved, badly, in a channel the brand does not observe, cannot correct, and never receives feedback from. Forrester's warning that one in three brands will damage customer trust with rushed self-service AI in 2026 is the mirror image of the same problem: institutions racing to deploy their own bots while remaining blind to the ones their customers already prefer.

One shift, three different balance sheets

The ResponseCompete for the question, not just the customer

The instinct will be to ship a chatbot. That is the move Forrester expects a third of brands to get wrong. The institutions that hold the advice relationship will do four things first:

• Instrument the off-property journey. Before you build anything, establish where your customers actually go when a money question forms, what they ask, and what they do with the answer. Signal maps the complete journey — including the parts your own channels never record.
• Audit what the models say about you. Your machine reputation is now a distribution channel and a compliance surface simultaneously. Echo shows you how AI systems characterize your products, rates and complaints — and whether your competitor is named in your own category answer.
• Own the questions AI handles badly. Job loss, rebalancing, sequencing debt against retirement, the emotional weight of a hard year — these are the documented gaps. They are also the moments where human accountability is worth paying for. Publish, structure and market against them deliberately.
• Keep the channels that still convert. Even amid AI adoption, Canadian bank customers named marketing communications (51%) and direct representative channels (47%) as their preferred way to receive guidance, well ahead of digital channels (31%). The lesson is not to abandon the relationship — it is to stop assuming it is being maintained by default.

Bottom LineParity arrived quietly. So should your response.

No customer announced they were switching advisors. Nobody closed an account. The share of consumers who credit AI with helping them make better financial decisions simply climbed to within a point of the share who credit the institution holding their money — and it did so without generating a single data point inside anyone's CRM. That is what makes this the most under-managed risk in financial services right now: it is invisible to every system built to detect it.

Institutions that treat this as a product problem will build another bot and measure its containment rate. Institutions that treat it as an intelligence problem will first go find out where the conversation moved, what is being said inside it, and which of their customers are acting on an answer they never saw. The brands that win know something others don't — and in this cycle, what they know is where their customers go when they stop asking them.