.jpg)
When an AI chatbot hallucinates, it’s expected: After one large language model misinterpreted an old joke and advised adding glue to pizza sauce, the internet laughed and moved on. But when an AI handling someone’s finances gets a number wrong, there are real consequences: missed payroll, or an incorrect tax return, or a business decision based on a guess instead of the company’s actual books. This is the tension that makes designing AI for fintech a fundamentally different (and fascinating) engineering challenge.
Intuit Intelligence was built to bridge that gap. It’s the system built into Intuit products that serve businesses of all sizes, from startups to mid-market, to the accounting firms that serve them. It combines AI, a customer’s actual financial data, and access to human experts in one place. For a user, it’s the difference between digging through reports and spreadsheets, and simply asking a question and getting an answer grounded in real numbers. Building a tool like this means earning a particular kind of trust: “Behind every transaction in our system is someone who stayed up too late worrying about their business,” says Maaron Bea, a senior product manager for AI at Intuit. “Making sure the AI is worthy of that trust is how we think about every design decision.”
To build financial intelligence that customers will actually trust, Bea’s colleagues focused on three core design principles: grounding every answer in real data rather than AI-generated guesses, showing the reasoning behind every recommendation, and knowing when to hand off to a human expert.
The first is a fundamental architecture choice to favor data retrieval over general information generation. For financial queries, Intuit Intelligence pulls a customer’s actual data. When it tells you your revenue is up 12 percent, that number came straight from your books. “You’re not generating plausible-sounding answers,” Bea says. “You’re querying actual data and translating it into plain language.” The knowledge base that makes this possible is not something any company can replicate by scraping the web. Intuit has spent more than 40 years helping people manage their affairs across TurboTax, Credit Karma, QuickBooks, Mailchimp, and Intuit Enterprise Suite. That history provides a deep understanding of how real financial workflows actually look, across industries, business sizes, and stages of growth. “Companies starting from scratch with web data can build impressive general-purpose AI, but they’re essentially trying to learn financial expertise from the outside in,” Bea says. “We’ve been building it from the inside out for decades.”
The second is showing the work: When Intuit Intelligence categorizes a transaction, it surfaces the reasoning and the data points behind the result, offering users the ability to review or correct it. “Closing that trust loop, showing the ‘why,’ is what lets people actually rely on the system,” Bea says. A customer can see that a charge was booked as a contractor expense rather than a capital expenditure, understand why, and easily correct it if the system got it wrong.
The third is knowing when to step aside. One of Intuit’s core benefits is building customer confidence, and one of the key ways they deliver this is by connecting customers to human experts when needed. When a customer hits a roadblock, signals are sent to the Intuit Intelligence platform to connect them to an expert, whether through a chat experience or a phone call, to ensure Intuit always has their back at all times and the customer is managing their books or taxes with confidence. “There’s always a safety net,” Bea says, “and that’s our edge.”
But accuracy is only the beginning—the next frontier for Intuit’s financial intelligence is autonomy. The industry conversation around AI agents tends toward a simple model: one agent, one job. Your accounting agent, your payroll agent, each in its own lane. Real business problems don’t work that way. “Think of it less like assigning one employee to one job,” Bea says, “and more like assembling the right team for the right problem in real time.” A question like “Can I afford to hire someone next quarter?” touches cash flow, payroll, tax obligations, and outstanding invoices simultaneously. Intuit is moving toward an architecture where specialized tools and skills assemble dynamically—the intelligence layer determines which capabilities a problem requires and orchestrates them into a single, coherent answer. The customer still sees what went into it and approves before anything is finalized.
What could that look like? A Monday morning when a user opens QuickBooks and finds the system has already reviewed the weekend's transactions, flagged an anomaly in a vendor invoice, drafted payroll for approval, and identified that they are trending behind on a quarterly target—with specific, data-backed suggestions for what to do about it. “The AI gets smarter about how to combine its capabilities,” Bea says. “The customer never gets less informed about what it’s doing.” For Intuit customers, this means precious time back to focus on what they love about running their business.
As that intelligence becomes more autonomous, the architecture of trust supporting it has to get stronger. And for the engineers building it, that’s the challenge worth showing up for. “When someone hands you their financial data, they’re not just giving you a data set. They’re giving you their rent check, their kid’s tuition, their shot at making payroll on Friday. And that’s what makes you show up every day ready to get it right.”
