Creating AI for financial software requires a different playbook than consumer AI Intuition the latest version of QuickBooks includes an example.
The company announced Intuit Intelligence, a system that coordinates specialized AI agents within the QuickBooks platform to handle tasks including tax compliance and payroll processing. These modern agents extend existing accounting and project management agents (which have also been updated), as well as a unified interface that allows users to query data in QuickBooks, third-party systems, and file uploads using natural language.
The modern solution follows years of investment and refinement of Intuit solutions GenOSenabling the company to build AI capabilities that reduce delays and improve accuracy.
But the real news isn’t what Intuit built – but how they built it and why their design decisions will make AI more useful. The company’s latest AI deployment represents an evolution based on hard-won lessons about what works and what doesn’t when implementing AI in financial contexts.
What the company learned was sobering: Even when its accounting agent improved its transaction categorization accuracy by an average of 20 percentage points, it still received complaints about errors.
“The use cases we’re trying to solve for customers include tax and finance; if you make a mistake in this world, you’ll lose customer trust in buckets and we’ll only get it back in buckets,” Joe Preston, Intuit’s vice president of product and design, told VentureBeat.
Trust architecture: query real data instead of generative answers
Intuit’s technical strategy focuses on a fundamental design decision. For financial and business intelligence queries, the system queries real data rather than generating answers through huge language models (LLM).
ANDis also extremely significant: this data is not in one place. Intuit’s technical implementation enables QuickBooks to ingest data from a wide variety of sources: native Intuit data, OAuth-connected third-party systems such as Square for Payments, and user-uploaded files such as spreadsheets containing supplier price lists or marketing campaign data. This creates a unified data layer that AI agents can reliably query.
“We’re actually checking your real details,” Preston explained. “It’s completely different than if you just copy-pasted a spreadsheet or PDF and pasted it into ChatGPT.”
This architecture choice means that the Intuit Intelligence system acts more as an orchestration layer. It is a natural language interface for operations on structured data. When a user asks about expected profitability or wants to perform HR and payroll services, the system translates the query in natural language into database operations based on verified financial data.
This matters because Intuit’s internal research has revealed widespread shadow exploit of artificial intelligence. In the survey, 25% of accountants using QuickBooks said they were already copying and pasting data into ChatGPT or Google Gemini for analysis.
Intuit’s approach treats AI as a query translation and orchestration engine, not a content generator. This reduces the risk of hallucinations that plague AI implementations in financial contexts.
Explainability as a design requirement, not an afterthought
Beyond the technical architecture, Intuit has made explainability a key user experience in its AI agents. This goes beyond just giving the right answers: it means showing users the reasoning behind automated decisions.
When an Intuit accounting agent categorizes a transaction, it doesn’t just display the result; shows the justification. This isn’t marketing copy about explainable AI, it’s an actual user interface displaying data points and logic.
“It’s about closing that trust loop and making sure customers understand why,” Alistair Simpson, vice president of design at Intuit, told VentureBeat.
This becomes especially significant when you consider Intuit’s user research: While half of compact businesses describe AI as helpful, almost a quarter haven’t used it at all. The explanation layer serves both populations: it builds the confidence of novices while providing context for experienced users to verify accuracy.
The design also forces human control at critical decision points. This approach goes beyond the interface. Intuit connects users directly with human experts embedded in the same workflows when automation reaches its limits or when users need validation.
Navigate the transition from forms to conversations
One of Intuit’s more captivating challenges is managing a fundamental change in user interfaces. Preston described it as one foot in the past and one foot in the future.
“It’s not just an Intuit thing, it’s the entire market,” Preston said. “Today we still have a lot of customers filling out forms and looking at tables full of data. We’re investing a lot in resisting and challenging the way we currently do it in our products, where you just fill out form after form or table after table because we see where the world is going, which is a really different form of interacting with these products.”
This creates a challenge in product design: how to serve users who are comfortable with customary interfaces while gradually introducing conversational and agent-based capabilities?
Intuit’s approach was to embed AI agents directly into existing workflows. This means you don’t have to force users to adopt entirely modern interaction patterns. The payment agent appears next to invoicing processes; an accounting agent streamlines the existing reconciliation process rather than replacing it. This incremental approach allows users to experience the benefits of AI without having to abandon familiar processes.
What enterprise AI developers can learn from Intuit’s approach
Intuit’s experience implementing AI in a financial context reveals several principles that are broadly applicable to enterprise AI initiatives.
Architecture matters for trust: In domains where accuracy is critical, consider whether you need content generation or data query translation. Intuit’s decision to treat AI as an orchestration and natural language interface layer dramatically reduces the risk of hallucinations and avoids using AI as a generative system.
Explainability must be designed, not bolted on: Showing users why the AI made a decision is not optional when trust is at stake. This requires thoughtful UX design. This may limit your model selection.
User control helps you maintain confidence while improving accuracy: Intuit accounting agent improved categorization accuracy by 20 percentage points. However, maintaining user override capability was necessary to implementation.
Gradual transition from familiar interfaces: Don’t force users to abandon forms in favor of conversations. First, embed AI capabilities into existing workflows. Let users experience the benefits before asking them to change their behavior.
Be candid about what is reactive and what is proactive: Current AI agents primarily respond to prompts and automate defined tasks. True proactive intelligence that makes strategic recommendations without prompts remains an evolving possibility.
Solve employee problems with tools, not just messaging: If AI is intended to support rather than replace employees, provide employees with AI tools. Show them how to exploit this technology.
For enterprises implementing AI, the Intuit journey offers clear guidance. The winning approach values credibility over demonstration of capability. In fields where errors have real consequences, this means investing in accuracy, transparency, and human oversight before pursuing the sophistication of conversational or autonomous operation.
Simpson puts the challenge succinctly: “We didn’t want it to be a bolt-on layer. We wanted customers to be in their natural workflow, and agents doing work for customers to be embedded in the workflow.”