Intuit Wins with AI but Loses on the Market: Anatomy of a Paradox
In mid-2025, Intuit reported a figure any CFO would endorse without hesitation: $90 million in annualized efficiencies stemming from its investments in artificial intelligence, achieved in just the first six months of the year. Simultaneously, its stock had dropped nearly 50% from recent highs, with some investors reporting losses of 86% on expanded positions. Two realities coexisting in the same balance sheet, in the same company, within the same quarter.
This is not a case of fraud or mismanagement. It is something more interesting and more challenging to resolve: it is the story of a company that learned to operate at two different speeds and a market that still does not know how to value that.
The Engine No One Disputes
Intuit processes $105 billion in tax refunds each year, automating 93% of the tax forms on its platform. TurboTax Live grew 47% in 2025, forcing the company to plan physical stores for 2026. It boasts 100 million customers across individuals and businesses, on which it has built its proprietary generative AI operating system.
None of this happened by accident. CEO Sasan Goodarzi initiated the strategic pivot towards AI in 2019. Three years later came the virtual expert platform, which uses AI to interpret financial documents and direct complex cases to human specialists. In November 2024, Intuit Assist launched for QuickBooks, featuring generative AI capabilities for processing tax filings. By January 2026, the company introduced its agentic AI platform, integrating Credit Karma and TurboTax, with automation of 90% of 1040 and 1099 forms, saving users a total of 6 million hours of manual work in the last tax season.
The number that intrigues me the most from this inventory is: 13,000 human experts integrated into the platform. Intuit did not bet on eliminating professionals; it aimed to make them more valuable when customers need them the most. This architecture of AI with human intelligence is, from an operational design standpoint, considerably harder to replicate than any language model.
Two Businesses Within One, and the Challenge of Measuring Them Equally
Here lies the crux of the problem. Intuit is not one company. It is, at least, two.
The first is a mature, highly profitable business: TurboTax and QuickBooks dominate their categories with decades of competitive advantage, massive installed bases, and predictable renewal prices. This segment generates the cash that funds everything else. It deserves to be measured with classic profitability metrics: operating margin, customer retention, revenue growth per user.
The second is a set of transformation bets: the agentic platform, the integration of Credit Karma as an annual financial engagement engine, the expansion into the mid-market with a customizable Enterprise Suite by verticals like construction, and the AI-scaled network of experts. This second business requires sustained investment, tolerance for adoption friction, and completely different metrics: learning speed, feature retention rates, and acquisition costs for new segments.
During the period known as the SaaSpocalypse, the market made no distinction between the two. It applied the same blanket discount to all software-as-a-service companies exposed to AI based on the logic that general-purpose language models could erode the value of any vertical application. The result was a massive sell-off that did not discriminate between companies that are truly being disintermediated and those that have spent years building the infrastructure that would make disintermediation difficult.
Intuit falls into the second category. Its competitive moat does not lie in having good tax software. It lies in having 100 million longitudinal financial profiles on which to train its models, plus a network of human experts that turns complexity into trust. That cannot be replicated with a subscription to a generic AI model.
The Real Risk That Technological Optimism Hides
Having said that, it would be a mistake to read this article as an uncritical defense of Intuit. There are legitimate frictions that deserve attention.
CTO Alex Balazs was explicit in January 2026, during an AWS event, about the stumbles in launching Intuit Assist and reimagining QuickBooks. Publicly acknowledging mistakes is a good cultural indicator, but errors in launching generative AI products also have concrete operational costs: customers who do not adopt, teams that have to redo work, extended development cycles. The $90 million in efficiencies are real, but the costs of post-launch adjustments do not appear in the same headline.
There is another, more structural risk vector. If autonomous AI agents mature enough for a user to submit their tax return without any kind of interface, the value of TurboTax as a user experience may shrink. Intuit knows this. Hence, the move towards physical stores in 2026 is not nostalgia for analog; it is a bet that TurboTax Live’s 47% increase indicates that as customer financial complexity grows, the demand for human support also increases. The physical touchpoint would be the premium end of that curve.
The strategic question that Goodarzi will have to answer in the next 18 months is not technological. It is about portfolio design: how much capital and how much organizational attention to allocate to protect and monetize the mature business, and how much to free up for the new agentic platforms to operate with sufficient autonomy to compete against AI-native alternatives that do not carry the legacy of a desktop product from the nineties.
When Execution Exceeds Market Recognition
The appointment of Sanjay Srivastava as AI director in July 2025 is a governance signal worth noting. When a company creates that role reporting directly to the CEO, it is not engaging in technological PR: it recognizes that AI has ceased to be a product feature and has become a function of corporate governance. AI decides how human experts are assigned, how forms are prioritized, and how the experience of 100 million customers is customized. This requires someone to be accountable for those decisions at the highest level.
The paradox of Intuit is not that it has failed in AI. It is that it built measurable operational results while the market applied the same discount to it as to companies that have nothing similar. The $90 million in efficiencies in just half a year is not an analyst's promise. It is a cost reduction already executed on an operational base that processes over one hundred billion in refunds.
Companies that learn to manage both a mature highly profitable business and multiple transformation bets measured with different metrics have, historically, better odds of surviving technological panic cycles than those that only know how to do one thing at a time. Intuit, with its included frictions, has the portfolio architecture to prove it.
