Intuit recently launched its new generative AI-powered assistant called Intuit Assist that provides personalized and contextual guidance for its 100 million customers across its entire product line of TurboTax, QuickBooks, Credit Karma, and Mailchimp. This is a major enhancement resulting from a decade of work by Intuit scientists to harness machine learning and AI to improve its offerings for consumers and small businesses.
AI needs data
For AI to be successful requires vast amounts of data. Intuit can check that box because not only does it have a staggering amount of business data, it also has plenty of AI experience.
Unlike many companies that have only begun to experiment with AI in the past six to twelve months, Intuit has spent a decade developing AI to innovate its products and streamline operations. The company already has 2 million AI models in production and 65 billion machine learning predictions per day.
On the data front, Intuit has 500,000 customer and financial attributes per small business and 60,000 tax and financial attributes per consumer. In addition, it has connections with more than 24,000 financial institutions.
These massive datasets give Intuit a complete and highly accurate 360 degree view of its customers and their businesses. All this information allows Intuit’s machine learning and generative AI models to deliver intelligent insights and personalized user recommendations about customers’ income, expenses, profitability and cash flows.
Intuit uses general and custom-trained financial-specific large language models (LLMs) for tax, accounting, marketing, cash flow, and personal finance. The system provides AI assistance for Intuit’s entire product line of TurboTax, QuickBooks, Credit Karma and Mailchimp—and, in some cases, across multiple products simultaneously.
Intuit Assist will be different from earlier systems that had limited interactions with users. Intuit Assist uses sophisticated AI to walk customers through complex workflows, provide personalized insights, act as a financial assistant on the user’s side, and even do many types of work for them if they choose.
Recent advances in AI have been driven by large language models (LLMs) such as OpenAI’s GPT-4. LLMs are trained on massive amounts of data scraped from the internet. Because the data is generalized, these models have limited capabilities focused on understanding language, reasoning, and search. Even though these out-of-the-box LLMs are not specialized for any particular task, Intuit used them as starting points for creating specialized models for its different businesses. While Intuit leverages these base models for experiences, the company doesn’t use its data to train or enhance these models. Intuit uses an AI technique called fine-tuning to train LLMs to create AI models that have specialized knowledge about specific subjects. Highly accurate generative AI models can be created from fine-tuned LLMs.
In addition to its own proprietary LLM, Intuit has created some generative AI models by fine-tuning general-purpose LLMs obtained from infrastructure companies such as OpenAI. Intuit uses its own data to fine-tune purpose-built LLMs to help solve tax accounting, cash flow, marketing and personal finance challenges.
Intuit Assist is an AI-powered financial assistant that provides users with highly personalized and intelligent financial insights and recommendations. It is being integrated into TurboTax, Credit Karma, QuickBooks, and Mailchimp.
The combination of customer data and generative AI is a key factor that will enable Intuit Assist to provide users with highly personalized and accurate recommendations.
Intuit Assist’s flexibility will give users the choice of many options. They can use it as a guide or it can assist if an action is needed. And, if a user isn’t comfortable with those options, they can click a button to talk to a human tax or bookkeeping expert in TurboTax or QuickBooks. If a user chooses, Intuit Assist can even take over and do everything for the user.
Intuit’s GenOS innovation
According to Alex Balazs, Intuit’s chief technology officer, Intuit Assist models are all built on Intuit’s generative AI operating system, GenOS.
Intuit’s overall platform contains over 60 petabytes of data and imports almost 20 billion transactions from financial institutions annually. For context, 60 petabytes of data is approximately 60 billion books. This data, proprietary to Intuit, is scrubbed and organized, then used to personalize customer AI interactions.
“These additions to our powerful data platform enable us to do the hard work for our customers intelligently,” Balazs said. “We have a modern, fully cloud-native platform that allows our developers to move fast and deliver revolutionary benefits. For our 100 million customers, we have increased our velocity by nine times in the last three years.
“GenOS can further accelerate our speed with its new capabilities, such as translating tax rules into code up to six times faster,” Balazs added. “We have hundreds of capabilities on our platform just with our AI capabilities alone.”
GenOS is comprised of four core generative AI components:
- GenRuntime: An intelligent layer that chooses the right LLM in real time and calls the right data access points to ensure that users receive accurate and complete responses according to their specific needs and situations.
- GenStudio: A dedicated development environment allowing Intuit’s developers to experiment with and refine generative AI experiences for users rapidly.
- GenUX: A library of user-interface components and consistent user flows that ensure a precise, enjoyable and transparent experience for interacting with the LLMs.
- Financial large language models: Custom-trained financial LLMs that specialize in addressing tax, accounting, marketing, cash flow and personal finance questions. These can provide actionable insights and invoke actions such as contacting a human expert when needed.
To say that Intuit Assist is impressive is an understatement. Intuit Assist is robust, sophisticated, and “smart.”
Intuit Assist understands context, anticipates needs, and provides tailored recommendations through the hybrid use of general LLMs and financial domain-specific AI models.
Intuit Assist is in the hands of tens of thousands of customers now as part of beta testing programs and will become available more widely on Intuit’s platform in the coming months.
After personally experiencing a number of new and weak AI products, it is refreshing to see Intuit’s example of how generative AI should be used to increase productivity and make life easier for users.
It is important to point out that Intuit has implemented responsible AI principles and practices to guide how it operates and scales its AI-driven platform. Intuit has taken the necessary precautions to safeguard private information in its applications. That is important for any business handling its customers’ data, but even more so for a company that has spent decades building trust with users as it handles all the details of their financial data.
During Intuit’s 2023 fourth-quarter earnings call, Intuit’s CEO, Sasan Goodarzi, expressed his optimism about the company’s future products and advancements enhanced by AI.
“With our durable AI-driven expert platform strategy and focus on innovating with GenAI across our products, we are moving at high velocity. This will help us put more money in our customers’ pockets, save them time, and ensure complete confidence in every financial decision they make.”
Intuit has a solid and forward-looking management team that understands and appreciates the value and yet-to-be-released future potential of AI. Based on conversations with Intuit’s key executives, it is the opinion of Moor Insights & Strategy that the company is well-positioned to continue to deliver future innovations, new capabilities and unique user experiences. Even though competition is likely to become heavily invested in AI, by remaining focused on AI innovation, Intuit should be able to maintain its lead by delivering future products and tools that will be difficult for other companies to match.