10 ways Intuit is revolutionizing personalization with Generative AI

Intuit’s generative AI strategy reflects the company’s vision of helping consumers and small businesses be more competitive and confident in every financial decision they make. Core to that vision is a heavy reliance on gen AI to personalize customers’ experiences using Intuit products, automate routine work and provide new insights to help customers achieve their financial goals. The company has more than 100 million customers worldwide across its TurboTax, Mint, Credit Karma, QuickBooks and Mailchimp products.

Intuit’s AI-driven platform strategy is working, as evidenced by double-digit growth in both total revenue and operating income. Total revenue for Q3 fiscal 2024, which ended on April 30, grew to $6.7 billion, a 12% increase year-over-year, and the company’s Small Business and Self-Employed Group revenue surged by 18% to $2.4 billion.

Intuit on how they are revolutionizing personalization with GenAI

During a fascinating interview at the VB Transform 2024, Nhung Ho, VP of AI at Intuit, discussed with Matt Marshall, VentureBeat CEO and editor-in-chief, how Intuit continues to refine its approach to using gen AI, AI and machine learning (ML) modeling to fine-tune its personalization strategies by customer segment, product and development strategy.

Ho provided insights into the practical challenges along the way and how Intuit’s dual goals of simplifying personal finance while enabling small business customers to be more efficient and informed using AI-based automation.

Intuit is focused on best democratizing AI across all of its applications, giving small businesses the ability to have real-time data and greater insights so they can keep more of what they earn while finding new revenue opportunities. The company launched Intuit Assist in September 2023. It is its first gen AI-powered financial assistant for small businesses and consumers. The tool is integrated with and capitalizes on Intuit’s proprietary Generative AI Operating System (GenOS). It’s designed to deliver personalized financial insights and recommendations across products, including TurboTax, QuickBooks, Credit Karma and Mailchimp.

Ashok Srivastava, Intuit’s chief data officer, told VentureBeat in an extensive interview last year that the GenOS infrastructure could be likened to an operating system for gen AI: “Think of a real operating system, like MacOS or Windows,” he said, referring to assistant, management and monitoring capabilities. Similarly, large language models (LLMs) need a way to coordinate their actions and access the resources they need. “I think this is a revolutionary idea,” Srivastava said.

Gen AI turbo-charging TurboTax

Ho told the audience at transform that “we’ve had a lot of time over the past year to experiment and put AI applications within our multitude of products. But the one that I want to talk about is the one that we saw the most scale, which is in TurboTax. This past tax season, 24 million customers got a generative AI experience that helped them fully unpack how we calculated their taxes and how we got here.”

“I mean, we all do our taxes, hopefully by the end of the year, and for many people, including myself, we get to the end, and you’re wondering, did I do this correctly?” she continued. “How did I get here? What got me here? And in the past, we didn’t have a good way to unpack and translate a huge series of numbers and calculations in a way that was digestible to a common person like me.”

Ho described how Intuit relies on Anthropic’s LLMs to provide accuracy. “What we did was we married the power of our tax engine, which guarantees accuracy, with an LLM, specifically Claude from Anthropic, to help us explain how we did that calculation,” she said. “And that was the magic that we were able to provide for our customers this last season. I think it was a really wonderful moment to feel the confidence like, ‘Hey, I know what’s going on. I don’t have to be scared. I don’t have to be worried that I did something wrong. And I feel like I’m a part of this and kind of taking power back in the scanning process.’”

Intuit has fine-tuned its approach to using gen AI to deliver greater personalization based on unique customer requirements. Ho explained to the Transform audience, “If you’re an example in the small business space, personalization means that we are using your data to modify our product and tune it to be where you’re at. If you’re a construction company versus a bakery, the way you set up your business is drastically different. And the set of services using us is drastically different. When you log in and you sign up our product, and you tell us what your condition is, we immediately start personalizing it so that you’re only using the parts that are most relevant to you rather than everything else.”

Based on the insights shared during the discussion, it’s apparent that Intuit relies on a core set of actions to help the company excel at personalization across all products using gen AI. These include:

Integrating AI to simplify complex tasks: Intuit relies on gen AI to break down complex tasks into understandable steps for users, as evidenced by TurboTax’s AI-driven explanations that simplify tax calculations for 24 million customers.

Automating routine operations: By integrating gen AI into every app to automate routine tasks, Intuit looks to help its customers’ employees focus more time on their core activities, which often include customer conversations. A prime example of this is how their AI tools automate accounting and marketing workflows for small businesses.

Ensuring high-quality data: Maintaining high-quality, clean data is crucial for AI applications to function effectively, according to Ho. Intuit continuously monitors and curates data to ensure accuracy in AI-driven processes. “High-quality data is what gives you that differentiated advantage over everyone else. Do you have the right data? Is that data clean? Is it actually correct?” asks Ho.

Manage latency in AI systems: It is essential to get ahead of latency bottlenecks in AI apps to excel at delivering user experiences at scale. Intuit is exploring and currently using AI apps beyond chat-based interfaces to reduce latency sensitivity. “These systems are slow. We live in an era where patience is kind of low, and we’re used to having things operate in the millisecond realm. In the generative AI world, we’re talking about tens of seconds,” Ho observed.

Adopting a multi-cloud strategy: By relying on a multi-cloud approach, Intuit is able to stay flexible and access the best AI models for a specific personalization use case. This strategy ensures that they can leverage the best tools available regardless of the cloud provider. “We chose to be fully agnostic… it’s just an API, it doesn’t matter… whichever the cloud provider builds the best ecosystem for this is the one that to me will win,” Ho advised.

Build scalable and secure AI systems: Intuit is doubling down on data scalability as a core to its AI security strategies. Hardening AI systems against attack also sets a foundation for more efficient innovation cycles. Ho says, “We wanted to allow everybody to innovate, but to do that, you have to do it safely. You have to ensure that your data is safe and secure to build a system that guarantees that.”

Create A composable architecture: Designing systems with fully composable microservices provides flexibility. Intuit uses a composable architecture to integrate various AI models and technologies. “We built a set of microservices and capabilities that are fully composable, which means that you can take and choose which piece you want to come together to create a new application,” Ho explained.

Personalize user experiences: Fine-tuning and personalizing AI apps and tools to meet the specific needs of different user segments is key. Intuit customizes AI tools for every type of small business in its customer base, with Ho referencing the wide variation in needs between bakeries and construction companies, for example.

Balance deterministic and probabilistic systems: Determining which AI systems should be deterministic and which can be probabilistic ensures accuracy, according to Ho. Intuit uses deterministic systems for critical tasks like tax calculations. “We separate the engine that does the calculations of your taxes. It’s deterministic. We would never put a probabilistic system to do your taxes,” Ho assured the audience after an attendee asked how accurate the Intuit AI models are when completing customers’ taxes.

Have a roadmap that expands AI application use cases: Push the boundaries of AI applications beyond initial use cases to further enhance adjacent, related business processes, Ho advised. Intuit applies AI to workflow augmentation and summarization tools to improve efficiency. “There are many applications for generative AI beyond just chat… think about every single part of your workflow right now and for your customers where you can apply AI to really provide that benefit of doing work for them,” Ho recommended.

Conclusion

Intuit’s advanced approach to using gen AI for personalization stands out in how models are used to further refine preferences and customer experiences by product and within a given customer segment. Their launch of Intuit Assist and GenOS reflects how far advanced they are in using gen AI to augment and improve customer experiences.