Skip to main content
SearchLoginLogin or Signup

In Conversation With Intuit Chief Data Officer Ashok Srivastava

An interview with Ashok Srivastava by Hamit Hamutcu.
Published onJan 30, 2025
In Conversation With Intuit Chief Data Officer Ashok Srivastava
·

Column Editor’s Note: In this Active Industrial Learning column article, I interview Ashok Srivastava, Senior Vice President and Chief Data Officer at Intuit, to discuss Intuit’s strategic use of AI and data science to enhance its financial products like TurboTax and QuickBooks. Intuit has developed proprietary AI technologies, such as GenOS, to offer personalized customer experiences and streamline operations. Srivastava stresses the importance of employee education and upskilling, alongside collaborations with academic institutions, to stay ahead in AI advancements. The conversation delves into the company’s significant improvements in development speed and customer satisfaction through its AI initiatives. Srivastava also highlights the need for ethical AI practices to prevent negative consequences like misinformation.


Keywords: generative AI, education, upskilling, AI community collaboration, machine learning, customer experience.


Overview

The power of data science in the business realm is tremendous, but the hype and complexity surrounding it can be overwhelming. HDSR’s Active Industrial Learning column focuses on applications of data science to businesses, with an eye toward practical considerations that can make data science initiatives more likely to succeed. The aim is to translate business concerns into the language of data science, and vice versa, to empower data science leaders at companies of all sizes and data maturity levels.

Intuit is a global financial technology platform company based in Mountain View, California. With over 18,000 global employees and 16 offices in seven countries, its mission is to power prosperity by helping consumers and small and midmarket businesses make better, faster, and more informed financial decisions. Since its founding in Palo Alto in 1983, Intuit has developed a suite of leading tax and accounting packages, most notably TurboTax and QuickBooks.

More recently, Intuit expanded into the personal finance and email marketing space with major acquisitions of Credit Karma in 2020 and Mailchimp in 2023. Now a $16.3 billion software company, Intuit is making a large push with generative AI.

Since joining Intuit in 2017, Ashok Srivastava has focused on making a better experience for customers. As senior vice president and chief data officer, he has led the development of data platforms, analytics modernization, and AI platform and product development for Intuit’s suite of tax and business software products, often in ways that required a complete reimagining of the company’s offerings. One such project included an overhaul of the QuickBooks platform to provide each customer with their very own AI model custom-tailored to their business. This meant Intuit would need to oversee millions of independent models, all of which would be refreshed on a regular basis.

That kind of innovation, he says, can only happen when team members are familiar with what’s happening at the cutting edge of AI, generative AI, machine learning, data science, and experimental design. For that reason, Ashok has always championed employee education and upskilling, as well as a close working relationship with academia. In his capacity, he has led numerous employee training initiatives and developed curricula for team leaders to deploy AI responsibly—a natural extension given his role as an adjunct professor at Stanford University.

I sat down with Ashok to discuss his vision for AI and Intuit and how education fosters his company’s mission to power prosperity around the world.

Interview

Hamit Hamutcu: I know you’ve been experimenting with and using generative AI technologies for a long time, certainly since before it dominated the conversation with the launch of ChatGPT in November 2022. Can you tell us about some of that early work?


Ashok Srivastava: We have had a team since I started investigating new AI technologies early in my time here at Intuit. Some early work that we did that was outside of the world of generative AI was to produce confidence intervals for time series prediction—something that I’m very interested in. The reason that I start there is to say that we’re not only technology adopters, but we’re technology builders. We build new algorithms and we continue to do that because those investments are really what lay the groundwork for future success. It also attracts some of the top people in the field to work at Intuit.

Back in November 2019, when OpenAI released GPT-2—three years before ChatGPT—we were experimenting with generative AI models like BERT1 at Intuit. As the more advanced capabilities came from OpenAI’s GPT-2 and GPT-3, we also began experimenting with those models. We were particularly interested in applications of generative AI that would help us learn how to help our human tax and bookkeeping experts to better assist our customers with their questions, including learning through conversation summarizations. This would make it easier, for example, to communicate more effectively with a customer who calls back, because we have a quick summary of their previous interactions with us. We’ve found that when our customers engage with our AI/GenAI self-help or AI experiences, the number of times a customer needs to reach out for help goes down by 11%—a significant scale. It also makes it easier to identify successful interactions and use them to train our workforce.


Hamit Hamutcu: Could you talk a little bit about the most recent applications of generative AI at Intuit—either internally or customer-facing? How have those use cases evolved over time?


Ashok Srivastava: Most of the work we do is customer-facing, so I’ll discuss some of it. In last year’s tax season, TurboTax processed 44 million U.S. tax returns and $107 billion in tax refunds. Millions of customers interacted with our generative AI experiences in Intuit Assist for TurboTax. Customers would request explanations related to their tax filing, especially pertaining to their final tax refund or payment. Those conversations were powered by generative AI. The result was a massive success because customers actually liked the experience and felt they gained a better understanding of their fiscal responsibilities. The product was tailor-made for them, it was personalized, and it really provided customers with the right insights to understand and fill out their tax forms.

This is just one example of something that we have discussed publicly. It’s kind of amazing to see how far it’s gone. As you know, this is not just about taking a large language model and deploying it for a specific use case. To make this a success, we built a proprietary generative AI operating system, called GenOS, which powered this experience. GenOS has several key components:

  • GenStudio, a dedicated environment for accessing and experimenting with an extensible catalog of best-in-class commercial, open-source, and proprietary large language models.

  • GenRuntime, an intelligent layer that chooses the right large language model in real time and calls the right data access points to ensure customers receive accurate and complete responses. It includes capabilities that will enable agentic workflows in the future, beyond basic prompting to autonomous planning, reasoning, and execution to tackle complex business workflows. And, it’s been enhanced with GenSRF (security, risk and fraud), an extensible and configurable framework that includes embedded safety, privacy, and security controls.

  • GenUX, a user experience (UX) framework with 140-plus UX components, widgets, and patterns that designers and front-end developers can use to build consistent customer interfaces and flows.

  • AI Workbench, a dedicated development environment for end-to-end application development.

Each of the components of GenOS power our generative AI customer experiences and allow our engineers to develop at high velocity. In fact, our development velocity has increased by 8 times over the past 4 years, as measured by the number of production releases per developer per week. We’ve also seen up to 30% faster coding in experiments with AI assistance.


Hamit Hamutcu: I know Intuit’s collaboration with academics and the broader AI community is important for you. I know you invest a lot of time in it. Can you share your approach to this, and how the AI community can work together for effective, responsible development and deployment of AI technologies?


Ashok Srivastava: It’s very important that a company like ours, a team like mine, be very aware of what's happening at the cutting edge of AI, machine learning, predictive and prescriptive analytics, experimental design, data systems, and all the other realms that we operate in. We’re not just taking AI or machine learning off the shelf and deploying it. The way I have always maintained a deep knowledge of cutting-edge research is by working closely with academic partners.

This gives us the ability to understand what is happening at the frontier of research and how to translate that into real-world applications. I'll give you a great example, where we have some challenging time series prediction problems—for instance, predicting the cash flow for a small business. It sounds easy, but it’s actually a very difficult problem, because cash flow prediction essentially means that you’re going to make a prediction about how much money is in the bank for a small business anywhere from a week to a year to help guide financial decision-making, whether it’s payroll, inventory management, or investing in an equipment upgrade. What makes this a complex problem is that even if invoices go out to the business customers, those customers may pay them with a delay. Yet, the small business owner has its own payments to make at certain times, and they need to access a line of credit to do so. Some payments may be recurring, whereas other payments may not be. The owner thus has a series of fluctuating incoming and outgoing payments that need to be balanced. These factors make the prediction problem challenging.

We are working closely with university partners to really understand how to model these types of time series for small business cash flow forecasting. We have written many academic papers on the subject (Han et al., 2021, 2022, 2023), in collaboration with the University of Texas at Austin, for example. Together, we developed a novel approach to cluster hierarchical time series (HTS) for efficient forecasting and data analysis. Inspired by a practically important but unstudied problem, we found that leveraging local information when clustering HTS leads to a better performance. The clustering procedure we proposed can cope with massive HTS with arbitrary lengths and structures. In addition to providing better insights, this method can speed up the forecasting process for a large number of hierarchical time series. We empirically show that our method substantially improves performance for large-scale clustering and forecasting tasks involving HTS. What is cool is that these kinds of modeling techniques actually make it into our QuickBooks product, and they help us make better forecasts.


Hamit Hamutcu: I can imagine it’s challenging, especially when you try to do it for millions of small businesses, so it’s also a scale issue. In a talk last year at a conference organized by the Institute for Experiential AI at Northeastern University, you mentioned that Intuit generates 60 billion machine learning predictions every day. Can you tell us about the key technologies, processes, and talent that you had to invest in to create that kind of scale?


Ashok Srivastava: I’ll start with the people. You need to have people who are skilled in the craft of AI, who have gone beyond just using tools, who are willing to make substantive changes in the platform and the data and the analytics in order to use them at such a scale. You know, 60 billion predictions a day is a big number, but what really matters is how the customers benefit from those predictions. The team that we’ve established is one that really invests in thinking about the customer benefit and then driving that benefit through, in this case, the use of machine learning predictions. I really want to have teams of people who are at the foundations of AI, who are building new deep learning models, who are building statistical models, and who know how to use tools and techniques and software that others have written to deploy them at scale.

On the process side, we have to be willing to change the way we’ve done things in the past. My organization is focused on driving business outcomes with analytics, AI, and data. Our analytics team provides critical business insights that shape our product and go-to-market strategies, and the data team helps accelerate those insights through robust data capabilities and services. And finally, our AI teams leverage both data and insights to create innovative and delightful ‘done for you’ experiences for our customers.

It’s great, but what’s even better is that we’re enabling teams outside of my core group to do AI. This group will always be pushing the foundations of AI, will always be working on the most critical use cases, but then you need thousands of developers outside of my organization and closer to products who should leverage what we can offer. So we’re investing in helping them have the best platform technologies, have the right training, the right skills, and the awareness of responsible AI so that they can go and do those deployments on their own.

The third thing is the technology investment. One example I’ll give you that's incredibly exciting is in the area of accounting automation. This is the heart of QuickBooks, where billions of transactions come in for our customers, and they’re categorized into different categories that are user-defined, rather than determined by Intuit. For example, if a customer cares about spending related to ‘restaurants,’ they can create that category. Another customer may instead be interested in ‘travel and entertainment’ expenditures.

Now, it turns out that making classifiers to do this is really hard because the categories are user-defined. We’ve literally reinvented our platform so that each person can have their own model that allocates spending to the user-defined categories. This means we have millions of models in operation at any given point in time, and those millions of models are getting updated regularly and frequently, because people’s preferences change and the transactions that come in are different.


Hamit Hamutcu: You said you have your own immediate organization, but then you’re also enabling others outside of that, and you gave the example of developers. But when you look at the rest of the organization—not just developers, but customer service, operations, HR, marketing, and many more—what’s your perspective on data and AI literacy? Are there any examples that you can cite as to how Intuit is approaching that?


Ashok Srivastava: This is such an important question, in my opinion, because the reinvention of the company happens with the people. What are we other than a collection of people that is galvanized under the single mission to power prosperity around the world? That’s what Intuit is. In order for us to reinvent ourselves, we have to reinvent our mindset, we have to reinvent what we know, we have to infuse our knowledge with new ideas and new ways of doing things.

One of the first actions that I took when I came here was to start an AI training program, and that training program has evolved substantially. Now we’re building training programs for everyone at the company so that they can know about the latest in AI and data, and they can experience it themselves, because while learning is good, doing is more powerful. For example, we built a sandbox called GenStudio, which is one of the components of our proprietary GenOS that enables any Intuit employee to interact with generative AI large language models in a safe and secure manner, with the right safeguards in place. It’s amazing to see how employees are using this capability. Program managers, product managers, executive assistants, people who do filming in our corporate communications department—all kinds of people use it in all kinds of ways. For developers, it’s an environment for rapidly experimenting with and refining generative AI experiences for users. For marketing and communications folks, it’s a tool for streamlining workflows and augmenting talent for a range of tasks—brainstorming, creating content, analyzing large amounts of data to identify trends, and more.

Our companywide training, coupled with the ability to use the technologies directly in a safe and secure manner, is what really makes this effective.

Education and upskilling are extremely important. The investment that we make in our employees, so that they learn the very best ways to use AI and data, is key to our success. We have ambitious plans for the company in this area. Recently, I wrote the curriculum for AI for product managers within the company, and I taught it myself globally to all our product managers. It received a Net Promoter Score of 100—the highest you can get.

The curriculum I developed was not overly technical but highlighted the key knowledge that we want our managers to have when it comes to deploying AI at Intuit. For example, does everyone need to know the intricacies of convex optimization and deep learning? No. But should they be aware of the ethical aspects? Yes. Should they have the opportunity to experiment themselves in a safe and secure environment? Absolutely. These are the types of things that we’re investing in, and it’s amazing because it brings in our partners from the legal team, the IT department, and the public relations team—so many teams come together in order to make this happen.


Hamit Hamutcu: I’m really glad I asked that question, and I feel like there might be a case study sometime next year. Personally, I’d love to look at that curriculum whenever you think it can be shared outside of the company, because I’m very interested in the definitions that you’ve made and the literacy requirements—what do we expect people to know in terms of that common language?

Looking ahead to the next year or two, what are you most excited about and what worries you most about advancements in AI?


Ashok Srivastava: Here’s what I’m excited about: We are uniquely positioned to help small businesses succeed, to make midmarket businesses succeed, to make consumers succeed from a financial perspective. We have the data, we have the artificial intelligence, and, most importantly, we have the team who can bring this to fruition.

As you look at what’s happening in society, there are a large number of people who are struggling to make ends meet. You know, 40% of Americans don’t have $400 readily available for an emergency, according to new Empower research (The Currency editors, 2024). We have an amazing opportunity to help drive better financial decisions.

You also asked what I’m concerned about. As you look in mainstream media and business coverage, as you read blogs, and as you listen to people from different vantage points, there is a concern that AI can have negative consequences, that it can be used for generating misinformation, that it can give people a false sense of what’s true and what’s not, it can give them lack of confidence in the way society is operating. This is my number one concern.

Humanity has evolved toward favoring the scientific method and empirical evidence. The use of that structure is what gives us the foundation of science. When society no longer can differentiate fact from fiction, that’s when things start to unravel. Unfortunately, AI can be used at scale by people with nefarious objectives, by people who want to attack the very foundation of society. So, it’s critical that we maintain the ability to differentiate fact from fiction, that we understand the value of peer review, that we understand the value of science and the scientific method, so that we don’t fall prey to those that want to harm us.


Hamit Hamutcu: Thank you. That was beautifully put. Is there anything else that you’d like to add?


Ashok Srivastava: Thank you, Hamit, for allowing me to discuss where we are as a company and where we’re going, and for emphasizing that our investments in people, data, and AI are foundational, they’re oriented toward a platform that can deliver ultimate customer benefits at enormous scale. The statistic I’ll give you is that our developer velocity has increased by a factor of 8 due to the platform investments that we’ve made.


Hamit Hamutcu: Can you clarify what you mean by velocity?


Ashok Srivastava: We have a very rigorous process of measuring development velocity. That means the number of releases that an engineer creates per week, and we’ve increased that by a factor of 8 due to platform investments. This doesn’t mean that we push our developers to type faster. What it really means is that we’ve created a platform that enables them to get their work done faster, and that we have created processes to eliminate barriers that slow them down.


Hamit Hamutcu: Thank you, Ashok!

Conclusion

With AI, Intuit is aiming to drive a new age of innovation on its platform for consumer and small business finance. Core to its innovative forecast, however, is education—both in terms of employee upskilling as well as academic partnerships. Intuit sees these connections as core to the fundamental promise of AI—that gains in efficiency, productivity, and knowledge turn on how well companies identify and take advantage of the things people do best.


Disclosure Statement

Ashok Srivastava and Hamit Hamutcu have no financial or nonfinancial disclosures to share for this article.


References

The Currency editors. (2024, April). Over 1 in 5 Americans have no emergency savings. Empower. https://www.empower.com/the-currency/money/over-1-in-5-americans-have-no-emergency-savings-research

Han, X., Dasgupta, S., & Ghosh, J. (2021). Simultaneously reconciled quantile forecasting of hierarchically related time series. ArXiv. https://doi.org/10.48550/arXiv.2102.12612

Han, X., Hu, J., & Ghosh, J. (2022). Dynamic combination of heterogeneous models for hierarchical time series. In K. S. Candan, T. N. Dinh, M. T. Thai, & T. Washio (Eds.), 2022 IEEE International Conference on Data Mining Workshops (pp. 1207–1216). IEEE. https://doi.org/10.1109/ICDMW58026.2022.00157

Han, X., Ren, T., Hu, J., Ghosh, J., & Ho, N. (2023). Efficient forecasting of large-scale hierarchical time series via multilevel clustering. Engineering Proceedings39(1), Article 31. https://doi.org/10.3390/engproc2023039031


©2025 Ashok Srivastava and Hamit Hamutcu. This article is licensed under a Creative Commons Attribution (CC BY 4.0) International license, except where otherwise indicated with respect to particular material included in the interview.

Comments
0
comment
No comments here
Why not start the discussion?