Intuit is a 35-year-old software company focused on personal and small business finance. A few months ago, Intuit hired its first Chief Data Officer to lead its 60+ data scientists. This speaks to how critical machine learning has become for products like QuickBooks, TurboTax, and Mint. Keeping track of your cashflow, taxes, and payroll can get complicated and time-consuming, especially if you are also trying to run a business. Fortunately, data can make this a whole lot easier.
Data science is becoming a key component for powering prosperity around the world. In this post, I’ll cover some of the areas we work on, as well as what it’s like to be a data scientist at Intuit.
Data science Project Areas
Data science at Intuit is focused on data products. We use machine learning to add new features and improve existing ones throughout our offerings. Here are some examples:
An extremely common task in any kind of accounting is keeping track of what each transaction is for: personal or business, entertainment or transport, assets or advertising. This categorization not only gives a high-level view of where all the money is going, but it also has important tax implications for both individuals and businesses.
Labeling each and every transaction by hand is, however, time-consuming. People would rather be running their business or spending their time doing almost anything else than manually marking gas, meals, rent. . . etc. Machine learning to the rescue! In products like Mint and QuickBooks, users have the option of connecting a bank account and having their transactions automatically categorized. Data scientists build the classification algorithms that make this possible. The result is a powerful system that is faster and cleaner to write than a comparable rules-based approach, and has the additional benefit of adapting over time based on how a user tags certain transactions.
For many people, their financial information is second only to their medical history when it comes to sensitive data. Intuit’s top value is “Integrity without Compromise” and we take our role as stewards of our customers’ data very seriously. To this end, there is a large team of data scientists, machine learning experts, and engineers dedicated to safeguarding the access and use of any data we come into contact with.
The team uses a suite of different tools to detect and prevent both fraud and other forms of abuse. These algorithms need to work in real-time and at scale. Because of this, data scientists in this group have particularly strong software engineering skills in addition to being well-versed in machine learning and statistics.
There are multiple teams at Intuit, including the one I work on, dedicated to customer success. Simply put, we make sure users achieve what they set out to do. We aim to make this as efficient as possible, while at the same time delivering a delightful experience. This is a broad area; in many ways the whole company works on customer success! In practice, my team ensures that customers get the help they need when they need it.
Personal and small business finances can be overwhelming and confusing at times. Fortunately there is a layered safety net to keep users on track. When a user clicks on the icon to search for help, the results are already pre-populated before they even type anything based on what the user has been doing in the product. After the user has typed a search string, there are algorithms to decide which results to show based on the knowledge base, community question, and answer forum. If they are still having trouble and call support, we help route the call to the right agent to assist them. Analysts then look at all these interactions, build dashboards, and report back to product teams on where we can improve. As data scientists, we build the machine learning algorithms that power these interactions. We are also looking for ways to make the experience even better: can we learn not just the part of the product a user was using, but what they were trying to do? Can we use natural language processing to understand a user’s question when they ask for help, and automatically recommend answers? Could we build an AI-powered assistant to deliver personalized guidance to each and every user? (See QuickBooks Assistant for an early beta of what such an assistant might look like!)
There is a lot of exciting work yet to be done.
Once a new or improved feature is ready to be rolled out, it needs to be evaluated for effectiveness. Every new data product is rigorously A/B tested to determine how much value it brings to our users. Often the data scientists who developed the particular feature are also the ones to run the tests. They are also heavily involved in bringing their models into production. This gives data scientists end-to-end responsibility over their models.
Fortunately there is already a strong experimentation infrastructure in place. The data team developed and open sourced an A/B testing platform called Wasabi. Wasabi makes it easy to assign users to different treatment buckets, adjust behavior via feature flags, and track outcomes. This allows data scientists to spend their time designing and analyzing experiments instead of figuring out how to run their tests in production.
This is just a partial list of examples of our data science project areas. There is active work happening on everything from risk models to recommender systems to image recognition. Stay tuned for future blog posts in which we will go into more detail on how we make finance an easy, frictionless experience.
Organization of the Data Team
Every data scientist at Intuit has a direct reporting line to the Chief Data Officer, but they can sit in different parts of the organization. Broadly, they either sit in a business unit such as the Consumer Group (TurboTax, Mint, etc.) or in the central Innovation and Advanced Technologies team. The business units are focused on delivering features for specific products, while work on the central team focuses on generally riskier innovations that can be used across products.
What it’s Like Doing Data Science at Intuit
Data science at Intuit is a community. Although data scientists work on their individual project teams, we are constantly working, learning, socializing, and growing data science at Intuit together. We have a company-wide data science meeting every other week to share what we’ve been working on. There are also study groups such as the working group on natural language processing and the data science journal club. Collaboration and personal connections is how work gets done.
The tools we use on a daily basis will be familiar to many data scientists. Most modeling work tends to be done in Python, some teams work in Scala and Spark, and there is a small contingent of R users. SQL is used to query databases. And now with Intuit’s recent partnership with AWS, every data scientist will have a common first-class cloud workflow for developing and deploying models.
Data scientists, and employees in general, are highly valued at Intuit and given a lot of autonomy. We are expected to drive projects, come up with solutions to customer problems, and also identify what those problems are. Every employee is expected to go on “follow-me-homes,” where we visit real customers to discover pain points firsthand. There is also company-wide 10% unstructured time for employees to work on anything they think might be valuable for Intuit, including learning new tools. Continuous learning and personal development is encouraged and horizontal movement between data science teams is common. This autonomy extends to hours: while most employees tend to stick to regular business hours (most meetings are between 9am – 5pm), some are in before 8am to use the on-site gym, and others get in past 10am. Quite a few people also work from home at least once a week. The culture at Intuit is remote-inclusive with nearly all meetings using (surprisingly good) videoconferencing software. All of this freedom allows employees to continually innovate and reinvent our products.
Personally, I enjoy having a real impact on our customers. Finances are a major part of life. For many, however, balancing their budget or filing their tax returns can be a source of unease, if not outright dread. By making these necessary tasks faster, smarter, easier, and more understandable, we are giving customers confidence and control so they can achieve their financial goals. Things like buying a home, putting their kids through college, saving for retirement, and building a business: that is what prosperity is all about.
Steve Brown is a Senior Data scientist at Intuit working on Customer Success as part of the Innovation and Advanced Technologies group. In addition to being interested in all things data, he is an enthusiastic and opinionated notetaker.