At Intuit, machine learning enhances products, CTO says

Artificial intelligence might be ultra-trendy at the moment, but at business and personal finance software vendor Intuit, machine learning and AI have been part of the company's technology arsenal for years.

More than 30 years old, Intuit is a multibillion-dollar company that has long employed AI tools to simplify finances -- from taxes to accounting -- for consumers and businesses alike. Yet, while Intuit has been in the business for a while, it is striving to evolve its tools with the times and has filed about 170 AI-related patents.

While Intuit machine learning and AI technology are increasingly being integrated into the Mountain View, Calilf., company's customer service and security efforts, AI's biggest role is with its smart products, says Tayloe Stansbury, executive vice president and CTO.

In this Q&A, Stansbury discusses the past, current and future AI and machine learning capabilities in Intuit's finance software products.

Editor's note: This interview has been edited for brevity and clarity.

Can you speak to how AI is used in Intuit'ssmart products, particularly in the tax and finance software?

Tayloe Stansbury: Starting in the tax space -- it turns out the tax law, even though it's really complicated, is just a bunch of rules.

Tayloe Stansbury

So, and this is going back to the AI of the '80s, why wouldn't you think about the tax law as a bunch of rules as first-class objects and represent it that way? That way users could do things like inference and infer the answers to questions we might otherwise explicitly ask users, thereby simplifying the experience users might have to go through to get their taxes done.

Similarly, if users aren't happy with their refund, you'll be able to do back-chaining to bring them to the parts of the questions that they filled in as part of their TurboTax experience that drove the most impact on their refund so they can be sure they got their numbers right.

That's just an example of old-school AI being applied to the problem of complex government regulation to try to make that simple for users.

What about more modern uses of AI, including Intuit machine learning tools?

Stansbury: In completing those tax returns, for example, more modern artificial intelligence, like machine learning and big data, is used to look at all the millions of tax returns that we see to try to identify deductions a user maybe ought to have taken or errors that otherwise might have been difficult to find.

To the extent that we can use data, the enrichment of the data and AI running across that data, we try to figure out what would be better choices for this user to make.

For instance, users can spend a lot of time filling out their deductions, like medical or charitable deductions, in the hopes they might qualify for a bigger deduction than the standard.

We can tell with pretty high accuracy using all the data they've entered up to that point whether they're likely to get a deduction bigger than the standard deduction or whether it's a waste of time for them to start digging up all those receipts. That can save about 40% of the time a customer has to spend filling out a tax return.

Then, in our accounting and management software, including Quickbooks and Mint, there are a couple of cases of modern AI there; for example, problems with cash forecasting. A lot of people make a lot of mistakes with cash forecasting. Our Mint customers, for example, spend about a quarter of a billion dollars on insufficient fund fees.

One of the things we're introducing is a forward projection of what your cash is likely to be based on things we already know about you and things we know about people like you or businesses like you -- so we can help you forecast when you might be bottoming out and let you focus on what-ifs.

As Intuit's machine learning tools continue to advance, how do you think this technology will grow, and where will it take the company?

Stansbury: Prosperity is hard. Running a small business is hard. For many people, running their own finances can be hard.

To the extent that we can use data, the enrichment of the data and AI running across that data, we try to figure out what would be better choices for this user to make.

Sometimes, the hardest part is getting them to make that choice. In order to get customers to respond well, you really have to build a profile for each class of customer, whether they be a consumer or a small business. And then, based on the data and the AI, identify what financial tradeoffs a customer might be willing to make.