In many ways, AI is the ideal financial partner: unemotional and impartial, able to make a host of calculations and find patterns in seconds. In practice, however, just under 10% of Americans actually rely on a robot to help them with their finances.
The missing piece? User confidence. In the abstract, trusting an AI robot with your finances might seem obvious, but the user experience with a bot might erode that trust.
Intuit, the global technology platform and industry leader in AI-powered financial tools, is tackling this disconnect across its TurboTax and QuickBooks products by paying close attention to situations where its customers need more than a cold computation to build trust. For example, even when a TurboTax user could read in the product what their refund was, most users still opted to call Intuit's customer support to hear the same information from a live tax expert—indicating to the Intuit team that users didn’t just trust the raw output of a computer. While millions of users get started using TurboTax and QuickBooks, the perceived complexity of managing their own finances often proved daunting. This overall lack of confidence became a significant barrier to adoption of the platform.
So, in 2017 Intuit decided it needed to take a different approach: build AI assistants that are better at guiding users, personalizing their experiences and, when necessary, connecting them with human experts. By using AI to make its tools more user-friendly and accessible, Intuit hopes to offer financial expertise at scale, empowering consumers and small businesses to overcome pervasive financial challenges from budgeting, to credit card debt, to saving for retirement and more.
Customer Confidence Starts with Better Customer Service
Tax filing can be a stressful prospect; some 73% of people expecting a refund say their annual return is important to their overall financial well-being. TurboTax uses machine learning to help its customers get the maximum return—and for many, that means opting for the standard deduction. But many users wanted more than TurboTax’s assessment. To start to address this confidence challenge, Intuit’s team realized they needed to build an AI-powered assistant that could not only do the work behind the scenes, but show its work.
When the assistant launched in 2018, the Intuit team noticed that users were calling customer support less, thanks to the ability to access more in-depth explanations about the filing process—a unique feature literally called ‘ExplainWhy.’ ExplainWhy is modeled as an incremental, interactive discourse between the user and the system—making it easier to address new questions as they arise and help the user “drill down” to a clear answer.
This capability in TurboTax is part of a larger paradigm shift in the way Intuit is approaching customer support to create confidence in its AI tools.
“Instead of seeing customer support as something that you do when everything else has failed,” Greg Coulombe, director of Intuit’s Conversational Experiences team, said, “now it's about turning around and using the ability of AI and conversational technology to actually give people the ability to not only self-serve, but to also get expertise.”
Another key element to building customer confidence comes down to personalization. Take, for example, a small business owner or self-employed individual using QuickBooks to manage their finances. As soon as that person logs into the QuickBooks platform, potential next steps are pre-populated based on their unique characteristics (such as their business type or where they left off at their last login) so that it’s easier to get started. When they open the customer support chat window or call the helpline, the AI assistant can use historical data points combined with those customer characteristics to guess why that customer is looking for help. From there, the customer can choose the question that fits their needs—or ask a different one.
“The tool already knows what problem I have,” said Loren Lacy, a group product manager who leads the conversational AI and digital platforms teams at Intuit. “So I'm going to have a lot more confidence in the answers or the guidance that it gives me than if it says, ‘What are you calling about?’ or ‘What do you want to know?’"
Meet the AI That Can Do Your Taxes
Building this conversational interface that offers each user a personalized journey while answering a variety of questions was an iterative process. Coulombe said the first AI assistant (launched in 2018) could answer only a small set of the most common customer questions. Over time, the Intuit team carefully tracked metrics to adjust, improve and expand its abilities.
Today, through natural language understanding and machine learning models, the AI assistant can comprehend a question or comment typed in a chat window or spoken on the phone—even if there are typos or homonyms in that person’s speech. And automated question answering allows the AI to respond to the customer, pulling from static, human-curated responses, or even dynamically, from a rich corpus of tax and accounting articles.
“We were fortunate that we could leverage some of the recent revolutions in deep natural language processing,” said Andrew Mattarella-Micke, a staff data scientist who specializes in machine learning for natural language at Intuit. “So for example, a number of transformer models have helped us understand, not just from a keyword-based perspective, but based on the whole context of [the customer’s] language, what they're saying and how they're communicating with us. And that really helps us be more flexible, more sensitive and more personalized in many different ways.”
Intuit’s AI leverages a variety of algorithms to cross-reference the user’s question(s) against Intuit’s body of knowledge and come up with the best answer. And in delivering an answer (known as fulfillment), the AI algorithm incorporates user data—down to the way the user phrased their question—to add a personalized touch.
The key to Intuit's AI is its flexibility: Rather than hard-coding tax logic into the assistant’s programming, the TurboTax Knowledge Engine makes the assistant more dynamic, using knowledge graphs to connect associated user data together. It’s what allows the AI assistant to understand and answer uncommon or abstract questions.
“One very common question that customers would ask last year is, ‘Why do I have to pay?’” said Coulombe. “Now, that's a really interesting question because it means one of two different things: ‘Why do I have to pay Federal Income Tax?’ or, ‘Why do I have to pay State Income Tax?’ And from the question itself, there's not enough information there to know which one the person meant. One of the things that a digital assistant is very powerful at, as compared to, say, a search engine, is the bot will ask, ‘Oh, did you mean this or did you mean that?’ It's a process called disambiguation.”
These abilities have allowed the AI assistant to assuage some of the initial confidence challenges that caused consumer churn from both TurboTax and QuickBooks. In 2020, for example, the number of TurboTax Live customers in season grew nearly 70% while average handle times declined 15%. The success rate of small businesses on QuickBooks is 10 points above industry average.
In the future, Lacy, Coulombe and Mattarella-Micke hope the AI assistant will instill even more confidence by offering more proactive guidance while striking the right tone. It’s a capability that’s still being explored internally:
“People love to experiment with emotion matching or emotion tagging,” Coulombe said. “If you sounded upset when you asked a question, or if we're about to give you bad news, our AI should act accordingly. For instance, when you say, ‘hey, what's my refund?’ the bot should adopt the appropriate tone and phrasing for your particular situation: whether you were to owe $10,000 in back taxes, or you were to get a $10,000 refund.”
So, Can It Tell Me How I Should Manage My Finances?
Despite the robust data and algorithms that drive the TurboTax and QuickBooks AI assistants, there are still scenarios that the tech is not equipped to handle. Sometimes it’s because the AI can’t answer a question, but more often the issue is whether or not it should answer a question.
“‘Is it better to take this deduction or that deduction?’ is really an advice question, for example, and you need to talk to a trained expert,” Lacy said.
This is Intuit’s ‘Live’ feature: available today in TurboTax and QuickBooks, and Mint in the future. Intuit’s AI assistant still plays a role; in addition to answering a user’s questions, it works to connect them to the best tax or bookkeeping expert to answer more complex questions. If a user is filing as a head of household, owns a home, and lives in New Jersey, they’ll get connected to a very different expert than if they rent an apartment in California, their filing status is single, and they have a robust investment portfolio. Intuit’s AI then delivers the expert a summary of the user’s journey in the platform so far. All sensitive information is masked in accordance with Intuit's data privacy and stewardship principles. The AI also takes notes throughout the Live call so that the expert can stay engaged with the user.
In the future, the AI assistant can be an even more robust player in this interaction: For example, with a customer’s permission, picking up cues during the conversation to serve up documents in real time, when mentioned. The goal, the Intuit team said, is to augment finance experts with AI to take care of tedious or repetitive tasks, so they can focus on what they do best: giving strategic advice and counsel.
“A bot or a product can tell you whether you can afford something, right?” Coulombe said, “But questions like, ‘What stock should I invest in?’ or ‘Should I hire an employee for my small business?’ or ‘Should I open a 529 plan now or wait five years?’ Those are all the kinds of things that a financial advisor might be way better qualified to answer.”
With the possibility of Mint Live on the horizon, there is a major opportunity to further democratize financial access across the population—an urgent need for many. Survey data suggests 75% of Americans manage their own finances, but only 33% of adults can answer four out of five financial literacy questions correctly. And when money questions arise, 60% turn to friends or family or don’t have someone they trust with questions about finance. Many of these consumers are in significant debt in the United States. They are overpaying for things like credit cards or auto loans.
“We're really trying to put the financial expert in the back pocket of every person,” Lacy said.
“Our goal,” added Coulombe, “is to use human expertise and AI to communicate to individuals and small businesses in the US and worldwide in a natural, smart and simple way to help them achieve financial prosperity.”
Intuit is the global technology platform behind products like TurboTax, QuickBooks, Credit Karma and Mint. Their mission is to power prosperity around the world by helping customers and communities overcome their most important financial challenges. Visit Intuit.com to learn more about how Intuit is using their AI-driven expert platform to solve challenging financial problems for consumers, small businesses, and the self-employed.
This story was produced by WIRED Brand Lab for Intuit.
