Hello and welcome to the NVIDIA AI Podcast.
I'm your host, Noah Kravitz. it's been said that there are only two certainties in our world death and taxes i'd like to propose an amendment to that to bring a little positivity into at least half of the equation I'm going to put forth today that there are two certainties in our world, AI and taxes.
Our guest today is here to speak about both of these subjects. how they interrelate, and also what's going on in the world of small businesses, which...
If you've been following the news over the past couple of months, you know is a very important topic to the small business community going through a lot of changes. and difficulties in the wake of the COVID-19 outbreak and everything that's happening to the economy since then.
So we will get into all of that with our returning guest.
Ashok Srivatsava is Senior Vice President and Chief Data Officer at Intuit.
He joined the podcast roughly a year ago to talk about Intuit's work using AI to help small businesses and other folks do their taxes, manage their finances, run their businesses.
And he's back and we couldn't be happier.
Ashok, thank you so much for returning to the NVIDIA AI podcast.
No, it's such a pleasure to be here. I'm looking forward to this discussion.
So why don't we start with a little bit of an open-ended question.
Obviously, since we last spoke about a year ago, A lot has gone on in the world, to say the least.
Why don't you tell us a little bit about What Intuit and your team in particular has been up to using AI, using deep learning, not only to help people prepare for the tax deadline, which is pushed back this year in the United States to July 15th.
So as we record, it's about a month out. but also kind of more generally what's going on in the small business community.
So the COVID-19 has had an incredible impact on our businesses, small businesses worldwide, but particularly in the United States.
And when you see what's happening to businesses just in your neighborhood being shuttered for weeks and then months.
It's really, really alarming. What I'm excited about is that in our way, we're thinking about how artificial intelligence, machine learning, and knowledge engineering can be used to help small businesses.
Probably one of the most exciting areas for me is the Intuit Aid Assist program, which really helps people who are small business owners figure out if they are eligible for loans from the government, how much that loan might be what relief they're eligible for, and what the terms on that loan would be.
It sounds like that should be simple, but it turns out that because of the number of rules and the number of questions, it's actually a daunting task for many small businesses.
So we've used artificial intelligence, in particular knowledge engineering, which allows us to isolate and focus on only the most important and most relevant questions for a small business owner so that that person can determine whether or not they're eligible and how much of a loan they can get.
And this is an exciting program because It helps us help our customers get the money that they need to make it through these really, really difficult and unprecedented times.
Yeah, as you spoke to, and you know better than I, but I know from friends and colleagues I work with that access to capital worries about just keeping the business afloat, being able to make payroll, keep people on insurance and such is top of mind for small businesses.
When you mention knowledge engineering to help people sort through all of the different programs and regulations, Can you unpack that a little bit and describe what knowledge engineering is?
We've defined artificial intelligence to have three major components.
One is machine learning. One is natural language processing and one is knowledge engineering.
So knowledge engineering is that part of artificial intelligence which takes rules that people have written.
Let's say there are regulations, rules from the government, but it converts them into computer code automatically.
And this then becomes the brains behind a lot of the questions that are asked in TurboTax, a lot of the questions that are asked in the Intuit Aid Assist program. and many other places within into its product lines.
What's really remarkable about it is that it codifies essentially what people have written down.
So it's a marriage of natural language processing, which understands language, but then converts it into computer code so that the computers can actually understand what's written and take action on it.
So we have made enormous investments in building this knowledge engineering capability because of the space that we work in.
We do a lot of work in compliance, a lot of work in taxes, a lot of work in accounting.
And so this is a great use of artificial intelligence.
Now, as sort of a layperson who has filed taxes for many years and also – read through and sometimes struggled to read through tax guidelines and other, let's just say financial documents, that sort of thing.
Do the particulars of this kind of language and regulatory language and compliance language and that kind of thing, Do they present any unique challenges to applying natural language processing and knowledge engineering as you just described?
You're not alone, I can tell you. I was trying to be diplomatic.
The U.S. tax code is 80,000 pages long. And I am amazed at its complexity.
And I'm also amazed by the team that we have that has taken that and essentially created the ability to read it by machine and to convert it into computer code.
So yes, absolutely, it is a daunting task.
And it's one that's really perfect for artificial intelligence because no matter how smart a person is, no matter how much they read, it's very difficult to say, oh yes, I've read the US tax code and I understand it.
But we actually have to create systems that do that.
That's our job. And so The use of artificial intelligence and the use of knowledge engineering and machine learning natural language processing in this context is really, really the key enabler in my opinion to help us to get to that level of understanding of the tax code.
Before we go any further, for folks who are listening, if they're interested and could use the help of the Aid Assist program, where can they go online to find out more about that?
So you can go to adassist.intuit.com. kind of widen our scope for a minute here.
And obviously, as we've alluded to in so many ways, 2020 is a year unlike any most of us have ever seen.
And the tax season is unlike others. The April 15th filing deadline has been pushed out to July 15th already.
And I don't know if you want to split this into sort of pre-COVID and COVID era or how you want to approach it.
What sorts of trends have you been seeing leading up to the tax season?
Or maybe if you just want to focus on the past couple of months –
Are there things that have emerged from user behaviors and the data you're seeing that are really just kind of notable?
So what we're seeing in the tax traffic and the use of TurboTax is that people are taking advantage of the additional time But I would say that because of COVID, the impact is more on the small business side.
We all know we have to pay our taxes, but how small businesses are dealing with COVID and how they're dealing with the need for capital immediately is really where we're seeing it.
Small businesses are having a hard time paying their bills and having a hard time supporting the employees that they have.
And I have to say, Noah, that as I... look at the data as a person who used to own a small business in the past, I can tell you that the most important thing is meeting payroll.
And making sure that you have enough money to pay your employees because they work so hard for the small businesses that they're working at.
And so that's where we're really seeing the strain.
That's where we're really seeing a lot of traffic and a lot of importance from the use of AI.
So let's talk a little bit about how that data comes in and what into its offerings are, because beyond the tax filing. software and products that you offer.
There's a whole suite of bookkeeping products.
And so is that data kind of coming in in real time and kind of giving Intuit a look at how the economy is sort of moving data? day, week to week, month to month.
The small business data does come in on a very regular basis.
So daily in real time, we start to get information about what's happening with small businesses from the bank transactions that they're uploading into QuickBooks or from other information like payroll and so forth.
And so in the data, we can see the trends and the impact of COVID-19, like I mentioned.
Probably the most important thing for us to note is that this effect of COVID-19 is going to last for a long time.
Even as the country opens up, we know that there are going to be more and more businesses that are impacted by it.
And so we're spending our time not only analyzing the data, but then thinking about new ways that we can use artificial intelligence in order to help small businesses make better financial decisions.
That's our overall mission. And we're really trying to think about it in this particular context.
And so if we can get into that a little bit, how are you using AI and deep learning to help the small and medium-sized businesses?
So I'll give you a couple of examples. One of them, which I'm really excited about, that's very relevant to COVID-19 is what we call cash flow forecasting for small businesses.
So small businesses need to support their customers.
They need to support their their employees.
But fundamentally, they need to make sure they have enough money in the bank to meet all of their obligations.
And this is where cash flow forecasting comes in.
So what we do, we built deep learning algorithms and other kinds of algorithms that help us forecast, let's say two weeks from now, how much money is the small business going to have in the bank?
What are the outlays are coming up? What are the major invoices they need to pay?
What are the major invoices that need to be paid by their customers so that they have money coming in?
This is one of the most important applications that we're focusing on for small businesses, and we're using deep learning in very, very novel ways.
So it's important for a small business to know how much money they're going to have in the future.
But if you think about it, it's very hard to say you're going to have $30,000 in the bank next Tuesday.
That's very hard because it's such a precise number.
You'd like to know the machine would ideally say, we think you're going to have $30,000 in the bank next Tuesday, plus or minus 10%.
Hmm. Figuring out that confidence interval turns out to be a hard problem.
Statisticians, people who work in time series forecasting, have worked on this for a long time.
We're building deep learning neural networks that can allow us to identify and predict not only the $30,000, but the confidence intervals as well.
And that really helps a small business. Because if you think about it, if I told you, Next Tuesday, you're going to have $30,000 in the bank and therefore you're going to be able to make payroll.
That's good. If I told you, Next Tuesday, you're going to have $30,000 plus or minus 10% confidence.
That's much better. And we want to give small business owners that ability to know not only what we think the number is, but how confident we are.
Because 10% might be okay, but if I said next Tuesday $30,000 plus or minus 100%, That's not very good.
You can't take action on it. And we want to be able to give people that level of information.
And so when you're talking about the margin of error on the confidence metric, Is that a function of – supply chain is maybe not the right word, but just all of the –
The people who your business does business with being able to meet their obligations and that sort of thing?
Yeah, it depends on a number of things. How often does the small business have to pay their suppliers?
How often do the customers pay back? Do they pay on time for invoice-based businesses?
When are credit card bills coming due? When do we have to pay the rent?
When is payroll due? All of these types of things get factored into this deep learning model.
I'd like to say for those people who are interested in this area, it's not just deep learning, but we're bringing together other areas. or quantile regression, new algorithms for bringing together nonlinear predictions.
It's a very complicated, very exciting area of machine learning and deep learning that we're exploring.
Our guest today is Ashok Srivastava. He is Senior Vice President and Chief Data Officer at Intuit, And we're talking about Intuit's work, particularly with small and medium-sized businesses, using AI and deep learning techniques to help them navigate the financial aspects of running a business.
But right now in particular, at the tough times, that small businesses are really up against in the wake of COVID-19.
I'm just going to open it up. What else are you working on now?
I know we talked in the past about chat bots and voice recognition.
I don't know if that's where you want to go.
But what other things are you and your team working on at Intuit applying AI to small business finance?
I'd love to tell you about the work in chatbots and natural language processing that we're doing.
In that area, one of the most important things that we need to do is to build systems that can understand what people are saying. common conversation, because when we have those kinds of capabilities, we can serve our customers better.
Let me give you an example. For instance, When a person is, let's say, using TurboTax or QuickBooks and they have a question, often they don't know what to search for.
Often they don't know exactly where to look for that information.
And so if they go into application and they start looking around, the artificial intelligence systems in the background are taking that information and making predictions about what their real question is, what their underlying issue might be.
And it's forming that opinion in a sense.
And then let's say a person actually asks a question in our help box.
What happens is that we look at the question, plus all of the feedback they've been providing us implicitly through the use of our product.
And then we do a match in the background and we say, we think these are the most important articles for you to look at, or we think this is where you need to get help.
That's the type of language processing that we're looking at.
Not only looking at what the person is typing in, but also bringing together their behavior, the way they're using their product in order to come together with the best recommendation of articles.
Another example, as we're looking at people's chat requests as they type into the chatbot.
We're trying to parse that information in order to understand what their underlying intent is, what their specific question might be, and provide an answer back.
And this uses very heavy deep learning technology because the understanding of the language is not being done through rules.
It's not being done through knowledge engineering systems like I described earlier.
It's actually being done through neural networks that go through the language and attempt to understand it, and then attempt to make predictions based on it.
How often do people search for the right or wrong things?
People are very smart and they tend to search for the things that are relevant.
But if you think about it, sometimes people have very specific situations.
So for instance, They might be working in a particular industry, like the farming industry, and because of the type of farm that they're working on or because of their geographical location, certain tax credits might be available to them that they may not know about.
So even when they're searching for That information, it's our job to give them the most relevant data possible so they can make the best decisions for themselves.
With an 80,000 page tax code to go through.
On behalf of taxpayers everywhere, we thank you for that service.
These are the questions that we ask on the podcast, and yet they feel so different just given the... the sort of abrupt halt and then what feels like a rapidly accelerating change in the economy and the way business is being done and sort of the what people are referring to as the new normal that's emerging right now.
But How do you see the work that you're doing kind of progressing over the next five years?
And frankly, is that answer different today than it would have been six months ago? before COVID-19?
Definitely. Let me give you some ideas that we're working on and that we're building.
So we've spoken a lot about language, we've spoken about rules and knowledge engineering, But there's another area that we're focusing on that I find incredibly exciting.
Again, think about small businesses, think about consumers that are trying to file their taxes and think about the sheer number of documents that they have to deal with, their receipts, There are W-2 forms, 1099s and so forth.
I don't have to imagine I can see them in my home office right now.
So I'm with you, Ashok. Yeah. Now, take a look at that stack and imagine the ability to just snap a picture or upload it in some other way and that the machine parses all of that information, extracts all the necessary information puts it into a database, and then your taxes are close to being done.
Just imagine that world. That's the world that we're trying to drive to with speed, with deep learning, and with a lot of other technologies.
And I'm really excited about it. Because again, as somebody who has done a lot of taxes, as somebody who has dealt with unfortunately a lot of paper, There's nothing that I would like more than to see all of that happen through the click of a button.
And it's easier said than done. Because if you think about it, all kinds of W-2s are out there.
Some of them are even handwritten still.
Some of them are in different forms, they look different.
And making a general purpose vision system, computer vision system that can look at all of that and get the right information with extremely high reliability. and high accuracy is a great challenge.
So I can't stress enough, I can't put enough importance on this idea that we have to be very, very accurate, that even a small error can have a costly impact.
That's an enormous area of investment for us.
I'd like to go into a bit more detail about the ability that we have and that we're building to use deep learning to extract information because The documents that we have and the other kinds of information that people are dealing with on a daily basis, the receipts, these are things that really... exemplify the problem.
Each receipt is different, each form is different, and building a system that can look through and pull out information requires a couple of things.
It requires deep learning. It also requires knowledge engineering in the background.
Because if you think about it, when a person has a W-2 form, all of the data on their have relationships to each other, and our knowledge engineering system knows that information and can make determinations about whether or not We're reading the form correctly based on the data that it has in the background.
So it's an ambitious project. These are ambitious areas to bring together knowledge engineering, language, and then also the deep learning and machine learning capabilities.
And what's the current status of, I'm gonna oversimplify it, but the ability for a user of an Intuit product to take a picture of a receipt or a W-2 form and upload it and then have your system work its magic.
Yeah, it can be done today. And we're rolling it out in stages.
Okay, but we're really excited to see that real people are using it and they're getting benefits and we're measuring those benefits as best we can.
People can use it today and it's being used on a daily basis.
For some reason, that reminds me to ask you to go back You mentioned that you were a small business owner previously.
What was your business? So we, my wife and I, after we finished graduate school, We decided to move to California and we got real jobs, but then we also decided that we would start an art and yoga center in Mountain View, California.
I'd say within a month or two of coming to the Bay Area, we found a retail space in downtown Mountain View and we started small business and we ran it for several years and It was such an incredible education for me to know what a small business has to do on a daily basis. to know what it means to make the rent.
We didn't have payroll, but we had to pay rent.
Then also to know what it means to deal with an inventory, to work with suppliers, to work with the students, the yoga students that came to the center.
It was really an incredible experience. And I've also, after that, Many years after that, I ran a different kind of small business.
My mom, unfortunately, suffered from dementia.
Almost eight years or so, I took care of her and I had a set of employees that used to come to help take care of her.
And that was the time that I really became very aware of the importance of meeting payroll.
Those people who took care of my mom were really really dedicated to it and I'm so thankful to them and I also wanted to make sure that I could meet payroll and so I was very I was viscerally aware of the importance of meeting payroll so that they could pay their bills and that my mom could be taken care of.
I've run different kinds of businesses over the years, and I'm very, very thankful for all the people that worked with me in those times.
So to kind of take that to a bit of a wrapping up point here, you mentioned previously the cash forecasting tools.
I may have gotten the name wrong, but that Intuit offers to small businesses.
Are there other roles that deep learning can play, whether it's now or in the future? particularly during tough times when small businesses are struggling to make rent to meet payroll and otherwise manage cash flow.
I think that artificial intelligence and deep learning in particular can really play a number of important areas.
If you think about that small business owner or the consumer, every day they're making decisions that affect their livelihood.
Those decisions need to be informed by data and by recommendations that are based on the judicious use of that data.
And so I think that for the foreseeable future, these two things are going to come into play again and again data and machine learning data artificial intelligence are going to come into play again and again to help people get the best information at the right time so that they can make the best decisions about their own financial future.
Well, Ashok, thank you for taking the time to come back onto the podcast.
We certainly look forward to having you on again, maybe next year at tax time.
And hopefully by then things will be. turned around, looking up better for small businesses and folks around the world and in the U.S., But in the meantime, we mentioned the aid assist program, but for folks who want to find out more either about the specific work that you and your teams are doing with AI. or more broadly about ways that Intuit can help them with their own small businesses, where would you recommend that they go online as a starting point?
The best place is intuit.com to see what we're doing and how we can serve our customer base.
So with everything we've talked about now, we see these are very serious and uncertain times we're going through.
But the work that you and your team are doing Ashok and just the advances that already have happened in AI and deep learning and machine learning more broadly, do give a lot of reason for folks to be optimistic about the long haul, if you will.
Yeah, no, I'm optimistic about the future because I think we've seen, not only over the past few years and the past few decades, We're seeing that people are more and more willing to use real data, objective understanding of facts, in order to make good decisions.
And artificial intelligence can play a critical role in that.
It can play a role to provide people with the right information at the right time. to provide insights that they may not have known so that they can make the best judgments for their own financial lives and for the financial lives that they impact.
And that's what I'm excited about. I really feel that as we move forward as a society, the more objective we are, the more data we use, the more clear headed we are in our thinking, the better off we're all going to be.
Literally could not have said it better myself.
Ashok, thank you again for your time and all the best to you and yours.
It was a real pleasure, Noah. Thank you so much.
Thank you.
