Jason Staats at Xerocon: Using AI starts with curiosity

Staats: Using AI starts with curiosity

August 16, 2024 4:30 PM

Artificial intelligence has been everywhere in the accounting profession lately, to much joy or consternation, depending on who you ask, but regardless of how one feels about it, there are always a lot of questions.

During small business platform Xero's annual Xerocon event in Nashville this week, Jason Staats—a CPA, YouTuber, podcaster and advisor who has spoken extensively about the impact of AI on the accounting profession—answered some of the questions, both from moderator Akankshu Dhawan, senior vice president of product management and EGM with Xero, and from the audience.

Staats discussed how people can get started with AI, as well as the pitfalls one might encounter on the way, along with some of his overall impressions on the current state of play in this growing field. While he said it's difficult to point out any one area where AI is being especially impactful, it's still hard to imagine any part of the profession that is not being impacted today by this new technology.

Q: How can accountants get started with AI?

"I think it starts with a degree of curiosity and, increasingly from really smart people like you in our industry, we have more folks going out and making helpful content on how to do AI stuff and launch cool automation communities for folks in tech and AI. I think lean into that curiosity. … You don't need to take two weeks off work to do a deep dive, you just need to be plugged into it and start experimenting in very basic ways, because this is sort of a personal assistant rather than the top down rollout of a practice management system. AI, especially large language models and chat assistants, it's almost bottom up. It starts with your personal workflows and getting comfortable with it. … It's this mental reworking on how do you work this into my day to day to be helpful to me."

Q: What are some things to be careful about with AI?

"For sure data security, privacy, making sure you understand all that stuff. The big difference with AI is understanding how model training, how the conversations you have with it could be trained into the new model. We are lightyears ahead of where we were 24 months or 18 months ago in that we now have completely secure ways of doing this stuff. The fact there are bad ways to use AI does not mean AI is bad, just don't use it in the bad ways. You need to define the boundaries where your team can innovate responsibly within this stuff.

Another thing to watch out for is thinking, 'What does it mean to roll this out in an organization?' As we roll this out, our big blocker here is incentives. Is my team incentivized to be higher output, or am I paying them to sit in a chair for X hours a week? … We can see people who talk out of one side of their mouth about AI, but at the same time their team is incentivized on the billable hours model, and there is literally no incentive for them to stop and think 'is there any better way for this process to be done?' If the system we build isn't going to reward them for that, we can hardly expect them to make space for that."

Q: Everyone has heard AI is expensive, so if you want to implement it successfully, what kind of budget do you need?

"I'd say the opposite—the best thing about generative AI is how cheap it is. You used to have to train your own models, so you'd drop X billion to do it. Now we have unbelievably good models driving most of the stuff we use that we can now access for free or $30 bucks a month. It doesn't take more than $30 a month per user to start off with ChatGPT or a secure way to start testing different workflows and that sort of thing. I actually think carving out a budget, saying we'll spend X on this thing, kind of defeats the purpose. What is going to drive curiosity and help the members of your team grow and develop this new skill set that is so important for the future of how we work? It does not need to be expensive; start simple, and don't make it the job of a subset of your firm. Because as soon as you do that, everyone else is like, 'Oh good, they're the AI people, I don't need to learn that.' That is a big risk. So get it in people's hands; some people will take to it, others won't, but it does not need to be a big, spendy thing."

Q: How soon do you think AI can start producing accurate books?

"What the people here understand is [there are clients who are] a hot mess on all sorts of issues from the Lulumon purchases on his credit card to writing off his haircuts to the random information he has to provide every month. AI is going to get us way closer [to accurate books] but if you do tax you know if someone buys something from Starbucks it can [still] be classified in seven different ways. AI will get us way closer, but people who understand this stuff understand we are a long way from someone snapping their fingers, connecting to a bank feed, and 'boop' it's done."

Q: A lot of firms want to adopt AI but they want to be sure the work done is accurate. How do you assure them?

"I mean, the same way you ensure it's correct now. We absolutely cannot trust the output of AI, but I've never known an accountant to trust anything or anyone—even colleagues. Someone else can do something, but the minute it crosses your desk, you rip up the carpet and make sure it is correct. I think this is what makes accounting firms very well suited for AI. You've already built the machine to ensure something is correct before it goes out the door. We're never going to plug something into AI and ship it sight unseen. But because we already do this—you remember what your first few years were like—we already manage this from junior to senior, to people making mistakes, and we build systems to manage that. AI can be a part of that, and I think we are well suited for that because we work that way."

How do you utilize AI today in making YouTube videos?

"There's some weird stuff in those videos that is hopefully very obviously AI. AI is impacting creative stuff in really amazing ways, from creating an avatar of, 'we've pumped all 300 episodes of the podcast into a chat assistant, and you can chat with it, and it can fetch information from the transcripts instantly across 4,000 hours of audio,' to increasingly making really compelling visuals with AI to really stick the joke. We're also, interestingly, moving away from stock footage libraries, and instead, there are more and more platforms for creating AI influencers, which is terrifying in so many ways. But we try to use those tools for good—like make some hypothetical person, and put him in different settings—and we do that a lot more now than with stock video libraries. So it totally impacts a ton of that stuff. It's not replacing anyone but increasing the capabilities we have."

Q: Will AI make it easier to be a sole proprietor? And if so will the population dramatically increase?

"I don't know if it will or not. It will certainly make solo people higher output. Did the computer lead to more solo outfits? I dunno. There are a lot of other aspects of running a firm with other humans besides needing the humans to get the work done. It will for sure help with some menial stuff you don't like, but every time we do that, we transition to, 'these are now the menial things I don't want to do.' You can for sure run a higher leverage, higher output firm than before. But if it were up to me, what would lead the decision of whether or not to build a team would be more about: Do I want to create cool employment opportunities for other people? Do I want to get to know these folks and build a relationship with this team, pour myself into them, and help them grow? So, yes and no."