Can AI Be Your Bookkeeper?

Jordan Buckner: [00:00:00] There's a lot of excitement right now around AI and it being able to take over every single industry and automate everything that you do with your business. But I'm always curious to find out where the limits are and what's possible right now. So for today's episode, I want to talk about AI, and if it can be the bookkeeper for your business, a lot of founders have a lot of issues with accounting and bookkeeping early on, especially with inventory based businesses.

And there's lots of different ways of solving it from a full, you know, outsourcing your accounting to a bookkeeper or accounting firm to using a couple more automated online platforms now, and AI is starting to find its way in. So want to figure out where. The technology is these days for this conversation, I've invited on Alice, who's the CEO of pocket, which is a online accounting software, and they are starting to integrate AI into that platform as well.

Alice, welcome. 

Alice Zhang: Thank you, Jordan. Glad to be here. 

Jordan Buckner: So traditionally founders or any [00:01:00] business owner would work with like an accountant and they would kind of go in and manually enter in transactions , into their bookkeeping software a lot of times, maybe a QuickBooks and over the last couple of years, there's been more automation.

I know you've been able to use automation AI to associate kind of transactions to a chart of account, reducing that workload of founders. How has that technology really been developing and what's kind of next with AI? 

Alice Zhang: Well, that's a loaded question. I'd be happy to share our experience so far, right?

Obviously that just one angle with one lens and happy to learn more from your perspective. I would say right now we're leveraging AI technology in two aspects. One is like you just mentioned I think categorization itself for e commerce or CPG businesses has always been a pain and a chore that most people don't like to do.

Like even accountant don't like to do. So what we have done is we [00:02:00] developed a I call AI bookkeeper. And our AI bookkeeper basically takes a company specific chart of accounts. And trying to, you know, figure out for all the transactions we have ingested. What's pick, like basically pick the best fitting category that we should put this transaction on there.

Obviously, I mean, the accounting category, accounting account, right? So we have been doing that. I would say I think from the feedback so far people love to see the instant effect because right now we do real time data ingestion of all cells. All expense and all you know, invoice and bills data.

Sometimes I would say obviously we target smaller startups, CPG companies. I would say under 10 mil, but even that range, sometimes daily transaction could be hundreds or thousands of transactions, so it's getting. To a point where it's just not that viable. You hire a I would say a human [00:03:00] accountant who can do that type of categorization manually for you.

Jordan Buckner: Yeah, I think that's a really good point because a lot of times if, even if you have like a QuickBooks online or so, if you work with a bookkeeper, maybe that information is updated every. Month, maybe if you're lucky every two weeks or every week, but rarely is it daily? And so it creates a much different understanding and information for your business.

If you can go in at any time and know that your data, you know, might not be a hundred percent accurate yet. If you need to make any adjustments, but it might be, you know, 80, 90 percent correct, or you can identify any potential issues or trends much faster than having to wait an entire month. 

Alice Zhang: Yes, exactly.

 I would say if it's a manual process which we used to do that is we probably wouldn't do daily categorization number 1. So you wouldn't get daily let's say consolidated sales visibility. You probably get it monthly or quarterly, right? And also we wouldn't be able to categorize on the older level or, you know, [00:04:00] even more granular because older could consist of multiple skills, like items could consist of discounts and platform fees and shipping fees, service fee or taxes.

So , it could be a rabbit hole. You know, if you want to go down, I think with AI, it is possible. I can categorize daily transactions and each order level and their order and each line item level, and then the, I can basically. Separate out different accounting category again, if it's shipping fits, shipping revenue versus.

Sales or product income. Right. Which has implication on your cost of good sold calculation, for example. So I do think ai I think improved efficiency so much, and I would say latency in terms of understanding, making sense of your number that perspective. 

Jordan Buckner: One thing that I'm really excited about with AI is the ability to surface.

Information and insights and even like flag concerns with a business, right? Like maybe one month your. Cost of goods [00:05:00] sales are much higher than expected, or your expense in one category dramatically changes to be able to flag, to say like, Oh, that's something to investigate further. And as a founder, you can have a better sense of what's changing in your business with the transactions.

Because one thing I realized, even as a founders, every time I'm buying ingredients or supplies or a service, I see is constantly changing. And over the course of a month or two months or a year. Sometimes that pricing can have dramatic shifts. And so I'm excited about, you know, something like AI to be able to flag like, Hey, this was a big change , in your business.

Do you think, you know, that's going to be possible just with AI soon? 

Alice Zhang: that's a great question. And we're experimenting something in that area as well. So obviously AI can be utilized in. Many functions of a finance operation. What we have to just talk about was, what was probably more of the fundamental function, which is literally just do the bookkeeping match the categories to transactions.

I would say another [00:06:00] function, which is probably a step up is I call analysis function. So we do have a beta. Product where I call it , your financial interpreter. So you can think of, you know, AI financial interpreter as your junior analyst, right? We are experimenting with. You know, AI financial interpreter, meaning if we train the AI to give some prompt, can the AI to properly flag certain, yeah your revenue has grown more than 20 percent over last month.

Congratulations. Here's your biggest contributor , to that growth, for example. So , we could do some of the initial analysis on that. I think to it, it relies on, I would love to work with my customer to figure out some commonality in terms of what type of, you know, reading, the bond or outside of the bond that should trigger AI to give you the flag, you know, we used to do it in a more coded way, right?

we see. We give you a bunch of maybe lower bound, higher bound. [00:07:00] And if it's above or below, , we send you hard coded notification. But with AI, I think it can be much more like a human can be much more, I would say intelligent, quote, unquote. I think that's something where we're experimenting is.

The LLM model by itself. Obviously we have the open AI kind of generic model, but we're also seeing some vertical models. If we can extract some of the industry, maybe like. Industry analysis for CPG, for example, for food and bev certain vertical we can fade it to the AI and AI probably hopefully would be able to, you know, quote, unquote, intelligent and spit out certain, I would say flag either say, Hey, you're doing great.

This is why you're doing so great. Maybe you want to double down. Next month, right? Or you know, there's sharp increase in expenses or , your cost of goods. So, as percentage somehow looks very. Unnormal and maybe you want to do something about that. So that's the intelligence part.

I think we're testing out at [00:08:00] that maybe. I think the further development may need a combination of the generic LLM model with some industry, I would say industry benchmark number in my view, and with the user specific data. 

Jordan Buckner: Yeah, I like that. What else are you seeing that you're excited about , with AI being able to do to a business in maybe like two years, five years, because I think the speed of this development is really fast.

What do you imagine being possible in the, you know, not too distant future? 

Alice Zhang: I think , where AI could probably make the most impact is AI could be more human like. Could, you know, understand, interpret the human communication part. I think AI come a long way, you know, we kind of started with AI.

Natural language processing and then we have the LL a model, which is like. A hundred a thousand times you know, knowledgeable and intelligent. So that's where I think, I think a lot of the accounting. Services or frictions coming from coming from either drop of communication, you know, between the [00:09:00] business people and the accountant that the financial services personnel or from misunderstanding.

I think that's where. AI could help the most. That's why I personally, I'm very excited about this AI analyst or AI interpreter, or in the future could be AI recommender, you know, really doing more like CFO recommendation, decision support. That obviously that would have to be. To the, I needs to have a judgment.

That's a whole nother speaking. We're talking, but I think probably it's, it's going to kind of like autonomous driving, you know, it's probably going to gradually chip away and first just replace the chore work. In second half, some analytical, let's say, called AI D and analytical analysis, but still largely driven by data, by benchmarking, by a little bit of judgment.

But that judgment by and large is , more just rational. You know, have some kind of basically reasoning, right, logic, capability. But when it [00:10:00] comes to judgment. Again, that's a whole nother area. I think it's probably gonna emerge pretty soon. You know, AI, CFO obviously I think there's probably some way to go.

I don't know if it's three to five years or 10 years but we can chip away some components of that. 

Jordan Buckner: Yeah. You know, it's exciting because you mentioning that reminds me the past two weeks I've been testing out the chatGPT is new. Advanced voice feature where it's more conversation like in, in the AI that you're talking to.

And it's funny because like, there's, it's still a little rough around the edges and the responses, but it does feel like you're having, you know, a conversation with an AI. But I could imagine in a couple of years that this is tied into a system and it's trained on. These are accounting practices for CPT businesses and things to do and not to do of almost like having a conversation either via text input or voice with an AI where you're reviewing your business and having reports and being able to ask [00:11:00] questions and it being able to instantly service the data , and then ask you questions as a, you know, as a business owner to understand your goals.

And then it almost becomes that. You know, assistant and , the insight into the data, because I think one of the most, you know, the valuable skills of having an accounting team is someone who both understands your business, but then also has access and expertise over the numbers, but you know, if you're an accountant, you're not going to know every single number and every single kind of chart of account.

But with assisted with AI or something that might be possible at some point. 

Alice Zhang: Yeah, totally. And you know, , that's why. You know, I named the company in my PocketCFO. I think that the vision is if everyone has a CFO in their pocket could be the AI CFO. I think obviously whether it can completely replace CFO or not, It's a question mark or I'm live question, but I would say it definitely can chip away some of the low Henry foods, such as for example, if you're not accounting trained you don't know what's the [00:12:00] difference between a poor, like simple question, like what's a pro based accounting versus cash based.

And tell me what's the pros and cons of each method. I think those kinds of questions even today, it can be simply answered by a AI bot, right? And also I think that there's this generic education part, like you don't. Again, we're talking about you don't even need to go to college.

You don't even need to take those specialized classes. You could literally get trained on the fly on your work. If especially if you're a CPG founder, you could ask AI all those questions. Tell me what's the difference between balance sheet and PNNL? I would say one of the biggest one is balance sheet.

The number is ending balance. It's a point number. But the number on your P and L is a sum number. It's a sum of all the transactions, but the balance sheet number , is NDS. Whatever your asset or liability and, and this point of time, right? So balance sheet, you don't have some. Column, but P and L, we have some column.

That's precisely. So those kind of like, very [00:13:00] simple question, I think, can be easily educated just by. Asking a, I co pilot in my view, right? Then you can get educated. And then obviously, then you, once you, you have those general. Building block education, then you can go into. More of, you know, yeah, then I can make a sense of my number on my financials, then I can like a read more about what's the implication of my balance sheet of inventories negative, for example, right?

And then you can carry a conversation with. , the AI, the AI could say , it could be negative because of the number of races could be, you don't have an opening balance, you know, could be your cost of goods sold somehow is joined too much out of your inventory. So yeah, it's very exciting.

Like literally you could have a virtual kind of CFO maybe should begin with as a virtual analyst, but in the future, a virtual in your pocket. So whether you talk to investors, you talk to bankers whether you're looking at your financials, it's [00:14:00] literally gonna lift you up so much, and you make decisions 

Jordan Buckner: I don't know.

I think that's super exciting. Actually, yeah, , I'm always like optimistically terrified of it because at the same time, right. Hey, in dealing with lots of numbers, I think a lot of founders don't know the intricacies of their financials and their numbers just from like an accounting standpoint, and there could be the sense of AI models as they're developing, but you know, hallucinations basically where they're Making up numbers, but presenting it as accurate still can happen.

And I think that's where a lot of founders are also concerned of, you know, if, as this is developing, how well do you, you know, what's the level of accuracy and our understanding of the accuracy. 

Alice Zhang: That's another great question. I fully agree. I think because of , the black box nature \ , of the transformer model of the LLM model every time you click a button, it gives you a different answer.

So. I think the solution as far as I [00:15:00] I see is I, I do think, especially when it comes to specialty in vertical AI, I think human or expert in the loop it is necessary. I think in the midterm, in the short term midterm, it's, especially when it comes to finance, when the stakes are high accuracy is required.

And there's liability involved in, you know, every financial statement you share out to, you know, to lender and investor. I think that's why , like, even us, , we have human in the loop. I do think the model would be, especially when it comes to compliance. Related or high stakes documentation there still needs to be a human, like a real CFO, a real controller a real CPA who can review and sign off on the data.

So. I think there could even be legal issues, obviously, around to completely give the control, the power to AI. But in the short to midterm, I [00:16:00] think the human, the expert in the loop part would still be applicable. 

Jordan Buckner: Yeah, I think that's super exciting. and being able to better understand like where things are and ask, I think the biggest thing is with AI can be on demand and so, right, like, you know, Which I love, I actually, I like talking to a person because there's a lot more nuance that can be understood, but you might have to schedule a call like a week out into the future or a certain time which is completely understandable.

Versus some of the cool things I'm excited about with AI is that like, Oh, I just have like a question right now. I can immediately ask and get an answer. And so I'm excited to see how those things develop over time. . Yeah, it sounds like, you know, AI. Can't quite completely replace , your bookkeeper right now nor do you necessarily need it to be, but it's going to be more and more powerful into the future and just become an everyday part of your, your business.

Alice Zhang: Yeah, I think my personal view is we should embrace it. Like for the exact reason you mentioned it saves time, saves money it can chip away much of the chore work or some answer simple questions, [00:17:00] as long as we're aware where the limit is and where we have the boundary, meaning if it's crossed some kind of boundary, if it's liability compliance.

or high stake. We should get a human expert involved. I think as long as we're aware of that, we should fully embrace it to make our work, I would say, more efficient and more effective. 

Jordan Buckner: Very well said. Alice, thanks so much for being on today and talking about this. 

Alice Zhang: You're welcome. Thank you for having me.