So we’re going to get started. Thanks for coming here today. This is AI in accounting, reality or marketing and just some housekeeping stuff. First, the sessions 30 minutes and the first 18 minutes of presentation, and then we’re open it up to questions. And that’s kind of the TED talk layout that we’re doing. Secondly, if you’ll check in on the Bizabo app and say you’re here, then they’ll send you all the material for today. And third is our presenter, Ron Baden. Ron is the vice president of CFO Tech at High Radius, which autonomous accounting is part of. And Ron started his career as an accountant, so he was one of us. And then he moved to consulting and then he moved to sales. And then he was CEO of a software as a service vendor in the IPM area in enterprise performance management. So statutory financial consolidation and planning. And I can’t think of a better person to do the presentation today on AI and accounting, marketing, our reality.
Ron Baden [01:12]
Thanks, Rick. So we’re going to talk about an accounting reality or marketing. It was a little bit weird listening to Sashi talk this morning because he was talking to all about AI, so I can’t say anything that’s counter to what he said. So kind of changed my presentation a little bit, but we’ll talk about is it reality marketing? Look at other departments, what it looks like in in accounting and then what we’re doing for and then we’ll open it up for questions. So before we start, I do have a poll question and we won’t do it on the app because the app sometimes works, sometimes doesn’t. Right, Rick? We’re not doing it on the app. We’re just going to do a show hand. So Rick said we should do a show of hands. And I said, I don’t know how many hands these people have because you can’t put up for all four. But let’s just do a quick show of hands for first. What do you think? A.I. in accounting is reality first. Raise your hand. Okay so have some kind of some beliefs. I’m not believe marketing so nobody believes it’s just marketing. How about more reality than marketing? Okay. And how about more marketing than reality? I’m with you. Okay. So kind of half half. Did you get that? And do you update bizabo or what? Okay, perfect. Okay. So as we were putting the presentation together, one thing that struck me as funny is that if you watch the movies and you believe movies and TV, this is the way they positioning air for a long time. Everybody know these three movies. Okay, everybody’s going to show me your age in this one. So this was iRobot, a robot that actually had a motion, right? This the Terminator? My favorite. He was a good guy, then a bad guy, then a good guy, then a dead bad guy, then another good guy and came back. And then who knows? The other movie classic, right? Decided it was playing checkers and actually launching a nuclear attack against someone else. Thankfully, TV and movies aren’t always right. There’s actually only really one TV show that ever predicted the future correctly. Anybody know what it is? If you said The Simpsons, you are 100% correct. The Simpsons has been almost spot on with all their predictions, including things like who is it? Richard Branson going into space. They got that one right. They got The Simpsons homer. And what’s your name? Marge actually wearing Oculus, which is pretty funny. And then the only other one, though, was Marty McFly, where they predicted the hoverboard way back. And then the hoverboard actually came out. No, his version was way cooler. But the reality is, if we watch what happens on TV and in movies and we believe that A.I. is a bad thing now, I’ll tell you about a little bit, just in my own life. So I’ve got two kids and we taught them, obviously, at a young age to learn. Right. So what do you do with kids? The first thing one of the first things you tell them is don’t touch the stove, the stove’s hot. And what do they do? Touch the stove, right? Because they’re not convinced it’s hot. Then they get burned and then they probably touch it a couple more times when it’s not on and they learn from it. The problem is computers don’t have that ability. They have to be programed like we have to tell them. And the computer will never make a decision of whether it’s hot or cold. It can’t where kids can, but we have to teach them. But what’s interesting is as as we took that poll at the beginning, what we found is that people don’t have a lot of belief yet. That’s a lot of mix of reality and marketing. But in other departments it’s really happening. Like, do you guys see this in your companies happening in other departments where sales, marketing, logistics are really taking off us too. So I work with a lot of software companies and if you look at marketing, is there anybody in here from marketing? Perfect. Good. I can make a joke about him. I always have to be careful because I make jokes about people. I’m like, Oh man, that was someone who actually is in that group. So sales would be the first person buying everything. You know, we as a recovering account, I always had to pay the sales invoices. It’s why I got in the sales because I’m like, they get to spend on everything. But sales would be the first folks buying every piece of AI, every piece of anything that they can. The next group is marketing. Marketing is equally as want to buy a sales are. And if you look at the different things that have happened in marketing, a lot of these have already been automated or have been had AI applied to it. So if you think of it used to be a stack of leads that would be backed up on an A desk. Now it’s in a CRM, Marketo, automated lead scoring, clarity, automated things like opportunity management. And so marketing’s done a really good job of embracing technology to use it and be better. Now, it’s never as Sashi said this earlier, it’s never going to be exacting and it’s never going to be where we don’t augment it. So even in marketing, we need to use it and apply human judgment to what’s being done. Now I want to show you. So video’s next, but we’re seeing it in finance already. So how many folks have automated their expense, expense claim process? Okay, perfect. So we’re starting to see it in a lot of places. And typically, this typically goes to sales like let’s be let’s be realistic. Like if sales didn’t put in bad expense report, we probably wouldn’t need to have an expense processing system like this, but we’re already starting to see it. It used to be lots of manual checking receipts to now I put mine in yesterday and I’m one of the worse there is. I put mine in yesterday and it actually read the receipt for me, put it in the system, told me the date didn’t match to what I put in, that I needed to split it out among the people that were there. The total didn’t add up because I had room service, but it at least it wasn’t a person that was doing that. So we’re starting to see some automation already, and that’s a good thing. Now I want to show you a video. Is this going to start recording? Okay.
Freeda Video [07:25]
Freeda. What do we have for today? I’ve updated your work list with four anomalous transactions totaling $792,457. Do you want to start with the first line item? Sure. The debit has posted incorrectly to software expenses and should be moved to the prepaid expense account instead. Should I create a task for this anomaly? Yes. Assign it to me. I have added the task to the October 2025 closed project. Do you want to open this task? Sure. Let me quickly review the suggested journal entry.
Ron Baden [08:11]
You stop. All right. What did you guys think about that? It’s pretty cool, right? How much do you think is real? Anybody want to venture a guess? I say all real. Yeah. Which parts do you think are real? Which parts you think are fake? All real. Okay, you get another Starbucks gift card. It’s not actually the case. I just love that you answered it that way. So the interface itself is real and Freeda is real. They’re both real things. They just haven’t worked together yet. Okay. And the reason I say that is to just bring up that. Okay. I would just want to bring up one point that it takes the learning, the teaching to make something like that actually work. So it takes a lot of programing in order to do that. Does anybody know if you ask and this is going to probably set off 55,000 alarms. If you ask Alexa what Ibadan means, is anybody ever tried it? Yeah, I have. If you ask Alexa what Ibadan means, it’s a country off the coast of South Africa. It has no idea what it is because we just haven’t trained it. And so the two things you saw are both real. So great answer, but it needs to be trained to learn for accounting, and that’s what we’re actually starting to get into. And if we look at so McKinsey did this study and said, we’re getting there, but we think that almost 80% of the tasks that we’re working on and that we have access to can be automated at some point. And you can see that tax and financial controls are a big portion of what we can actually automate. So those are areas to go look at from an automation perspective. And so we’re making progress, but there’s still a long way to go. And they kind of split into two areas. And I apologize for the chart. I’m like four feet away and I can’t read it, so I’ll move closer to it and just kind of give you an idea of it. But on the top part, this says these are automated for processes that enhance user experience. So things like general accounting operations, you wouldn’t think that necessarily was for people’s experience. But if you think of somebody who has to put an invoice into the system and wait to see if it was actually paid and did it come out right or did it come out wrong? Their experience is actually poor, waiting for that data to come back. And so that’s an area where people have automated, very transactional, and at the bottom you get to tax. Now we get into some high value, really high value things like Treasury, risk management, biz dev, external relations, where these are way more strategic things and can be automated more and they’ve got a lot of value add in those activities. So those are areas to really look at from an automation perspective. And just to give you an idea, because I thought this was really interesting when I first heard about it, I was like, I have no idea what it is. Right. And when I heard of a bot, someone said that a bot was going to be running. I thought it was like the iRobot, you know, like what you see is because it bot is short for iRobot. A bot is short for robot, just like fan is short for fanatic, right? We kind of shorten these things, but a bot is just a piece of code that’s running that’s been taught to do something. It’s not like a robot, not like you’d see in a manufacturing plant or with, you know, automation from that perspective. But there’s a lot of examples of AI and ML that are happening that folks should be aware of. So the first one is just process management apps. So an example would be like SAP Concur, which we talked about, which is automating these expense reports. So it comes in somebody can put their expense report in via their credit card and the thing that it really impacts is a better user experience. So lots of applications out there that offer that kind of impact. And then we start getting into RPA where it’s process automation, process automation, and it’s really about the bot, a piece of code that can automate something. So imagine a recurring journal entry, the correction of an invoice that comes in every month incorrectly, an omission on someone’s expense report where they didn’t put it in. Those are all examples of where you can very easily automate a process that’s just going to be with a small piece of code and it’s high volume and transactional. And then we get into the actual machine learning and artificial intelligence side of it. What’s interesting is I didn’t realize this. Do you guys know that when I’m probably going to break my own, I don’t say this word SIRI because if I do it, everyone’s phones are going to go off in the audience. We’re going to get, what, 25 answers to this question. Do you guys know that at the Super Bowl, did anybody see Google’s ad for Alexa? Everybody say it. It triggered like 26 million Alexa devices. And they all knew that they went off at the same time. And they knew the exact reach of the Super Bowl ad by saying that name on TV. They all went off at the same time. So if I. Did that and I said, siri are I in here? Same reaction would happen. I would know exactly. But it took Google two years of programing. Alexa with a thousand people listening to the questions that people are asking it. They were actually searching the web, typing the responses, and then it was learning from that. So that’s actually how machine learning and AI actually works and the way we do it. So we have an app that’ll sit on the general ledger and what it does is the same thing. We provide it like templates for all the automation things and then it spends time learning what happened. What did the accountants do over the last six months to 24 months? And once it learns it, then it can run it. Does anybody know how many times you have to say, okay, before you can actually train a machine? Rory does. Because I’ve told him this. Anybody in a 47, it’s actually four times. Yeah. After the fourth time of clicking learn it will then know and it knows to then process that. And they learn this from credit cards from from when you go travel. This used to happen to me, you know. Ron, are you in Buffalo? Yes. Okay. We will allow your transaction of process. Ron, are you invited to travel a lot? Ron, are you in Boston? Yes. And at the after the fourth time, it actually new. So an example of what we would do for machine learning and AI. And then you get your packaged applications, things like the closed management dashboard or things like the AI engine. That’s very, very specific. If you look at Zendesk is a really good app where they’ve taught their application and been able to package it because they know it’s specific to what they’re doing. So think about Zendesk, a ticketing system. If you put a ticket in on Zendesk on one of your apps and said, I forgot my password, you’d get a reasonable response back, right? That said, okay, click here, forgot password and change it. Now imagine you did a general. I forgot my password to the internet. You’d get some Nigerian friends telling you you won $10 million, your warranty is expired and whatever else the other one would be. But there are companies that are learning how to package it up so that when we ask these things, what is Ebola, it knows exactly how to respond to it because it’s got the criteria. And just some examples of folks that have seen these are on IR side, the receivables, but just very good success and taking processes and automating them down into less people time, better results. So they’re both here. I know the known is here. I don’t know if Staples is, but if they are, you’d certainly ask them about this. But I know there are two very good examples of automating the process in terms of doing better with AI, ML or just even automation. And so just to give you a sense, and I think we touched on this earlier, but what HighRadius has been traditionally known for obviously is this top layer, the order to cash section with these five apps. Year, year and a half ago we introduced the Treasury concept and then over the last few months we’ve released the bottom two. So going from the office of the CFO just in order to cash in and Treasury now into accounting with the record, the report will eventually take that to consolidation into more of what we’re Kivus doing with the with the workbench and then for FPNA delivering an app that somewhere between a better version of Excel and an easier version than the EPM tools and each of them does the same thing. So there’s a solid section, which is what others have done already. So think about Record Report. Lots of companies have done a closed dashboard, lots of them have done a reconciliation app. But what we did is then say, okay, what’s different about it? And applying technology like AI is we built an app that sits on the general ledger and corrects entries as they’re happening. I think Sashi talked about this earlier. And so as much as I’m interested to talk to people about, Hey, you can solve your record to report problem, everyone wants to solve that. Can I make my general ledger perfect problem? Right, because like me, I was an accountant and I was a super bad accountant, so don’t ask me about my kind of experience. I understood the concepts. I just couldn’t sit still severe. A.D.D. That’s why I’m up here and I’m not an accountant anymore. But I understood the concepts and like, the one thing that drove me crazy is they said, you have to be exact to six decimal places, exact six decimal places. And then we’d go into clothes and they’d say, Oh yeah, forget about these. These are non-material items and you don’t have to worry about them. I’m like, Wait, where all I have to be exacting, but now you’re telling me to ignore these hundreds of transactions. And so what we’re starting to see is the starting point for a lot of folks is applying a I to something like that bucket. We can’t pop. Simply have humans do it because it’s just too big. But imagine if we told it to look for a set of transactions that are repeatable and to report back whether they’re high, medium or low, and what the impact to the financials are going to be, to great, easy use to adopting and records report and RFP. And we get this constant argument and prick, you’re supposed to give me a warning. Am I getting. I think I’m right at. Probably 2 minutes. Of course not. Okay. Imperfect an RFP and we see this constant argument. And it’s funny because when we started building the product, we’re thinking about what would be cool features to deliver for FPNA. And we had this whole list like Excel Hell and, you know, calculations and too much data. And where we kept coming back to was this story about the CRO presenting in a quarterly business review or QPR. Right? And the sales manager puts up and says, Here’s my best rep in the enterprise and then says, Here’s my best rep. And mid-market and finance says, Well, wait, which is your best rap? You know, who’s your best rap? It goes, Well, it depends. You know, if you depend what metrics or if it’s R by cost, you know, that’s a good way we can define it. So they go off, they start doing that, they come back to the QPR and what do you think happens? They put it back up there and finance goes, wait, where did you get those costs for h.r. Says, Where do you where did you get those costs? Like you don’t have access to cost data? And they said, oh, we went to everybody’s hire, you know, are their offers and we took the data and we wrote it in there. And then they said, But that’s not the true cost because you didn’t include recruiting and you didn’t include all these things. So now the CEO goes, okay, now finance or h.r. You do it or finance. So they do it in celsius. Wait, that’s not the right sales data. Now you have an argument of whose data is right. And so what we built was a collaborative tool that leverages the ability to use EML, the machine, learning to learn about the intersections of the data. So how does this employee number respond to this, this number and sales, things like that? And so I radius is really committed to I think Sashi said we’re going to be out of business in ten years. Lisette is coming in seven years. Okay, if we don’t grasp this, we’re going to be out of business in seven years. So so I’m going to wrap up, but this guy’s going to take a picture and I want a really good one. So I’m going to stop talking for 1/2 and pose while he takes a picture those for you think one my better okay are you video. Oh dang it I knew that 100% knew that that was going to be the case. Can I give you a Starbucks gift card to stop? And I’m like, I’m only kidding. You can you can keep going. It’s okay. You can keep going. Yeah. I totally agreed with Sashi that in seven years if we don’t get keep video. All right, so that’s actually my last slide. So we’ll take questions. If anybody has questions about anything related to AI, the future, how we’re using in accounting.
Yeah you know, you should that subject to future collections. Yeah. You have the exceptions highlighted. So what’s the challenge that is in the collections? I’m just curious as to why the technology is not what it is.
Ron Baden [21:34]
Yeah, it’s it’s very close. We just haven’t programed it for things like, like for example, the anomalies engine does things like somebody booked an entry and it should be like prepaid expense and they booked it to software expense. We just haven’t told it yet that that’s an anomaly like that’s wrong booking. So it is they, that’s why I said they both actually do exist but it’s the coaching for it. So they didn’t understand exactly what to do on the general ledger that allows it to go fix those entries.
And then and then blockchain. Is that anywhere you guys in the back of your mind?
Ron Baden [22:11]
It definitely is. Yeah. I’m glad the guy stopped recording for this part of the of the discussion. No, no, no. So blockchain is right. I mean, it’s the technology that’s going to allow these transactions to happen digitally and at speed. We found a couple of use cases like shockingly, the use case of converting Excel formulas for huge data sets is a great example of where to apply blockchain. We just haven’t found the technology we want to work with yet. So but it is I mean, I’ve heard a number of discussions. We just haven’t really picked the right spot yet for us to apply it.
Are you learning? Teaching like that example.
Is that just looking back through history to see. All these different types of entry point for months, etc.. You actually got to tell it, you know, as we go through this, as you’re going through all this, right?
Ron Baden [23:08]
Yep. So it happens in three ways. The first way is we tell it, we have a bunch of templates that we’ve built that we tell it what it should be, right? So basic accounting rules and it’s about prepaid expenses and accruals and things like that. Then we actually take the rules from our customers. So they give us a list and say, if this happens, do it this way. And we take that and we apply that as a template. And then we take those and we go learn. So we like to get 24 months of data is the right number. Anything after six months is too little. So 6 to 24 is the right number. But we can learn all of the data and all of the things that are happening in that 24 month period. So that’s actually how we apply it and get it to know your specific goal, your structure. The thing that we’re actually working on that’s going to be a little bit more interesting is the ability to train the humans who’ve made the mistakes once. So think of someone who’s sitting. I know it’s a but, but that’s where we’re actually headed, is to say, this person keeps doing this thing, call it an AP invoice. We keep looking at it and correcting it. How do we tell them and get that behavior to change so that they actually stop doing it? And that’s the part that we’re working on. We already built out the can we reported. Can we auto correct it? Can we do the digital binder? Can we make the entry in your general ledger? Can we do both sides of it so that it’s trackable? Give it to the auditors. That part’s all done. The part we’re trying to figure out is how do we enhance the experience for the user who doesn’t want to code it wrong, right. Doesn’t want to spend 30 days making mistakes in the general ledger. So that’s the part we’re extending it to. Yeah.
Love the way that you break that down to the three different stages of. What percentage of your automation are you actually? As you go from the stage one to stage two? Yep. Yep.
Ron Baden [25:11]
So we ran it for one of our enterprise customers. And I’m not going to say the name because the date is rather embarrassing, but we did it for 24 months for companies, 500,000 transactions a month. At post closed data, we found 5000 anomalies still after they were closing. So we’re getting about 80 to 85% once we do the learning in automation. So of the transactions that they never needed to look at again, we were able to correct 80% of them. Those 5000 after now that we learned them and we templates system we’re getting into the low 90% range. So their close went from 15 days to seven and all seven days are high value tasks.
So where’s the theory would be that it would be a further.
Ron Baden [26:00]
Upward. Well, from your side upward band? Yes.
It sounds like you’re hitting a point, but then it’s more of cracked. Goes off first as well as quickly.
Ron Baden [26:14]
That’s right. Yep. Yep. Yep. And it actually we’re it’s yeah. I mean, we know the girls like I mean, that’s one of the interesting things that I really liked about what we did is we had the integrations from the other product line, so we were able to leverage them, right? So we had all the girls, lots of the banks, and so we were able to leverage those integrations to make it a much faster implementation and and deployment. So we have lots of customers who are doing nothing but buying the anomalies engine to just make their GL perfect. All right. So just think about, you know, being able at the end of the month to say, I’m certifying and I know these transactions are done and we’ll just work on the high value tasks from the from the GL. So it’s a really interesting way to add automation and add some air, but I like the automation aspect first because you can templates, then you know that you’re looking for the right transactions and then it finds the other ones. So. Any other questions? I can bribe you with more gift cards.