In their latest video and podcast, Randy Johnston and Brian Tankersley, CPA, review AI systems for accounting firms, starting with Acumatica ERP. Watch the video, or listen to the audio podcast below (transcript below):
Or use the below podcast player to listen:
Transcript (Note: There may be typos due to automated transcription errors.)
Brian F. Tankersley, CPA.CITP, CGMA 00:00
Welcome to the accounting Technology Lab sponsored by CPA practice advisor with your host Randy Johnston and Brian Tankersley,
Randy Johnston 00:10
good day. Welcome to this version of the accounting technology lab with my co host Brian Tankersley, Brian spent a good bit of travel here lately with the Acumatica Summit and the Zoho analyst day, and we thought it would be helpful for you to learn a little bit about the AI that Acumatica announced at Summit with Acumatica AI studio. Now the concepts here of acumaticas AI is the way we think this will be incorporated into a variety of products. Remember, we think that the general tools, the chat, gpts and the copilot, 365, and so forth, will be used for general productivity, but a lot of you will get the benefit of having AI directly in your systems. So Brian, what did you wind up learning in Las Vegas from the Acumatica team.
Brian F. Tankersley, CPA.CITP, CGMA 01:02
Well, I will double just first say that had a great event over there, you know? And the the thing that I’ve been struggling with, and I think a lot of us have been struggling with, is, what, what’s the vision of how AI is going to transform my work? Okay, we can see chat GPT, and it’s got a lot of G wiz things, and, you know, it’s nice, it’ll summarize emails and things like that for you. But what does it really look like when we when we have it actually solve real business problems, and it touches ERP data? Okay? And so Acumatica came out, and they said, basically they have, they have a three pronged approach, okay, they have an AI advisor that is that is kind of going to do, you know, it’s going to go through and identify exceptions and anomalies and do some basic things in the background. They have a tool that will help automate workflows and help accelerate those and then they even have an AI assistant plan a little further down the roadmap. Now, when we look at the architecture in here, the idea is that three of these will be apps, but a lot of these things will be just kind of running in the background. And again, they’ve layered on top of these, on top of these data sources that are already there in Acumatica. And remember, it’s a cloud first platform, so it’s got APIs and all the kinds of things that you need to connect into it. But they’ve layered in some intelligent assistance, including a help a help desk tool, a Help tool, a task assistant, automation agent. They’ve got an AI studio to help do automation and insights. They have an AI services gateway, and then you have the ability, or will have the ability, soon, to connect up to external, external large language models and machine learning models and other things. And then Acumatica is coming out with their own AI services models so that you your data isn’t used it, you know, isn’t used publicly in some of these public models. You know, we’ve had problems where data has leaked in some of the, some of the open AI based models between different different tenants and things like that. And so we really need to go, we really need to make sure that we’re protecting that data. So
Randy Johnston 03:22
Brian, to that point through the years, we have watched Acumatica have architectural ideas that I think are very innovative. Number one. Number two, this idea of keeping the AI private is going to become much more of a deal breaker or a big kingpin concept, as we see it, and a lot of the publishers are trying to figure out how to get that done right now and beyond that, I think they have a roadmap here that can be leveraged both with their own model and with external models, but they’re very cognizant about the privacy of the data, which again aligns with the way we ask people to think about things.
Brian F. Tankersley, CPA.CITP, CGMA 04:11
And I would, I would just remind you that when we look at Acumatica and Zoho, they’re very cognizant of the privacy, and they are very concerned about any kind of use of data for any other purpose. On the other hand, some of our other traditional vendors have said that we’re going to use your AI. We’re going to use your data to train our AI, and there’s no way to opt out of it, okay, yeah.
Randy Johnston 04:35
And you know, with the with the new models like deep seek that have been introduced, there’s much concerned about the privacy of those models. And just like the meta Lama model, now that it’s being used by the federal government, again, we’re kind of concerned. We not kind of, we are concerned about the leakage on that particular model. And. And so that’ll continue to be an issue. But you know, like many of the vendors in the last 12 months, the focus has been, how do we leverage AI on our platforms, and how do we do that effectively? Yeah,
Brian F. Tankersley, CPA.CITP, CGMA 05:15
and I don’t want to turn this podcast into Brian ranting about privacy again. We’re not gonna, we’re not going there. But I do want to say that I do want to go through and kind of break things down here a little bit. The AI Insights tool is the first phase of this. This is currently in beta, and it’s going to be in Acumatica. The the r1 2025 release that’s should be generally available in March. The idea is that you’re going to have real time dynamic data extraction and anomaly detection, also some automation for handling of exceptions and some predictive forecasting. And so the model that they’re going to use here is it’s they’re basically using this generic inquiry query language technology, and they’re using small language models here. Now remember the small language models are less prone to the as you recall from where we’ve talked about large language models, the small language models are less prone to the hallucination that is endemic in large language models. So they’re more likely to get things right the first time. Also, the anomaly detection in here, again, has some machine language powered services. If you’re using the Acumatica hosting, they’re going to be able to use that, you know, use all Acumatica hosted instances as potential sources without without leaking the data between tenants. So the idea is that it can look at different people in similar industries and identify what really is an exception, what’s not better that we also have some integration with automation tools and then some frontline components. They also have some forecasting services, and we actually saw some things at Zoho that we’re also looking at forecasting as another component of their of their AI, and they’ve got components to make them make this stuff more visualizable. So
Randy Johnston 07:06
just piling on to that thinking there, Brian, as I consider the various accounting software vendors, almost all of them are doing something in an FPA forecasting mode. So where we used to always recommend adding on third parties to get that done, I expect more and more of them to be included.
Brian F. Tankersley, CPA.CITP, CGMA 07:30
And I will say, I will tell you all that I culled through literally hundreds of slides for the episodes we’re doing this month because Mike the CTO at Acumatica, or Michael and and Raju, the chief evangelist at at Zoho. They, they are two of the best thinkers and explainers of these tech of this, these very technical concepts. And so what we’ve tried to do is boil it down so that we can talk about computer science without talking about computer science, if that makes any sense. Because you know this stuff isn’t brain surgery, okay, but there, there are some complex concepts. But you know the you’ve been doing the work yourself in many of these cases, looking for the anomalies and so forth. We just gotta, we just gotta get the language lined up. Anyway. So this is an example of this is a screen capture that they actually showed from the keynote stage. So the idea here is to look for anomalous transactions in payables. And you can see in here that it identified that there are 15 bills out there for which there is a late payment alarm. And then it identifies the 15 bills that are unpaid that should be paid at this point in time. And so So again, it’s it’s identified those, and then you can go through and review them and decide whether you’re going to act on them or not. They also mentioned some use of this in their operations areas, and particularly in field service, where they can use machine learning to track system performance. They can do some predictive analytics, they can do some root cause forensics, and then maintenance planning. The next phase was aI automation, and this was actually one of the more interesting parts. There’s actually a demo that’s about that I stripped out. That’s about seven minutes that we won’t have time to go through in this podcast. But what they actually use this to do is to automate complex workflows, do some customization, integrate AI services and then protect sensitive information. So the idea here is that you invoke this, it generates a prompt. It goes to the AI. It it reasons, it transmits the request. It invokes whatever AI action is needed on a limited basis. The response is received. It reasons, it updates the entity, and then does the action in here. So the the the example that we had in here again, would you. Would automate complex workflows. We do code free customization. It integrates external AI services and protects the sensitive information. Now, the example that Mike showed from his breakout, that also ended up in the day two keynote, was about about five or six minutes where he went through and showed how they were using the Acumatica service service tracking systems and and CRM data and the operations tracking information to take a case where they were a tech support case and to boil it down and to to summarize the tech support case into a resolution memo. This is a very time intensive thing, and you have to look at all the related data. And so they’re actually, they he actually showed a an AI, an AI bot, and some automation that would actually something that would actually write a draft of these case closure memos so that people could learn from them quickly, without having to go through and again, spend an hour or two creating them. Because, again, I’ve created those things, and it’s a beast of a problem to kind of document what the real issues are, but this thing will take the case notes and boil it together pretty amazingly. Now, again, for this, you’ve got an AI gateway, you’ve got that will manage the connections to the services. It will again track the usage statistics, and then it It limits the sensitive information it flows through. You then have a prompt editor that you can use. And in the prompt editor that they had, they actually referred to database fields that were coming out of JIRA and some of the other tools that they were using to manage the case management. And so they they retrieved case numbers and customer names and all kinds of other pieces like that directly by referring to referring to names in almost like mail merge fields. And so they actually prescribed the format in the prompt, and then it created this, this formatted thing. Then they have a processing engine again that goes through and does it. Okay? Now, these agents, so, so again, and I will tell you that that was the most impressive demo I’ve seen in a long time. Now, don’t get me wrong here. Okay, folks, I’ve been impressed with Mike the Acumatica CTO for a long time. And he always, but he always finds a way to blow me away with his demos that he does because he’s, he’s thinking, you know, he’s, he’s like a chess grand master. He’s five moves ahead of me and and he comes in answering the question that I that, I wanted to ask and then answering the two that come up once he tells me the answer to that. Okay, anyway. Now, looking at the AI agents, this is a way that you can basically just ask natural language questions. Of this, you’ve seen this in Power BI, you’ve seen this and a lot of other other things in here. Now, what’s different here is they have some automation to trigger AI driven workflows related to these, where it notifies you when you reach exception conditions. And again, you can dynamically present ERP data and and again, you get you get information to track how things are actually going in the field. In here, we also have a chat engine. Acumatica has their AI that that, again, will do chat and voice interface. So you can do it the way you want to. We can have the acumat is doing their own AI service. And again, the reason there is that we don’t want anybody’s AI trained off of off of your data. And so the idea here is that they would take anonymized data and and they would just just send the information out to it that was needed to answer the query. So you could, you could, for example, use this tool to analyze, let’s say, human resource data, without necessarily linking it to any PII. So so the engine could make good conclusions off of fields other than PII. And so it could, it could go forward. We also have some AI automation and insights in here. But again, this just kind of shows you how this, how this works. Again, you’re going to enter the prompt in there will then be some chat processing that’ll get translated into API calls by the AI studio. It will then go out to the ERP, access the API, pull the information that’s needed, go in and reason, generate the response, and then complete this. And I gotta tell you those, those closure memos, those case closure memos that I talked about earlier. I I wish that we had the time in the podcast to show you the nature of what they looked like, because it was everybody in the room. I mean, they the guy, Mike, had 300 people in the room, and they were just, you know, for the breakout session, you know, a 300 person breakout session, you know, just think. About that for a minute, and we were all just gobsmacked by it. So I got to tell you that they’re really doing it right now. Different pieces of this are going to show up in different times. Some of this stuff isn’t going to show up until late this year, early next year. But what I think they did a great job of, and what you need to know is that Acumatica did a great job of taking AI boiling it down in to solve real world problems, you know, sticking, you know, and and, and real work that human beings would actually do that would be on their job descriptions. So anyway, Randy, that’s, that’s what I saw.
Randy Johnston 15:37
Well, Brian, I had to let you run because I knew you were impressed with what you’d seen. I had reviewed the materials and had a fairly good understanding of where everything would go. But you know, in terms of innovation from a mid market vendor that understands what they’re trying to get done, Acumatica has done a very nice job advancing themselves again, and we expect and will cover what we see going on in other mid market competitors as they go through their various events over the next little while. So Brian, any parting thoughts here then on Acumatica summit, I know next year, it will no longer be in Las Vegas, but it’s been moved to Seattle.
Brian F. Tankersley, CPA.CITP, CGMA 16:25
Yeah, they’re going to go to the new, new convention center in Seattle. You know, the the thing I would just generally say to you about Acumatica and about mid market and the AI functions we’re talking about here, is that Acumatica is mid market. Okay, this is stuff that some of it’s already been done at a very technical level, at the enterprise level, it’s now making its way into mid market, and it will soon make its way into entry level accounting. And so I’m showing this to you because I want you to understand, we want you to understand, here at the accounting Technology Lab what’s possible, so that you start to create your expectations and your visions for what you’re going to do with AI to make your life better every day in your practice,
Randy Johnston 17:09
a delight, such a pleasure to have you listening in we’ll see you again in another accounting Technology Lab. Good day.
Brian F. Tankersley, CPA.CITP, CGMA 17:20 Thank you for sharing your time with us. We’ll be back next Saturday with a new episode of the technology lab from CPA practice advisor. Have a great week.
= END =
Thanks for reading CPA Practice Advisor!
Subscribe Already registered? Log In
Need more information? Read the FAQs