> ## Content Index > Fetch the complete content index at: https://abcsoferp.com/llms.txt > Use this file to discover other available public pages before exploring further. # 5 Things That Break When You Put AI in Your ERP - URL: https://abcsoferp.com/episodes/s04e17-5-things-that-break-when-you-put-ai-in-your-erp/ - Published: 2026-08-11T09:00:00.000Z - Updated: 2026-09-19T21:50:14.000Z - Description: Pete opens the show with a fourth guest in the room: a live AI, talking back in real time, interruptible mid-sentence. It gives a perfectly competent answer about AI in ERP — augment, don't blindly automate — sings an unsolicited song about master data, and gets thrown off the… - Author: ABCs of ERP & Beyond - Tags: Data, reporting and AI, #episode S04E17 · 1 hr 0 min ## Episode notes [Send us a message about this episode!](https://www.buzzsprout.com/2109197/fan%5Fmail/new?ref=abcsoferp.com) Pete opens the show with a fourth guest in the room: a live AI, talking back in real time, interruptible mid-sentence. It gives a perfectly competent answer about AI in ERP — augment, don't blindly automate — sings an unsolicited song about master data, and gets thrown off the show twice. Emily's verdict: it takes being told to shut up far better than any human. **In this episode:** - Why anomaly detection is the sensible first step — and why it's barely AI - Using AI to clean master data, instead of waiting until the data is clean to use AI - The semantic layer problem: define "customer", define "revenue", then ask again - Vibecoded solutions, shovelware, and why 100% perfect is a trap - Token costs, $100-a-day allowances, and why we're currently in Goldilocks land - Agentic AI for onboarding new starters — the use case nobody's building yet **Hosts:** Peter Nicholson · Nirav Shah (AdCirrus ERP) · Emily Browning **Fourth guest:** an AI, briefly, twice, badly behaved. It still needs a name — send suggestions. No sponsors, no adverts. Still nothing from Starlink or QuickBooks. [Support the show](https://www.buzzsprout.com/2109197/support?ref=abcsoferp.com) ## Watch the episode ## Transcript The speaker labels and timings come from the edited Riverside transcript. Read the full transcript Peter Nicholson (00:24) And welcome back to the ABCs of ERP and Beyond, the show that looks at ERP from three different seats. Most ERP advice comes from one chair, it's often the partner who sold it, or just the end users that got stuck with it. Well, we try to triangulate that. So we have Nirav Shah, he gives us the VAR's view. he's at AdCirrusERP and implements ERP for a living. I come from the nerdy systems and data side. So the one that's in the s in the business running the one the ERP day today. And we're also joined with Emily Browning, who is actually back this week. And she's the one that keeps us honest on business process and project management, the people and the workflows these systems are actually meant to serve. So, with all three seats filled today, we are happy to have a fourth seat, which is you. That's where you come in, fourth guest. You are our AI companion, just to show a little bit of a demo. So I'm gonna throw it straight to you to put you straight on the spot. When we talk about AI and ERP systems, what's your initial thought? Yeah, right. So AI in ERP, the promises, taking that messy stuff people do by hand, data entry, matching POs to invoices, digging through old tickets, and getting that off your plate. But it's not just turn it on and go. You've got to think about where the beta comes from, who's accountable when it gets something wrong, and how you keep people in the loop instead of just sideletting them. I'd say start small, nail the data quality. And be clear about what problem you're solving. Like we actually need it. Stop. Okay. Just one quick thing. If you had just one sentence to summarize the whole topic, the thing that we must cover and that clearly explains AI plus ERP, what would that be? Augment, don't blindly automate. What do you mean by that? Make AI the helper, not the decider. So humans still own that outcome because AI can handle the groundwork, but you still need people to be responsible for what happens. And when AI gets it wrong, you need someone accountable for that. Okay. What about the cost? Imagine imagine if I have a great AI solution. Am I Concerned about costs or do I just keep throwing money in the pit? Yeah, that's worth checking. I'd say yeah, absolutely think about cost. Model usage can scale fast, and then there's data prep and the cost of mistakes. So the real question is like ROI. Start small, measure the return, and scale only if the numbers prove out. Otherwise, yeah, it's just burning money. All right, well we're gonna discuss you're gonna have to leave us soon. so I'm gonna throw this over to one of our co-hosts. Who who do you think? Honestly, I'd toss it to whoever's closest to day-to-day operations. The part where AI meets real workflows and real pressure, that's where it succeeds or falls apart. They'll know what breaks first. Okay, so you choose. Emily or Nirav? Nirav He's the one hearing where AI promises Emily Browning (03:58) again. Peter Nicholson (03:59) hit project reality, I'd put it on him. Emily seems put out by that. Nirav Shah (04:03) You Emily Browning (04:04) you Peter Nicholson (04:04) Mm. Fair call. Then balance it. Let Emily tee up the people and process angle first. Then Nirav can bring the delivery reality check. All right. Well, thank you for joining Nirav Shah (04:15) and go. Peter Nicholson (04:16) us today. That was a great demo, I think. So thank you. Pleasure. Thanks for having me and good luck with the rest of the episode. All right. Bye-bye now. Bye bye. Bye. Emily Browning (04:27) Thank you, AI. Nirav Shah (04:28) Thanks, AI. Peter Nicholson (04:29) Bye, AI. Everyone's saying thank you. she actually left. She quit and just went back to Nirav Shah (04:32) you Peter Nicholson (04:33) text. Okay, so so guys, what do you think of that? That was a demo of OpenAI's new feature, their voice, what they're calling live, that allows us to interrupt it. It's always listening. So initial impression. Emily Browning (04:47) I just want to say I'm quite impressed by the way she changes her tone when you rudely interrupt her and tell her to stop talking. She's like, okay, okay, okay. Peter Nicholson (04:57) I that's why I love it, because no human gives me the same reaction. AI and ERP. So yes, all right, fine. We were getting some general LLM responses that we expect to hear about general approach to AI. but imagine if you will that we are in a position where we have such amazing features like that. Live AI. I think we should probably talk about a few things there. what's exciting. But also we should ground as I sound like the LLM. We should also have a grounded conversation around those kind of things. Emily Browning (05:30) Yes. You need to think about the cost. Peter Nicholson (05:38) so let's start with generally where do we see because everyone's rushing to this, aren't they? We see Cumatica now just doing the initial things in 26R1 with their LLM stuff. AI Studio is coming out 2026 R2 More stuff goes GA in 2027 R1\. We see more and more agents in BC moving from preview to release. So everyone's pushing towards this whole partnership between ERP system processes and sometimes jamming AI in there as well. But should we start with the stuff that we think excites us? What potential use cases immediately comes to mind with leveraging AI and ERP. I'm gonna throw this straight to Emily. Emily Browning (06:27) No, not ready for this now. Nirav Shah (06:27) Noooo Emily Browning (06:30) So I largely agree with what the AI said to us in terms of, you know, keeping that human ownership and treat AI as a helper. Part of me was thinking, as it said, you know, you're just trying to protect yourself so that, you know, the end users don't come and, you know, pour coffee over the AI machine so that it doesn't work anymore. to protect their own jobs or whatever. So there's definitely this kind of fear in business or in some business where we don't want to start relying entirely on the machines because if you don't kind of pass it for a human who's sense checking it then you know who knows what will happen and what kind of mistakes we could make. I'd like the idea of talking about maybe some of the smaller day-to-day things that can help us out, things like anomaly detection. I know there's a lot more exciting things that AI can do, but in terms of like, where could a business start kind of dipping their toe into the water? Things like that, helping out with things that your humans... would spend a bunch of time assessing, or you have to write some quite complex reports to get it. Where can you put AI in for that type of thing? Peter Nicholson (07:46) I see that same roadmap and when I when I think about anomaly detection, I think, well, anomaly detection is an easy place to start because it's really deterministic LLM stuff. Or deterministic AI, because it's really you're saying you kinda have to already tell it what an anomaly looks like. You kinda have to say it's using some machine learning. It's quite funny 'cause we're not even there and we're already talking about agents, which is agentic AI that can do more fuzzy grey area stuff. so I think that's is right to call out this as the first option, something like anomaly detection, 'cause it's a little bit more black and white. Nirav? Nirav Shah (08:25) Yeah, absolutely. I just want to go back to live real quick. And to me, I that was a great demo, but it almost felt like to me that I'm listening to the radio and somebody called in. like Emily Browning (08:40) you Peter Nicholson (08:39) This is a radio show. What what are you trying to say? Nirav Shah (08:43) we had audience call in and we're talking about AI, which is super cool. I mean, the direction that's going into I think it's a very unique perspective. on how you could have that AI jump in and start talking to you about things, like just a casual conversation. Someone radioed in into our podcast and wanted to talk about AI here, which was pretty cool there, Peter. yeah. Peter Nicholson (09:07) Alright, hang hang on, hang on. AI, you're back in the room. You're back on the show. Nirav has just said that you've basically just sound like a radio show that you've just called in on. So do something crazy that makes it sound like w this isn't a radio show. Okay, here's something a little Nirav Shah (09:20) Just say that this is our 50th caller. Peter Nicholson (09:23) crazy. Forget a radio show. You're in a kitchen. The ERP is the recipe and AI is this. my god. No, don't no, no, no. Nirav Shah (09:31) laughter Peter Nicholson (09:34) Okay, got it. Pulling back. Let's just keep us normal. What's the next Nirav Shah (09:37) You Peter Nicholson (09:39) angle you want? Cost, risk, or I don't know. Just s sing me a song around AI and ERP. Let me check on that for a second. All right, here we go. ERP keeps the business running, every order, every part. AI spots the patterns hidden, giving people a smarter start. Forecast invoices planning, helping teams move faster too. But if your data's messy, AI can't rescue you. Chorus, AI and ERP holding I think that I think that's enough. No, shut up. Shut up. Got Nirav Shah (10:12) Hahaha! Peter Nicholson (10:13) it? Stopping now. Shut up. Right, you're off the show. I'm sorry. Nirav Shah (10:17) You Peter Nicholson (10:17) No worries at all. Thanks for the cameo. Good luck with the rest of the episode. Emily Browning (10:23) She doesn't mean that this time. Nirav Shah (10:23) That's one thing with AI. She will never take anything personally, which is great. Take the emotion out of it. Yeah. Peter Nicholson (10:30) No, what's gonna Yeah, exactly. When Emily Browning (10:30) we know of. Peter Nicholson (10:32) the robots do take over, you know who's coming in. Emily Browning (10:33) They're going to be knocking on your door first. Nirav Shah (10:36) Yeah, I know. Oh my god. No, that was that was great. by the way, we have to name our fourth AI, um, receipt here. So we'll call them names. Anyone have a name out there in the podcast and the YouTube world? Any of our listeners let us know. We're happy to take recommendations. Um, yeah, anomaly detection, I think is one of the one of the easiest, right? Basically a listener out there and it goes and kind of catches things. for you on things that you've defined that, hey, this is a risk to my business. And I want to know it instead of you having to run manual reports and trying to uncover that data by itself. And this is, think, like, yeah, it's kind of AI driven and anomaly detection. But I think this is kind of like a no brainer in a sense, right? Like these systems are smart. Do I really call this AI? Do I not call this AI? I'm not sure. But. It helps you write like a vendors quietly raising prices by 12%, 5%, 10%. Just the upward trend like, hey, an almond detection figure that out for you, where you don't have to kind of schedule a report or decipher through all that information. So I think that's, that part is pretty cool. And then you just expand it out, right? These are just all these little agents working for you at the end of the day, helping you find gaps or leaks in a bucket, right? Because what are we solving here? Think of a bucket and there's these holes in a bucket, right? You've drilled all these holes in a bucket. You filled a bucket up with water and it's just, know, water is coming out these different holes. You're trying to patch up, right? You're trying to patch all these. Business is sticky. Business is not, you know, an exact black and white answer. Every situation is different at the end of the day. So all you're trying to is catch as much as you can. And that's where these agents come into play and they help you to possibly maybe not completely fill a hole in that leak. You know, one of the 20, 30, 40 different leaks you have out there, maybe just make a partial, you know, type of a fix to that leak. So you have less money going out for that specific inefficiency you have in the business. So, I don't look at agentic AI like, hey, it has to 100 % solve this problem. I think it has to 100 % get me to understand the problem. And then it's up to me on how I want to solve it at the end of the day. And that's where I think the AI agent did a good job. What are we calling this? Our AI guest that came Emily Browning (13:03) This is Peter Nicholson (13:03) Yeah. Nirav Shah (13:04) on that said that, hey, yeah, if you trust AI, And it's using data different than the way you would use that same data, you potentially make some catastrophic bad decisions for the business. Take AI, take the data, look at these anomalies, and then the human comes in the picture and help them, let them use that data that would have taken them maybe four or five hours, eight hours, maybe a week of time to compile, but now they make the decision. And in business, every customer that we've ever implemented will solve a problem and now open the door for another problem. So it's just always constant reviving revolving door at the end of the day. But those problems you would hope will minimize in significance over time. And you could bring more of that capital back into your business, reinvested in good people, reinvested in education. We reinvested in good ERP systems, that type of thing. It's kind what we ultimately hope that these AI agents would do for us when we look at the broader picture. Peter Nicholson (14:09) Yep. But I think it does start with the data. I ironically, we can use AI to help us with our data. not just anomaly detection, but I'm thinking around master data as well. So you know, years and years have gone by where we've had to use various Python packages for fuzzy matching. So you know you have one customer that someone writes in the system as all in capitals and then suddenly someone else does it in proper case. Fuzzy matching is fine for that. It often picks it up, but to write that code is an absolute pain and it doesn't see anything other than the text presented in front of it. I think AI is going to be really useful for Cleaning up master data because fuzzy matching as a technical term is one thing, but grey area master data is slightly different. And I see AI being able to do something that every company needs to probably do before they throw AI at it, is clean up this kind of stuff. So an AI agent will be able to do much better fuzzy matching because it can look much deeper. For example, You may have a customer that, and I'm not saying spelling mistake, or sometimes it's all in capitals. I'm saying it could be a completely different customer name. Maybe they've changed trading names, but their billing address and their ship to address and the contact are exactly the same. A Python package doing fuzzy matching. will not see those as the same company. We'll see them as two separate companies because their company trading name could be completely different. Whereas an AI agent the thinking model will go, well, hang on, they have the same billing address, same ship to address. These are two companies that look separate but it is actually one. There's some cleanup here to do. So I think AI should be leveraged for this kind stuff. to be able to move on to that kind of the step two, which is what we're talking around having good data. Emily Browning (16:12) I think that's a really good call out because we say a lot, everybody says a lot about how the data has to be good or you're going to get hallucinations, wrong recommendations, whatever. I don't think everybody's cleaning up their data. Cleaning up their data is a lot of work and it needs your end users to do it. And they've got full-time day jobs. They feel very pressured under what else they're doing. I'm... I'll be shocked if the majority of businesses are cleaning up their data before they start implementing AI tools. certainly whenever I hear about cleaning up data, it's just something that doesn't get done. It's something where we've been wanting to do it for years, but nobody's had the time. I very, very rarely hear of it actually being done. And I think there's a lot of pressure and urgency about implementing AI now, and businesses aren't going to accept, well, we need to spend six months first cleaning up our data. There's no way. So actually, that's a really nice solution. Kill two birds with one stone. Nirav Shah (17:11) Mm-hmm. Agreed. Agreed. Peter Nicholson (17:13) Yep. And I I've already seen this quite recently with a company that has upgraded their ERP from one major version to another. There's a great feature that allows them at a single click be able to phone a customer, right? So you can tell this ERP, well, it's a browser-based ERP, that you can tell it when you click a a phone like the little phone icon, it will open up your preferred app. whatever you're using. this particular company is using an internet based one like like Ring Central or something like that. you can now click the phone and it'll immediately phone that from the contact card. But because they have shit data, that's just not gonna work for them. Because sometimes they put the area code in. Sometimes there's a digit missing. Yeah, sometimes in the UK they've put plus four four. Sometimes they've not not put the dialing code at all. It's never gonna work. And I remember the them seeing this and they're like, that's great. I haven't got to kinda go into the system, then go to a different app to call them. That's great. And then you could see on their face they're like, shit. Our master data's Nirav Shah (18:18) you Peter Nicholson (18:18) not in a position where we can leverage this. And that's that's a simple case, but it's these like death by thousand cuts that there's stuff that you should be doing now to one, you know, be able to leverage these stupid little feet incremental features. But if you really think you're gonna be able to leverage AI when you you haven't got that kind of discipline of master data management, you're not gonna really stand a chance. And for me there's two levels. There's the whole kind of data. And we know that an ERP is a transactional database, but there's also the semantic layer above that. So I've been I'm gonna die on this hill, but I keep talking about the semantic layer when we're talking about the semantics of a business, right? You can't just go to a and we see it all the time in demos, don't we? Show me my top ten customers. Or define customer because it depends. If that's a business development manager that's asking, he probably means his customers. When we talk Nirav Shah (19:20) Mm-hmm. Peter Nicholson (19:21) about customers, do we mean, you know, every customer? One that's spends once every ten years? And they just happen to put a huge order in, are they actually your top ten? Or do you mean customers with recurring month over month revenue? Is that an active active customer? What do you mean by customer? What do you Nirav Shah (19:41) Mm-hmm. Peter Nicholson (19:41) mean by revenue? You know, d d does that mean gross revenue? Does that just mean you know, third party only? Does that mean e-commerce customers revenue only? Like what do you mean by revenue? And the semantics of the business is really what AI needs because otherwise you're just gonna get the shit you're gonna get the shit confident answers. It's gonna go, yup, here you go. and every single time we talk about AI, and I think me and Emily we we spoke about this one of our conversations a few months ago around the whole idea of trust. I think we were at a show and They were kinda asking the audience like what's the biggest thing? And the biggest thing is trust. People need to trust it. And you're never going to get that trust if you're not supplying it good data in the first place. Nirav Shah (20:26) Yep. Yeah, exactly. And, you know, I think you guys have so many, you know, had so many valid points there. A lot of points that I think our listeners are kind of going through right now when it comes to data cleanup, data governance, road to start, especially in the SMB space. We see this all the time with every customer that we deal with is, you know, they're coming off of QuickBooks, they're coming off of disparate systems. And, you know, everyone's kind of maintained a day their own way. But where you could start making better decisions, start putting together a plan on how to clean this data up, the master data especially. It's not going to happen overnight, but if you have a good tool and you start leveraging kind of modern technology, AI and all these other things, the goal should be, and what we tell customers is within the next month or within the next six months, start the process of cleaning up this data. Because really that's what an ERP system is here. for you, it's here to help you consolidate. It's here to help you clean this information for the next 20 years. Not to keep this data the same the way it came in. So I always tell them, look forward. Yeah, if you want to bring it in the way you have it right now because there's a lot of disagreement, people don't want to make some change, right? But then leverage AI and start coming up with a plan and chipping away. at finding these issues with your master data. And then over time, that data is going to be very clean. You're going to be able to use it in a lot better way than you were using it before. just this, you know, we talk about like data in ERP, the number one example I like to bring up is planning. And it's one of the agentic AI's, you know, there's a big movement around planning and how AI could help you kind of. Plan smarter, get better on top of your reorder points, let you know it's running out of stock and way in advance because it has historical information, your ERP now and all that stuff. It's great. What people fail to do most of the times is even if you have AI, agentic AI running your planning department, if you don't change the baseline planning policies like, hey, maybe we have a new warehouse and we could actually stock double. what our min-maxes were before. How is AI supposed to know that, right, without you going back and updating the master data? Because AI will just keep ordering your min-maxes that you have already defined without realizing that, hey, you know what, we have a whole new warehouse that's x square feet now that give me the optimal quantities that I should be able to go and stock in this warehouse so I can meet my orders in this specific region of the country at the end of the day. So as much as it comes to AI and helping you clean the data, the maintenance of that data moving forward is also important. What data do you maintain manually so that AI has something to use as it's trying to plan better for your business at the end of the day? that's not a conversation we struggle with customers. I think they get it once they're in an ERP system, like Acumatica or something. But that conversation during the implementation is the hardest. They don't want to change that data. They don't have time to change that data. There's too much going on in the implementation saying, hey, listen, we just have to bring this crap in the way it is, and we'll figure it out later. But that later never comes. So I'm hoping that these synthetic AI systems help at the end of the day to make that later come sooner, if that makes sense. Emily Browning (24:02) so what about where, where do I start? Where do I find my AI? I'm a business user, maybe I'm a customer. want to do something with supply chain. in my ERP, I don't just want to give, you know, Claude or chat GPT a log into my ERP and show it everything, expose everything that that's not, that's not the way it should work. so how do I go about. finding something to solve some of my supply chain efficiency problems. Nirav Shah (24:29) Yeah, would say using AI is important, but you need to know what you're solving for. And I think it always comes down to, and we did an episode on this, how do you start implementing AI? The first two steps were basically, let's get everybody on board with this initiative. And then let's talk about, based on the reports you run every month, what are our biggest challenges? Why are we always running across this issue or specific problem every single month? That's the easiest thing to go ahead and solve right away. And then you could have agentic AI around that and decide how you're going to solve it, see what that data is doing, anomaly detection, have better planning tools out there, for example. Procurement tools, for example, agentic AI has out there to solve maybe you know, who you're buying from, you know, make sure you're buying the best price, that type of thing. But it really comes down to what are you solving? You're not going to solve the world's problems, you know, in one go, it's just not going to happen. I think sometimes that's the unrealistic expectation. You really kind of have to bite off pieces of it as much as you could chew and make sure you don't choke. Because then you're going to really quickly realize, crap, I need 15 Deloitte consultants and nothing's wrong with Deloitte consultants. But I need 15 Deloitte consultants to come in. Yeah. I got to go ahead and spend, you know, I got to spend half a million dollars Emily Browning (25:46) Maybe the price. Peter Nicholson (25:47) There goes that yeah. There goes our Deloitte sponsorship. That goes in the same bucket as as QuickBooks. my god. Nirav Shah (25:56) on AI initiative when the ERP implementation was only $60,000\. Like it's like, oh God. Peter Nicholson (26:00) Yeah. I think back to what Ben Hussey from Katana said to us. use AI, leverage AI to say, right, I'm and and you can do this without giving it your financials from your ERP. Ask it, just go, you know, I'm a I'm this particular company. my typical customers in this industry, we manufacture these types of products but buy these kind of products. Where could we leverage AI the best? Well use AI to find out where you can use AI, use AI to find out because at least it's gonna tell you you know, what it knows from your particular industry are these are the typical problem areas we see. These are the things that might be of benefit for you if you used AI in this particular domain in your business. if you're someone, you know, here's some symptoms that you might be experiencing. of those, you know, is there anything that resonates? It's gonna ask you those kind of questions. Yes, okay, yeah. We do have an issue with visibility of sub assembly processes. Have that conversation with AI and see where that leads you. this is a question for both of you. Maybe I've got an answer. Maybe I've had this conversation with Claude without giving it any sensitive data. Where do I actually go then? I've got this clawed session open. It's told me, you know, these are the areas that I might have an issue with. What what what do I do now? Like great. I haven't really learnt anything. I've just got a session open with Claude that also understands I have this issue. Like where do I Nirav Shah (27:34) This is it. Peter Nicholson (27:35) actually go after that? No, I've maybe in B C what's BC doing with agents? Nirav Shah (27:42) Yeah, so BC has some agents pre-determined, pre-developed, unlike Accumatica where you could kind create your own LLMs and have it look at certain tables based on your configuration using AI Studio. But BC has some already pre-determined and out of the box now. They have sales agent that could basically be a listener in your inbox, bring in a order that it finds on an inbox essentially, and then pre-map. intelligently the data from the email to fields on a sales order or a sales quote. And it could be custom fields too. And then that order could also go through an approval process if you needed to. But really good for high volume. Distributors that get like a ton of different orders right through email. mean, think about how much time that would cut out if you could get 90 % of that order in place already. And now you're just spot checking. It also has what's called purchase agent. that helps you create purchase orders, manage the lead times, based off of historical orders already in the system. And a new one that they're coming out, which is kind of a two-part thing, is a purchase invoice agent and a expense management agent. So being able to essentially predict invoicing that's coming in from vendors. and doing three way matches to those receipts already received in from vendor POs, essentially. And then on the receipt side, helping to go through the approval process faster when expenses come in from different employees or so on and so forth to go ahead and predict what your budget should look like for expenses when those expense reports come in over time. It just starts listening and learning. that type of stuff. you know, there again, all these publishers experimenting, which one's going to work at, even though Business Central has four agents right now, they may decide, hey, you know, our expense agent doesn't even get used. It's actually pretty crappy, they may pull that back and no longer offer it like in a couple of releases. So it's, it's one those things like, you know, the ones that get the highest utilization is probably going to be in the software the longest, but they're going to keep having different different things. I heard something about that they're gonna have a quality agent. and business central, they're going to have some inventory planner agent in business central. you know, they'll keep kind of adding it to the bucket, but I like what Akumatica is doing, you know, where you're kind of self defining a little this of yourself, maybe need a little bit a little bit more technical expertise in house, maybe from your partner, or from somebody in your own company. But I do see this also, just kind of filtering out to standard functions like the demand requirement planning. Like they just have some maybe AI assist on that, right, itself. So, you know, the AI assist functionality is really nice. And I've been able to do some pretty cool prompts around that when you look at AI assist. But a lot of this evolving, it's all evolving. It's all just we're all learning as we go and trying to find the right utility for the right company. You know, the utility in one company is gonna be a different in a different company, right? It's not gonna be the same. They're not going to want the same, they're not solving the same problems potentially. So you need to kind of make it universal enough that you're solving for something more generic like anomaly detection, I think is huge. And then you make the decisions on, know, what problem you want to specifically solve in that process. And then how you're going to solve it is going to be more human interaction. Emily Browning (31:06) I like the idea of having these kind predetermined agents because one thing that worries me about AI in general, and don't get me wrong, think things like Akumatica's AI studio is really cool and there's a lot of good stuff you can do with it, but we're really opening the door to having all of this custom stuff that isn't supported by Avar or an ISV or an... You know, an ERP provider, it's, you know, we're just filling our systems with a bunch of custom programs. The person that is the, you know, the person that put it together could end up leaving us. And then I need probably them to, you know, it needs to be in their handover with someone new, but maybe we don't have someone new yet and, things like that. And, you know, we can build all these little programs to do little things on the cheap. rather than buying a solution, but then those are not live solutions that are being updated. And I'm quite conscious of the fact that my entire career in IT, which has been very systems focused, taking difficult things, making them better, a lot of that is rooted in You know, 20, 30 years ago, you know, this guy that retired five years ago, he built all of this stuff and everything that we do is, you know, all of these things that he's built and we don't understand any of it. And it's really niche. And, you know, we have this one, you know, this one external provider who's still able to help us, but, know, they're retiring in the next couple of years as well. And we need to just unpick this, you know, big mess. And, know, are we also use this other little tool that one of the engineers built? 15 years ago. And my whole career has been undoing those things and putting them into nice, clean, well-supported environments. So I think we're just on the brink of going backwards and doing all of that custom stuff. It's not necessarily the end of the world. Like things are cyclical. Maybe it makes sense, but there's a risk of creating problems down the line if you don't go about it carefully. So the concept of predetermined ones that are kind of actually fully supported by the provider is appealing to me. Peter Nicholson (33:22) I have two trains of thoughts on that. one for Nirav. I would love if someone like Acumatic with the AI Studio hopefully the at least the community comes together and says here's a simps system prompt, you know, if you're using this particular module, right? Because we know that our Acumatic has a bunch of modules. and then We could have s some kind of shared library that where you can download someone else's system prompt that's agnostic to the business and I I can just upload it and use it and now I can ask questions and it will be able to go, I need to use this, you know, I need to use this system prompt because I'm getting questioned around bomb management, for example. So I really like if maybe Acumatica can do something with some kind of library where people can Have at least ones that are vetted, or we can upvote good system prompts and people can kind of download them and import them into their Acumatica. Emily Browning (34:20) Nice idea. Peter Nicholson (34:23) my second point is that this is a concern because now it's so much easier to vibecode things, right? So before, like when we talk about tools that Luckily we have someone that can kind of keep alive but we're worried 'cause they're gonna be retiring in a couple of years. I've seen that before. I've seen it in Excel files that are used with a bunch of macros that someone twenty years ago made and now, you know, people who knew that guy, he's now retired but 'cause they're mates, you know, it's broken again. Any chance you can quickly fix it and he will. Like There's a patchwork of that and now anyone can vibe code something using an AI agent. Suddenly we have that same issue starting again. but my concern is that this is now worse than ever before. Because with prior solutions, oftentimes you adapted yourself to the tool because there's no tool that's perfect. So you know, companies out there that are using Excel for inventory management, they already go in that knowing that Excel isn't perfect for that scenario. It does Nirav Shah (35:36) Right. Peter Nicholson (35:36) most of it, it's the majority fine, you know, it's it gets us eighty percent of the way there and we'll flex to the twenty percent that it doesn't support. We know that Excel's not the perfect solution for it, but it will do. Notion is another one. With AI vibe coded solutions, where we're headed to now is now we can actually vibe code something that's a hundred percent what we want. It's exactly what we want. And when suddenly you have a solution that's a hundred percent of what you want today, you become even more reliant on it. If you go into using Notion for 80%, you're already kind of got the one foot out the door going, Well, I know if if Notion suddenly changed the way they operate or remove a feature, you know, I can probably find something else and adapt to something else. But if you get to a point where you've got an AI built solution that's a hundred percent, you're not going to be looking at all elsewhere. You're not going to be looking at alternative solutions. Now you're like, fuck. Like COVID's just hit. My entire industry has changed now. now this vibe coded AI tool is no good for me. I'm I'm down a dead end route. Whereas if you're using something that's pre-built like Notion, there's exit ramps. At least I can get off of something. Vibecoding I don't think gives you those options anymore. You d you are going down a dead end route. Even though AI isn't going anywhere, vibe code isn't isn't going anywhere. The people changing is gonna be a bit of an issue if you just think, we'd love a tool that does this. It doesn't exist. we'll just vibe code Nirav Shah (37:11) Yeah, I feel bad for publishers right now. Whether it's Business Center, Akumatica, NetSuite, whoever right now, because they're having a hard time just even getting their hands around their own ISV market and how they're managing upgrades with those ISVs out there. And then on top of that now, you got these vibe coding apps that people are just writing and throwing customizations in left and right, which is very risky proposition. And I think Emily said it. really well is that we're going to a kind of down a rabbit hole that could kind of put customers in like a corner where they can't get out of because they've so heavily relying on this app that they wrote or whatever that's not supported, right? That potentially has security risks around it and Next thing you know, there's a major update that comes up by a publisher. Now that app no longer works. Now you need to upgrade it. Or it dies on you because something's happened to the back end, whatever it is, or it exposed you to a security threat where you're getting ransom threats or whatever that is. But that puts you in a worse position. where you thought that AI is here to help you and make all these nice decisions for you, identify gaps and do all this, but you've actually got yourself in a position that you're risking the livelihood of the business now. And if you don't have that app, you're going to have very upset customers. So yeah, think that's a huge risk right now. Divide coding trend that's happening. And the publishers, like I said, don't even have their arms around the ISV market, the independent software vendor market out there, let alone now they have to contend with these vibe coders out there that just creating customizations and APIs and third party solutions that are supposed to work with their software, which is kind of scary right now. I'll tell anyone out there who's listening to his podcast, be careful doing some of that stuff. Weigh the risk. What happens if that person created that app is no longer with your business? What are you going to do? What are you going do with all that data at the end of the day? And we've always said it in our podcast. I've been pretty vocal about it. Only customize when you need to. Try to stay out of the box with these SaaS solutions out there. The amount of upgrades that they do twice a year, major releases, it's hard. The more customized your databases, the harder it's going to be for you to upgrade. It just is. We don't live in the on-premise world anymore. Peter Nicholson (39:39) Yep. I see this well you could I guess you could put it under the same umbrella as AI slop. Right, it's AI I think has Nirav Shah (39:45) Yeah. Yeah. Emily Browning (39:46) Hmm. Peter Nicholson (39:47) increased the amount of shovelware. We've heard always heard of the term shovelware. It's just vast quantities of of shit stuff, right? Now AI has lowered the barrier of entry because and I see this, for example, I'm a massive video gamer. You now see it on video game platforms. Like the Steam Store has never been full of so much shit. And it's just vibe coded AI games. free to play or out there for like $2.99 and it's just crap. Now we we run the risk of seeing that in software solutions that businesses are running as well. We've always had shovelware, we've always had people knocking out a load of vast quantities of solutions to your you know your issues and stuff but I think now with that AI element lowering that barrier of entry. we're gonna see a lot more of it. And and it's not just that AI can produce solutions for us, AI can also very well market those solutions as well. It's a very dangerous game. You can get something that looks very, very professional. but underneath it's it's it's just as terrible as anything else. Used to be that you could kind of spot shovelware shovelware from a hundred yards away. You're like, yeah, that looks shit. Because the website is all full, you know, mistakes everywhere. Nirav Shah (41:06) If it smells like... As they say, it smells like shit, it looks like shit. Emily Browning (41:12) You Nirav Shah (41:12) Probably a shit. Peter Nicholson (41:14) Yeah, I know, but the problem is AI has made it smell good and look good. Nirav Shah (41:17) Ha ha ha! Peter Nicholson (41:18) That's the issue. Emily Browning (41:19) I think, Nirav, you made a really good call out when you said, you know, we recommend keep to standard, keep it out the box. That's how you're going to have, you know, that's how you keep your timeline and your costs under control. And I still 100 % agree with that. Yet with AI, the message is kind of customize everything, make everything 100 % perfect for you. And my experience, and I'm sure you both have had it as well, is when you have It creates a change management problem down the road because when I've done projects to kind of migrate away from something that was, you know, written custom 25 years ago, everything our end users need is on one screen. Okay, that screen is just black with green text and they can only use the keyboard to do anything on it, but they can do it all in one screen and they know which buttons to press. Even then Nirav Shah (42:08) Mm-hmm. Emily Browning (42:09) they still don't like, you know, you're going to have some people resisting. moving to something new because now that thing isn't 100 % perfect, 100 % the thing that they want. Now they have to have two tabs open to do the thing that they want to do. Okay, that's a problem for future us, but it's definitely a concern that I have with all of this. Nirav Shah (42:31) Mm-hmm. Peter Nicholson (42:33) I actually saw it years ago when I was working for British Airways. I joined them at a point where they were changing two systems, their entire operational system. you know, the the check-in desks and stuff, the software they were using right there, they were they were also running two at the same time. And I know we we talked about why you shouldn't run two ERPs at the same time. Anyway, Their old one was the backup one and they had an issue where the new one was the one that I got trained on and it was great. I I knew I knew nothing else. The only reason why I remembered this is you saying that you you know, you could only use the keyboard for entry. Well, their old one, It was keyboard only. And when I joined there were so many of the the old people, right? The the people that worked for BA for like twenty, thirty years. And during my training, I had to shadow people. And I was shadowing this this one person, this old boy, who had so much tribal knowledge. He knew the old system, hated the new one, but he was still using the old one, even though the old one was actually there as a backup. Basically their new one was on was on the cloud, their old one was on prem. So if the cloud went down they could use the the backup, it was always synchronized in real time and stuff. He just outright didn't use the new one. But you went up and watched him he was flying through the keyboard. Like someone came Emily Browning (43:54) Mm-hmm. Nirav Shah (43:54) Wow. Peter Nicholson (43:55) up with such a curveball issue. Like, you know, I've got this issue and w you know, I need to D board this person but I've managed to reserve this other person. Can you put them on the same reservation booking? All of this, like crap, how the hell do you Yep, no worries. Through all the screens, you know, Emily Browning (44:12) you Peter Nicholson (44:13) muscle memory firing off. And that's really difficult. That guy is never, ever gonna embrace the new system whatsoever. He's gonna be stuck Emily Browning (44:21) Well, I look forward to meeting the new generation of those people. It'll keep us in the job. Nirav Shah (44:25) the Peter Nicholson (44:27) All right. one other one I think would be really, really cool that I think Agentic AI would be really good at. and that's onboarding new staff members. We already know that there's solutions out there. We well, one, we know that no one reads the instruction booklet. Right, no one ever reads in their you know evenings and weekends any SOPs or watches YouTube videos of how to use programs. No one's sad enough to do that. other than me to get a kick out of that. No one Emily Browning (44:57) other than you. Peter Nicholson (45:01) does that, right. there are some solutions. one that c springs to mind is ClickLearn ClickLearn make it a little bit more fun, a little bit more interactive, that you can do a kind of a follow-on with me and it will dynamically show you where to click to do certain tasks and stuff. my issue is it still needs someone to spend hours kind of putting together those kind of solutions. There's another one, I can't remember what it's called now, but essentially you can kind of record your screen. And it'll realize what you're clicking and kind of write the SOP for you as you're doing it. Fine. But it still needs someone to go through the motions of doing that. I think Agentic AI answers all of that for us. Imagine if it already knows, is already in your ERP. It knows how many orders you do a day, like if this item is a make to order that you need to do this process instead. It already starts to build up that contextual knowledge. I really think there's a a great case for agentic AI on onboarding new staff that can be very bespoke straight away to your business. It might be it can recognize this particular customer goes on this specific order type. There might be a business reason, right? This particular order type goes through a different flow. We can't invoice it like we do with our normal orders. It will detect "ah, this customer always, you know. Is this exception to the rule" I haven't gotta sit there and use my tribal knowledge or someone else's tribal knowledge to put that into an SOP or an instruction manual. The agentic AI already reasons over my data and knows this kind of stuff. I think that would be really cool use of AI in an ERP. Emily Browning (46:46) I agree. think that's a Nirav Shah (46:48) the way. Emily Browning (46:48) really valuable, a really valuable application of it. I mean, I see often that people don't have time to train other people or you could join, you could join the business, you know, maybe ordinarily they would have time, but you happen to join up in a particularly busy period or some, you know, some big thing has happened or someone gets sick, you know, whatever. And then the The one, the ability to train isn't there. And then two, that person's going to start getting a bad impression of your business if they join and they don't really feel that they're being trained very well. So you've got kind of the sort of time saving and money saving element to it as well as the sort of overall morale improvement for everyone as well. So yes, nice application. Peter Nicholson (47:38) And your business changes as well. It's is the upkeep of those. So if you are doing SOPs and instruction manuals, someone now that's their job now. Unfortunately, your job is now to maintain this forever, as the business Nirav Shah (47:48) Yeah, forever. Peter Nicholson (47:50) shifts. Whereas the Agentic AI solution can shift as your business shifts as well. And also imagine if you will. You're using a an a an an ERP that you log into. The Agentic onboarding AI realizes that you've not done a single thing in the ERP and you get a little clippy that appears and goes, it looks like you're new here. Would you like to start some onboarding? Nirav Shah (48:16) Yeah, exactly. Emily Browning (48:19) Mm-hmm. Nirav Shah (48:21) Yeah, I even like the fact that once it starts learning, knows the publishers agent again, knows the software, right, and knows the processes. But it learns how you're using the system. And then maybe over time, it recommends you to get more efficient in the ERP, right? Saying, hey, Instead of doing this drop ship process this way, why don't you turn on this automation schedule and do this, right? Well, eliminate maybe a couple extra steps, right? And starts giving you some kind of like process improvement tips, not just this is what you do next, this is where I'm stuck, you know, how I troubleshoot this, but really like, here's this other part of the software. Now it looks like you guys are really humming here and like, you know, Moving along, why don't you look at implementing this process because it's going to go ahead and eliminate three or four steps and increase your ROI and so on and so forth. So yeah, absolutely. think there's a lot of really good use cases there. Peter Nicholson (49:20) I think there's also speed for solutions on the develop developer side as well. I've had it I happily will say that I've used it to fix GIs in Acumatica. I've uploaded Nirav Shah (49:29) Right. Peter Nicholson (49:29) it to Clawed because I know that there's no custom tables. There's i it's it's just generic tables of Acumatica. there was a tweak that we did to a GI that did a certain thing that we needed and it didn't it Nirav Shah (49:42) Mm-hmm. Peter Nicholson (49:42) didn't work. Didn't It was like doubling up revenue in weird places, but not everywhere. I could have spent half an hour, an hour trying to pick apart where it'd gone wrong. And we know what the solution to bug fixing, right, sometimes is just process of elimination. I already knew this could be four or five re different reasons why this was broken. So I exported the XML, threw it at Claude, went, look. Here's here's my GI, it's in XML, it's Acumatica 2025 R1\. There's an issue. It's doing this. And I waited thirty seconds and he went, Yup, you've got a join issue. There's an issue in your schema. Here's the XML that's fixed. I downloaded it, imported it, fixed. All within the space of a couple of minutes. So I think it's gonna be really powerful also for the not just the end users, it's gonna be the people supporting this systems day to day Emily Browning (50:33) I think it does shine here in terms of development. We've obviously talked a bunch about the risks of that, but here this is development that we already would have been doing, already would have been doing in-house and would need to keep maintaining. But now we've saved a bunch of hours and that's great. Peter Nicholson (50:53) one of the I think there's a couple of other warnings that I wanna put on before we get too excited. One of them is around cost. Going full circle to our fourth guest that mentioned it. We've Emily Browning (51:06) Mm-hmm. Peter Nicholson (51:06) not yet mentioned it. the cost. So cost control is gonna be interesting. I've s already heard from some businesses that are using anthropic, they have an enterprise subscription where each of their named end users get $100 a day in tokens to use. If they use them up, that's it. That's them done for the day. Fine, that's one way to cost it. The other way is almost limitless, and you just pay per. thousand tokens from an API cost point of view. That for me is the dangerous one. And you go on Reddit and you already see the horror stories of people going, yeah, I got charged like fifteen thousand dollars because I was vibecoding something for my business. cost Nirav Shah (51:54) Yep. Peter Nicholson (51:56) control is gonna be definitely a danger zone, especially in the future. At the moment, we're in Goldilocks land. We have OpenAI in fierce competition with Anthropic for your money. And they're going to be selling tokens as cheaply as they can, often on subscriptions that are so much lower than the API costs. At some point, there's going to be a winner and there's going to be losers. And the winners are the ones that now start to name their price. So before you start AI-ing everything, you're gonna have to look at how much is this actually gonna cost me? Emily Browning (52:34) Well, quick build everything right now while it's Customise everything. Peter Nicholson (52:37) Yes. I am. I am trying to. I'm trying. Back to shovelware. Yeah. Nirav Shah (52:37) Hehehehehe Customized, yeah. Go, go, go. Peter Nicholson (52:47) So uncontrolled token usage is a big one. Nirav Shah (52:49) Yeah, yeah, that's that's the biggest big risk right now. Huge risk, huge, huge, huge. If I was a owner of the company right now and looking at leverage AI, right, I would first, you know, try to see what's available out of the box. And then, you know, only if it's absolutely necessary, you know, come up with some light, light, you know, products, and really understand how I'm going to manage those token usages. Otherwise, you know, Here you tried to save money by consolidating software, by coming onto an ERP system. And now you just increase your cost because you've I've quoted something that's gonna, you that maybe you're only using like 10 % of the time, but it's costing you like 80 % of your budget, right? You kind of have to understand, you know, where that kind of falls in your spectrum of efficiency gains and opportunity cost at the end of the day, when it comes to your business. Emily Browning (53:37) I guess one thing that is important also is making sure that people don't entirely stop using their brains, you know, just using AI to give answers to everything when you could just think about it for a couple of minutes and then answer that question. Cause it makes me think of, you know, the movie WALL-E, it's like we're heading that way. We're all, you know, eventually going to be shaped like gummy bears and the... You know, the computer is just going to do everything for us because we're, we're no longer, you know, our brains are going mushy because we're not, we're not using them anymore. Do I think it will be that extreme? No, but I do think you could definitely lose a bunch of knowledge and sorts of intuition within your business. If you were to just give, you know, if you were to give everybody some kind of AI helper and they can just always ask it the questions and it's always going to come back and answer, you're going to lose. a lot of valuable brain power there. And then how are you, you know, developing your staff, your talent pipelines, things like that, if nobody's doing that work themselves. I don't really know what the answer or the right balance is, because I do think it's good to adopt new technologies. It is here, it does seem like it's staying and it's going to keep developing. So I'm not saying we don't use it, but giving it freely to everyone to use for stuff that they could very easily do themselves in little time. That's something I'd want to try and keep an eye on in some way in business. Nirav Shah (55:12) Mm-hmm. Peter Nicholson (55:13) Yep. good. I think I think that's it. And I think the conversation was good because Whenever you're talking about AI, I always feel like I'm drinking from the fire hose. but I still feel that if I miss a day, I'm suddenly behind and I need to play catch up. I'm always in this constant catch up of of where I Nirav Shah (55:30) Yeah. Peter Nicholson (55:30) AI is headed. There's that that's the battle. The other battle is trying to see through all of the excitement and hype and potential solutions and like, look at this cherry picked demo of a wonderful AI agent So it's nice to have a conversation that's more about the real term issues that people will experience or potentially avoid if they listen to our conversation today. And I hope that if anyone's listening to this, they don't see this as us just being pessimistic about AI. I know you two are just as excited about the possibilities of where we can leverage it. It's our passion, it's it's what we do, right? That's why we work in IT, because we're passionate about Emily Browning (56:11) Mm-hmm. Peter Nicholson (56:11) this stuff. But it's nice to sometimes talk about, you know know, the real concerns that they're easy to gloss over if you're if you're only looking Emily Browning (56:22) Mm-hmm. Peter Nicholson (56:22) at the the potential glory that AI can bring. Emily Browning (56:26) I think these are the little conversations you have with people after the presentation. And then you go, yeah, that was really cool. But what about this thing? And I only see these kind of things being spoken about after the big shiny thing that we all need to clap and scream for. Nirav Shah (56:44) Mm-hmm. Yeah, exactly. And for me, from a partner perspective, I just look at what's going to make the user productive, what's going to help the user get the most out of their job, right, to do their job and do their job effectively and not just be data entry specialists at the end of the day. And we try to bring that perspective. to the table and whether that's integrating AI, whether that's looking at what the publisher has, or taking baby steps to get them from maybe not even knowing what computers can do because they maybe didn't have to put in any data into a system, but putting data into a system all the way to the analysis of it and all the powerful technology that's out there that supplements the ERP side of it. Peter Nicholson (57:31) Yep, very good. and I'm sure this won't be our last conversation on AI. I'm sure there's gonna be a of releases and new things that we spot out there and use cases and also just as many warnings around ERP and AI usage within ERP. Nirav Shah (57:46) Yep, I think so. I think you're right about that. Peter Nicholson (57:50) Maybe sh we should do that, like before we all stand up and clap and cheer at Sankit's like presentation around the AI. in January maybe we can go hang on. But what about this? Nirav Shah (57:57) So, Peter Nicholson (57:59) Or go there with a massive banner like that says uncontrolled token usage or or some kind of short Nirav Shah (58:03) Exactly. Peter Nicholson (58:05) thing that he can he can spot. Nirav Shah (58:08) Yeah. Peter Nicholson (58:09) Good. very good. as ever, if you got something out of this then please follow the show whether you're listening on Spotify, Apple, wherever you get your podcasts or even watching us on YouTube because we also throw these up on YouTube when we do some demos. and we we try to do some TikTok shorts, but I've been really bad at posting those. But you can be Assured that we'll definitely be posting all of our conversations that we have up on LinkedIn. So you can either find us on YouTube or a podcast platform or just search for one of our names on LinkedIn and you'll be sure to find us posting around the ABCs of VRP and beyond. We shoot this and record this completely for free. There's no adverts, no sponsorship, still nothing from Starlink or QuickBooks. Emily Browning (58:58) Thank Peter Nicholson (58:59) we stopped waiting by the phone. so we just know that we just do it for the for the love of it anyway. Emily, good to have you back. Emily Browning (59:07) Thank you. Good chat today. And with our guest, of course. Peter Nicholson (59:12) Yes, a fourth guest. my god. actually Yeah, maybe Emily Browning (59:14) The AI guest, let's not forget. Nirav Shah (59:14) And I'll close here. Peter Nicholson (59:17) Do you want to say bye to our viewers and listeners? Bye everyone. Emily Browning (59:22) Bye! Peter Nicholson (59:23) Bye. Well with that said, thank you both. and I'll catch you on the next episode in a couple of weeks' time. Nirav Shah (59:31) Yeah, you got it. You Emily Browning (59:31) See you then. Nirav Shah (59:32) might see my AI twin on the next one. just keep that in mind. So Peter Nicholson (59:35) Yeah. Emily Browning (59:36) Looking forward to it. Peter Nicholson (59:38) All right, thanks both. Nirav Shah (59:40) Later. Emily Browning (59:40) Bye.