EP 175: Kaden Webb - Differential Refresh over OData for Generic Inquiries

[Music] Welcome to the ERP SEAL podcast. My name is Tim Rodman. We are live today on Friday, June 12th, 2026 and this is episode number 175. We got it up on the screen there. You can find related links for today's episode at ERPseal.com/episode15 And even though it only says "Kaden's name" on there, we've got both "Kaden" and "Lisa" We'll get into why that is, but I want to do some little intros to start off And then we're going to dive right into a super geeky topic, which is doing a differential refresh over OData for generic inquiries in Accumatica or in this case a flavor of Accumatica, James Software. So and also I just realized you guys are actually the first guests on the change of name from the AVGFORMS.com real talk podcast to ERPseal.com podcast. Same thing just a new name and I just realized as I hit the button I need to change the color so it's still figuring it out, but I got the thumbnail at least updated with some new colors. So thanks for being the first victims on a rebranded podcast name and maybe to kick it off all I'll let each one of you introduce yourself. "Kaden" maybe you could guess just a brief, where'd you grow up? Where'd you go to school and how'd you land into this funny world that's been marketed ERP? Yeah, absolutely. I appreciate you having me on Tim. I'm excited to see what you could do with ERPseal and switching over from AVGFORMS. I know a lot of us have been quite far into AVGFORMS and we're curious to see what you do with everything. So I'm from Colorado. I went to school at CU Boulder, learning finance, business and computer science and now I'm taking that into the ERP world. I actually started in SAP, moved into BST Global which is an ERP for architects, engineers, and construction and then finally found myself into the acumatica world. I don't know, BST Global, haven't heard of that one. How do you like acumatica compared to the other two that you have experienced with? Yeah, I love acumatica much more than the other two. BST Global, it's very specialized for a unique space and it doesn't have all the aspects that you would find in acumatica like a manufacturing and all the other stuff. It's mostly project services focus on engineering projects. So much different field, much different capabilities and acumatica is definitely my favorite of the three. Awesome, awesome. Well yeah, thanks for coming on and the reason why your name is the only one in the title is simply a character limitation. I can only fit 75 characters into the title so that's why I did that but you were like, "Hey, we all need to have Lisa on because I guess we connected online because of Lisa." And so Lisa, thanks for joining us well. I'll let you give a you to Mike there and give us your origin story of where you grew up, where you went to school and how you got into mid-market ERP. Super stoked to be here. So thank you very much. And I cherish this community for a long time. I needed it. I grew up and live in Oklahoma. I was a James prime client and I went to college in Florida. That was just a side part. But then I became a James prime client and I needed you to teach me how to use James prime because I got hired to do the reports and didn't know how to use acrimatic at all. So I needed to learn very, very fast and it was a new product for the company at the time as well and no one there knew how to use it. So I had to figure it out really fast on my own. So I used you and your community and it was great. And then I just became addicted to it. I love the product. I loved James prime and I then began to search to try to get employed by James prime and I'm super excited. I love this product. So you're at a customer site and now you're working for the publisher. And we should say so we just did. Oh, I can't remember now. Last episode or two episodes ago we did an episode on James and their AI assistant. But catch people up. You don't know what James is. It's basically acrimatic with its own project accounting modules. That a good way of describing it. Otherwise it's just normal arcomatic. Yeah. Yeah. So and that's built all within the acrimatic framework. So it's not like this bolt on thing. It's it's native, but they don't leverage the regular project accounting module at all. They built their own. And it's it has to do with government contractors and something called DCA compliance. And there's a kind of I didn't realize how much headache goes into that on the infrastructure style. Like there's I'm not going to say the dollar amount, but a shocking amount of dollar amount investment that goes into making an infrastructure that the government gives their thumbs up to essentially. It's fed ramp. I think is the certification that you get for that. And so it's a niche to be in. And that's what James is. But it's all built on the acrimatic platform. So yeah, exactly. It's it's the same. It's the same skeleton underneath like you have your generic inquiries, business events, dashboards, all the fun stuff that you have in acrimatica. They've just tailored it specifically to the GovCon industries. So when we talk about like a lot of the similar issues and use cases between James or acrimatica, a lot of the same rules apply. So to speak, especially from reporting. Right. Yeah, exactly the same. And even functionally, like the finance module, the finance module, the distribution size, the distribution inventory, like all that's the same, right? It's just the project accounting area. If I understand it, right? That's different. Yeah, that's correct. Right. So when we go ahead, yeah. Yeah, so when it comes to like a generic inquiries, business events, O data, we can speak to both communities in a way. Perfect. And that's what we'll do. Yes, this is relevant to anyone in the acrimatica space before we just dive right into the geeky side of things. Lisa, you said something that I wanted to ask you about. So you said something about community and I needed it. And I tell me more about it. I like that phrase because for me, working in front of a laptop and so much of us work remotely nowadays since COVID, it's going to become the norm. Like it's great. It's convenient. I think there's more pros than cons, but there are cons. And one of those is you feel disconnected. And yeah, talk to me about what you've met when you said I needed it related to community. I was thrown to the wolves. They said create these queries. We need reporting. We need dashboards. We need things for our contracts. And I didn't know how to do it. I didn't know how to write in inquiry. I knew what a query was, but I didn't know how to write in inquiry in acrimatica. And I didn't know how to use report designer. And I just didn't know how to do any of it. And I was having errors. And I would go to your community and find other people that had that and figure it out based on their postings. They would even share generic inquiries. And I would get generic inquiries from the posting it. It was just an amazing community at the time in which I needed it the most. Love that. Love hearing that. And so you got some technical help, but it's not the part that really resonates with me is I almost hear like the emotional health as well. Just being like, hey, I'm not completely alone in this, right? Yeah. I saw other people having the same. It's like I didn't know where to start with this. No one even in my company knew how to help me. So I was on an island. That those were my people. Those org forum people were my people for a while. Those were the only there was no one else at my company writing generic inquiries other than me. So I didn't have anyone but that community. And there are several people in that org forum community that are regular posters. So those were like my little superheroes that I would read and figure out what they were doing. So if why it is doing it, Lisa Ford's going to try to do it. So it was amazing. It's probably why a lot of people are probably sad to see org forums get a deconamation in a way because it's useful. We like to see the information from it. And the interesting thing is you know how AI web searches work of how whenever you ask a question about acumatica, how it pulls that from a web, org forums has become like a source for that information whenever we ask chat GVT or Gemini or something, how do you do this in acumatica? It lists isn't it sources often that this is org forums. So there's something useful about the information on there and how it's training these models and helping people search for the answer, even if the form is closed, the information is still there. So, I want to ask you more about this because we're going to dive into the different refresh OData geekiness let's get there, but because this has been a recent change, like I just sent out the communication blast earlier this week on, and I wrote up a whole big long blog post, actually let me put a banner up just so we can have it here officially on the screen. It's oggforms.com/thankyou, I'll put it up on the screen there. So anyway, just because this is recent, I'm just curious, when you say some people are sad to see it get decommissioned, all I'm doing is moving it to read only mode. I look at the recent posts and I think the last one was a month ago, the one before that was two months ago. So there hasn't been much posting lately. I think community.accom.com that's a better place to go now. I'm just curious from your standpoint, I'm keeping it read only mode so you can still find all the posts, nothing changes there. What makes you sad? Why wouldn't you just post on [email protected] for anything you need? Yeah, there's something cool about the independent community of it. You're one of the leading independent voices in the acumenica, if not the most leading independent voice. So the fact that you are able to create your own acumenica community is pretty cool. But yeah, the new acumenica community is incredibly useful. But the fact that you put it in read only still, it makes it useful that we can still search Google or search these AI models to still gather that old information people resolved. Interesting. What are your thoughts on it, Lisa? Yeah, I'm still curious how people think about it. With Kate and I do see a need for both though. We have our own James Prime acumenica community for our users as well, which is a great resource for the same reason. When you're searching in AI and you want an answer, you want some help. You're what I used to be the first day novice. You need someone to walk you through those. You need a community. So I love dog forums. It's my place with my peeps, but I'm also happy to see them to use the acumenica forums, because I'm not an actual acumenica client. So I can't post in the same way acumenica clients can, but I can in my community. But I like them both for different reasons. They have their pluses. Interesting. So if you are a James person and you need a community, reach out to that James Prime acumenica community. What is the URL for that? I'll put it up on the screen. I'll get it for you. You can drop it in the private chat or the comments, whichever one of which you do. I'm curious on your thoughts. You both use the word independent when talking about the art forums community. I'm curious your thoughts on something I'm going to put on the screen here. At AI, create this image. For me, this is how it feels psychologically when I'm thinking about the difference between vibe coding and AI agents. I think there might be an application of what the community says. I'll put this on the screen. To me, vibe coding is when you just sort of sit back and you're just having fun. It's a hobby or prototyping something, just seeing if it's possible. But then to actually make something, AI agents are technically a different thing. AI agents are something that's working on their own and they're supposed to be more valuable, etc. But for me, feeling wise, it starts to feel much more corporate. The image here, the clipboard says efficiency, scale, control, and the coffee mug than the clipboard tasks, workflows, data, KPIs, optimize, performance metrics. And I even like the human irrelevance index that AI is doing there. So there's this corporate feel. The reason why I put this image on the screen is I wonder your thoughts on the difference between an independent forum versus an official publisher forum. Do you feel like there's a similar thing? One, you feel like you've just got the Hawaiian shirt on, you're on vacation, whereas the other one, you feel like you're at work with some professional tires. Is that part of what it is? Yeah, and that's honestly a great analogy. I even think about the difference between the SAP ecosystem and the acumenica ecosystem, two, like SAP is this big giant. It's very strict about what you can and can do, whereas acumenica is like more enabling for individuals and individual companies, the small to medium-sized business market. So the individuality is something I appreciate. And this meme right here, it does speak volumes, especially for a community like yours. At least your thoughts. I think he's nailed at all of it. That's exactly how I would, if I could draw, this is how I would draw it. That would lay it out. Yeah, I feel similarly. But I'm never sure. Am I just weird? Or yeah, it's good to hear that you guys feel the same way. I mean, it's like midnight some nights and I have a problem. I'm trying to think. I feel a little apprehensive about posting that on it. Tell me more about how it comes. Do you feel like you don't want to look dumb? Is that part of what? Yeah. Yeah. But you don't feel bad looking dumb or forms. Because everybody else is having the same issues. And they're just my same. We don't know what everyone is. I can't make any conclusions about you. I just, I'm getting help from you or I'm helping you or that type of thing. Love hearing that. Interesting. Something this meme reminds me of is the difference between the AI models, like a Chad GBT, Grock, Gemini or Claude. Grock is much more casual. You'll be shocked at some of the things Grock will tell you. And Chad GBT, along its evolution, it went from being more open. I'll say whatever I want to see to be in corporatized, very structured and cut kind of like the AI agent image you have on the right side. And so it's interesting to see even in the AI models, like which ones are more personable, casual versus which ones are strict and lose that human element. It is super interesting. They do sort of, they begin to take on a personality, don't they? Oh, yeah. Absolutely. Yeah. That's fast. Correct. Which I think is going to have incredibly negative mental health impacts on the next generation. I'm super sad about that, but I'm using it and I am very intentional. I use the word interaction now. That's the word I've decided on at least for now. I've really been trying to train myself to not say the word I talked to Chad GBT about this or I had a conversation with Chad GBT. I specifically am trying to force myself to use the phrase, I had an interaction with Chad GBT because I want to keep that line there that this is not a person. This is a machine that I'm interacting with. Yeah, it sure starts to feel like a person or maybe it doesn't. Yeah, and underneath these LLMs, it's just computer modeling. It's like talking to a calculator. Like, no one's going to say you talked to your calculator. Totally. Totally. So, we'll see where it goes. There's definitely a dark side, but it's also fun. I haven't had fun with the AI stuff as well. But I think with all that, oh, I wasn't going to put one more thing on the screen. So, on the topic of fun, community, vacation, Hawaiian shirt style, I am keeping that going at orgforms.com/discord. So, if anyone wants to still have that bar instead of office or vacation instead of work vibe, that's definitely still going on just not on the official web forms. It's in a discord server. And I've actually been enjoying it better. Anyone can join that. It's more flat. Instead of the form structure, you have a topic, topic, topic, and you got to go into a topic to see it. It's just more flat. Like, everything is more conversational. And so, yeah, we'll keep that going. And anything, like my shutdown, org forums, or put it in read only mode announcement, has actually reinvigorated the discord side of things. It's only been around for about six months. Anyways, it's kind of funny that my saying this is ending has actually started instead, just over on the discord side of anyways. We'll keep that going for that more Hawaiian shirt atmosphere, I guess. Awesome. So, it looks like the community still continues in grows. Yeah, exactly. Yeah. So, the ERPCL.com, I'm broadening the focus to be been market ERP in general on topics like reporting, which we'll be talking about today. But then the org forums discord server, that'll stay focused on the acumenic or crowd. Our people, right? Our roots, yeah. Our people, the acumenic of bar. And I've seen your discord. And one of the episodes you're talking about how you used AI to create a bot in discord. Oh, yeah. That's a good one. Yeah. Yeah. Yeah, I did it like as a prototype. Yeah, but yeah, discord opens up the opportunity. There's all sorts of cool technical stuff. that you can do in this court that I couldn't do in the WordPress plug-in that I was using on org form. So yeah, from a technical standpoint, it's better too. But not even, I'm not even doing any of that. Yeah, there's bots and different stuff you could do in this court, but just from a like, "Hey, I got a notification on my phone when someone posts standpoint and I know I'm interacting in real time without hitting posts, waiting for it to say, you know, it's just more instantaneous. Just from that basic standpoint, I've been enjoying it." But, yeah, I don't know, do you have any knowledge on this court bots or any other technical things you could do there? Yeah, absolutely. You could even have like an AI assistant that almost joins as a member in the chat. Like, this is the accommodica assistant bot and you can tag it and ask it any question you want at any moment. Maybe some community members are discussing a topic, have questions or are confused about something. They just say, "Hey, assistant bot, what's the answer to this accommodica question?" And then the AI just comes into your chat and answers for you. Yeah, there's all sorts of, and it's a video game origins where Discord came from. So, there's all sorts of geeky things you could do in there. We'll see what we wind up doing. My thought on the bot specifically is that I probably created dedicated channel to that because I want to keep the bar human, that's the overall vibe I go for as human communication. But, yeah, we could tinker with stuff like that in isolated areas on there. Yeah, there's also such a cool technical stuff there. For now, it's for me, it's a human-human connection mechanism. That's what I try to focus on there. That's just me personally. Yeah, absolutely. So, 20 minutes in, but I think that was all fun to talk about. Let's dive into differential refresh, huh? So, differential refresh over OData with generic inquiries. I love how easy it is to connect to OData. I could do it from Excel. I could do it from Power BI. They know how to speak OData. They know nothing about Akumatica. They don't care. But Akumatica can speak OData. And OData is the way you pass data between the two. But the problem is that when you start to get on larger data sets, when you're dealing with hundreds of thousands of records, that's usually when I run into it. OData starts to time out on you. So, the refresh will just time out or just take forever. It'll log your system down. And the more Akumatica is moving into larger, larger businesses. And, James, I feel like your average customer size is larger than the average Akumatica customer size, at least when it comes to transaction volume. So, you deal with this, I think, more than the average Akumatica customer does. But bottom line, when you start to get into that, you need to be able to refresh a large data set. And the best way to do that is with a differential refresh, where it's only refreshing the data that has changed since the last time it pulled the data. And this is the same technique that Velicso uses, which is a popular Excel add-on in the Akumatica space. They call it smart refresh, I think. But they, when you hit that smart refresh, it only pulls in the data that's changed. And so, you're still getting the up-to-date data, but it's not having to pull the whole chunk every time. So, that's what we're talking about here. And what do I miss? Help me out there. It sounds like you described it pretty well. Like you said, if users run these large inquiries or large reports that pull entire data sets, maybe from the beginning to the end of time, maybe just out of, because they didn't know otherwise, they didn't know like this is going to pull a million records. And it's going to bog down the entire system, or have everyone say, why is Akumatica slow today? We have no idea that someone ran all million records from the beginning of the end of time. This is when that differential refresh becomes important. To save that production server, possibly move it over into a staging area or a data warehouse and have reports like Power BI be hitting that server instead of your live production, because people depend on this live production site for business services. At least, I think you have some, you have any real-world, painful stories of people refreshing ODATA in large terms? Absolutely. Yeah, lots. So, I don't think that they realize that they're, when this happens, that they're on one CPU. So, you know, Person A is accessing reports in Power BI and they don't have their settings in Power BI set to differential refresh. They don't realize that that's a setting. And also, the inquiries need to be set up to do that. And to do that, you need to have something in the inquiry like latest, last modified date time. That's what it's using. That's what you need to do. You don't even have to mark it visible on the UI or visible. You just need to have it in the inquiry. So, they don't realize that. So, they will affect processing like labor batches by doing these refreshes. And by also not having their Power BI Excel or Velixo set for those smart refreshes or differential refreshes. And granted, if you're in Excel, you kind of have to use Power Query to get that done. It's not capable regularly. But with that setting, you want to help yourself out and not refresh because like our job costs will have millions of lines. You don't want that. So, I'm going to put it up on the screen. Go ahead, Kate. I'm looking at something like yeah, go ahead, Kate. And there are a large number of limitations inside Accumatica. If someone does run a massive generic inquiry from the beginning of end of time for a million records, how do you even put that to an end? And there is no just stop this button. You can look at the resource monitor or the event log, but there's no stop this report button. So, once someone runs that report, the production server is going to get bogged down and you don't want labor processing or whatever is going on to be affected by any of those mistakes. Oh, yeah, I didn't realize that. I guess you have to stop it on the Power BI or Excel side. You can't kill it on the Accumatica side, is that what you're saying? And often you can't even kill it like on the Excel side. The way you would kill it. Since Accumatica runs on top of SQL, you would literally have to run like a SQL command or a database command to say, no, kill this query. Gotcha. Okay. So, if you do that in prime, you have to contact me and I get it killed for you. Got it. So, if it's slow, if you're having that, then it's not just affecting you. It's affecting all of your users. So, that's something to know. So, it is time to reach out and get that killed. Now, you say prime, you mean, James, James, right. Right. And so, it might be similar. You know, I don't know. Accumatica has done some stuff recently with system resource monitoring. I'm not up to date on that. Maybe in 2026 or one, they have it. James is always a few versions behind. But yeah, it might be the same thing with Accumatica. You might have to open a support case to get a long running process to stop. But the field we mentioned earlier, I'll put up on the screen. I looked it up to confirm it. It is called last modified date time. You're going to find that field all over the system on the vast majority of data access classes that you use when you build a generic inquiry. And like you were saying, Lisa, as long as you put this field on the results grid tab in the generic inquiry, you don't have to make it visible on the screen. But as long as you put it in there, you can access it over O data. And this is the one that you're going to want to leverage to do the differential refresh. But like so does the same thing. And the Accumatica framework does a great job of actually making sure this field gets updated every time a record gets updated. So it's very reliable. And this is the foundation of doing a differential refresh, right? Yes. Absolutely. Great point. And even just having that column or that deck field inside of your generic inquiry in and of itself is not enough. It needs to be parameterized or even doing checks outside of Accumatica and differential comparisons, perhaps use in Python or using SQL Server integrated services to make sure you're only capturing those changed records. But merely just having last modified date inside the generic inquiry. That's how you make the field available. But you need to hit the parameter in a way. So if you're doing O data, you're able to do, I think it's dollar sign and the word filter to apply a filter to the O data call. And that's where you would use the field there. Or I think in more recent versions, can't remember when this came out, you could do something on the actual parameter tab in the generic inquiry and pass a parameter value in there. You do it that way, you have to make sure to use the parameter in the conditions tab, a parameter by itself does nothing. You have to use it in the conditions tab. It's like a wear clause in SQL. And then you apply the parameter to filter the data. That's it. Get that right? Yeah, absolutely. You'd have to create that parameter to make it available to hit with a way filtered dollar sign. So you'll create the last modified date parameter that allows the endpoint to hit it. And then like you said, it goes in the conditions tab to make sure this generic inquiry is only giving records that apply to that last modified date. And oh, go data is just like APIs, honestly. ODATA is basically an API applied to a standard protocol, the open data protocol. It's like a simpler API, right? And it's a flat API where it's always returning a rectangle of data, whereas an API can have this whole nested tree structure that gets crazy. ODATA is like a real simple API that's just like, here's a table back, right? That's my way of thinking of it. >> Yeah, APIs can be developed and built to do all kinds of things to interact with a software and interact against that software. >> ODATA is one standardized type of API call in a way that everyone, each system follows that protocol to pull that ODATA. >> And I look, it works real, it's real simple and it's designed for reporting, right? It's not designed for writing data back into the system. >> Right, it's designed for reporting, using get data inside Excel or inside Power BI to bring that data into your other software. >> So we talked about the generic inquiry, ODATA, the last modified date time. That part, I've done all that, so I'm familiar with that. I've tried several times to get a working differential refresh in place and have never gotten to the finish line. So I don't remember where you mentioned it, Kaden, LinkedIn, somewhere, you're like, I've done it and I was like, wanna talk to you about this. Before I get into your more geeky way of doing it though, I wanna mention that up until now, the only way I've known how to do it is I recommend a product called DataSelf because they have like just a simple app that allows you to do the differential refresh without writing any code. So if you just want the easy way, look at DataSelf, they have something called ETL+ that allows you to do a differential refresh over a generic inquiry. But before we get into your more technical way of doing it, Lisa, you said something about people don't know about the checkbox and power BI or something, but I'm not familiar with that. What were we talking about then? - They don't know how to make sure that you're refreshing, either on a schedule, you can do that or you're actually applying that differential refresh. If you're on just a standard subscription, it's a little harder, but if you're on a professional subscription, it's easier to do. And I have-- - I don't even know this. - Yeah, and the full look so is the same. I don't right now. The list is-- - I will get it while we're talking. But yeah. - And if we don't get it on the episode, I'll put it in the related links at erpc.com/episode175. But you think just your gut feel on the fly here-- - Yeah, they don't know. They don't know. - Is it just pro license, the $14 a month license? - Yeah. - $23,000, you need more. - Yeah, you can just do it easily with that. Yeah. And then the other thing that they don't realize is that when you put those volatile expressions in the GI, you're just killing that different. Don't put those. So let's say in the-- - And they can take longer. - Yep. - Okay. - So you don't want your volatile expressions in your GI, you want those in your valixial sheet or your power BI or your Excel, you don't want them in the inquiry. - So on Power BI specifically, yeah. If you can find it on this episode, great otherwise we're-- - Look at it. - But you do think it's just a check box somewhere and that'll handle the differential refurb? - It's a setting. It's your setting. It's your choose to do that. Yeah. And a lot of my clients, I'll say, you're running a lot of batches of blah, blah, blah. So it would be better if you don't need it live all the time to schedule it. And so I show them how to schedule it. And it's those settings, it's your choices. It's not as great in Excel, you can't do it as great. There's a few smart or enable because what they don't realize is hey, I'm in Excel, every time I click into a new cell, I'm refreshing. They don't know that. - Well also, I don't think there's a difference to refresh in Excel. You're just talking about my party at com. - Yeah, the only way to do it in Excel is power query. - Okay. Well even in Power Query, I haven't, and this is the part I tried. I haven't found a way to do it in Power Query. And I still don't understand but as something to do with query folding. Oh, data, that means the macchi-matica doesn't support whatever Power Query needs. Have you found a way to do it in pure Power Query in Excel? - No. - Yeah, so we're saying at least checkbox wise or easy wise, there's no pure way to do it in Excel, but you can do it with a checkbox in Power BI.com. - Well, you can set your settings. - Okay. - Just a checkbox, but I do have that. I'll get that out. - When you get it, maybe it'll make more sense. But if it's as easy as I'm hoping it is, Kaden, I'm now much less enthusiastic about what you're gonna talk about. Well, let's talk about that. - You still need that. So that's what I was gonna say. So you still need your data warehouse option, your data self or Kaden's methodology. - Okay, so the difference where refresh you're talking about is having Power BI pull from like a SQL database over a differential refresh. - Yeah. - What you're talking about. Okay, all right, yeah. You can't go straight from Power BI to Achaematica OData. Okay, yeah, I've run into the same thing. All right, so Kaden, I'm still enthusiastic about this. All right, so we're talking about now, instead of going straight from Power BI.com to Achaematica OData, we're gonna put something in the middle. Is that right? I'll let you talk about that piece, Kaden. - Yeah, absolutely. And that piece in the middle, it could be a staging database, a caching database or a data warehouse, as you mentioned. And instead of having Excel directly, just do a full data pull against Achaematica, Excel would be hitting that database server instead to alleviate any draw from the production server. So there's multiple databases you can use. You can use a SQL server database. You can use a Postgres database. There's benefits to each. I would, I recommend like Azure SQL because that's in the Microsoft architecture structure. So if you're using the Microsoft products already, like Achaematica is a.NET type of software already in the Microsoft structure. It just fits naturally. And there's variations of Azure, like Azure government. Maybe you're a GovCon company or your FedRAM for GCC high. You would use Azure government for that compliance. - No, it makes sense. Yeah, and then you're also when you're going to the Power BI, part of it, you're not even leaving the Microsoft Cloud, just from one part of the Microsoft Cloud to the other from Azure to Power BI. Is that usually the first one you reach for when you realize, hey, we really need a, I'm just going to call it a data warehouse or SQL database. You tend to reach for Azure SQL. You also mentioned Postgres though with what are your thoughts on differences? - Yeah, so Azure, it's the Microsoft SQL structure and it's in the cloud or you can just have an on-premises SQL server if you want. They just manage in Microsoft SQL server services. Pull against that. The time that you would use Postgres, Postgres is more like an open source side of things. If you're running Linux, you just want to have a flexible database that you spin up, not so formal and inside the structure like SQL. Postgres is amazing and many people use it for their software. - Not sure if you guys have picked up on this, but I've been hearing that Acumatica is actually moving towards Postgres and replacing Microsoft SQL server with that. I don't know if they'll actually replace it or it would just be another option, but yeah, I heard that as part of the whole modern UI, the move to.NET Core, the ability to run on Linux, Apache web servers instead of needing Microsoft's IIS. That's what I've heard because you would need some database engine in that world, but yeah, we'll see if it actually happens. But yeah, it's a popular one. I know ODO runs on Postgres. And so yeah, what would be the difference? Like other than it being cheaper, if you already have a Linux infrastructure in place, then you could just spin one up for free rather than paying the whatever subscription, the Azure side, other than the cost part, what are some differences between the two? - Yeah, like you said, the ability to run on Linux, and that's actually super interesting you mentioned about how Acumatica is evolving to try to be able to run on Postgres or Linux because it's always been confined into Windows or Microsoft. But Postgres does kind of represent like the Linux world of things and compared to SQL. That's how I think about it. - So a lot of it just has to do with current environment that you're working in at that particular client. And if there are more, already have a Linux environment set up versus no infrastructure, you'd probably reach for Azure SQL, is that fair? Right, and a lot of this will run on like a virtual private server. Maybe you spend up a VPS. And these VPS sites, maybe it's AWS, you can instantly install and bring up a Linux computer. Since Linux is free, you just install Linux, it spends up a VPS for you and you just quickly install Postgres on it super fast to get set up. Whereas Windows you'd have to pay the licensing fee, it might be a little more difficult to set up. Makes sense. I'm searching here just to round out that Postgres side. Let me put a screenshot up. So this is from the past, Akymatic Summit in Seattle about five, six months ago. So here's a slide from one of the sessions. Almost there. Put that up on the screen. So yeah, it wasn't just a rumor. This was actually a slide from a presentation. I'll read it for the sake of the audio version of the podcast says core technology, Postgres SQL support. It's more cost effective, the Microsoft SQL, Postgres SQL outperforms mySQL, which is also open source on performance and scalability. The project was initiated in 2024 and it's approaching the final stage. The beta is available. I don't remember this in Akymatic of 2026 R1. That's for non production, district tinkering. And the official release is targeted for Akymatic of 2026 R2. So that'd be cool. You can see it's kind of blurry, but in the screenshot on the right, you can see the server type drop down has an option there called Postgres SQL server. So that'd be cool. So you're going to spin up some database, whether it's Azure SQL Postgres, and then this is the hard part, I think. How do we then different refresh from Akymatic of OData into that database? Talk me on that. Yeah, great question. And this is when we build an ETL. And an ETL is an extract, transform, and load. The extract, we're pulling it from Akymatic. We can transform it in any way if we need to, especially if you're trying to do an integration to another system. You might need to change the way that data is formatted or presented. And then you load it. And the load goes into our new Postgres or Azure SQL database, whatever you prefer. And there's multiple ways to run ETLs. There's lots of ways to skin the cat. You can do ETLs with Python. You can run SQL server integrated services where you're sort of building that flow. You're getting the data, you're transforming it, and then you're loading it into your staging table or your data warehouse. You find that you reach for the SQL server integration services. Or I think there's a more modern version of that. Is that the right word for it now? Yeah, there probably is a more modern version like in the Azure landscape. I'd have to research a little more on that before I can conclude. I'm blanking on the name. Yeah, there's a name for a newer version. But would you find you would reach for that if you were in the Azure SQL world? Do you mind we'll just stay within Azure for the ETL piece as well? Yeah, absolutely. If you're in that Microsoft Windows framework, you have Azure SQL or just an on-premises SQL database. You can stick with those Microsoft products like SQL Server integrated services where you're building that flow of grabbing it from Acumatica and putting it into your SQL database. And then if you're on the Postgres side and more the Linux environment, that's probably where you would reach for Python to handle the ETL piece for you. Python programming language. Yeah, exactly. And since a lot of Linux and Postgres, you're working inside a terminal, you're running commands. This is where Python is really helpful. And Python is incredibly capable for data science or for business intelligence type work and building these ETLs. So, yeah, let's dive into the Python side first. Maybe we can back into the Microsoft side. So on the Python side, my personal experience, always known, has known that Python's a popular language. I've always known that it's popular, especially among quote-quote data scientists, people working with data. And my understanding of it is it has less to do with the language itself because when I've looked at Python before, it looks very similar to me to PHP. So when I run like theogforums.com site and the ERPCL.com, they both run on WordPress. WordPress was made in PHP. And so all tweak things in both of those sites with PHP codes. I'm more familiar with that. You look at Python, it looks very similar. My understanding is it's not the language that has made it popular. It's just that people kind of gravitated toward that particular language and they built libraries in Python. And it's the libraries that are just kind of like do stuff for you without you having to build it from scratch. Is that your understanding of why it's so powerful? Yeah, I know. PHP, I sort of think about as the website building world like HTML, CSS, PHP, all that. Python, it's more like application development as well. And that's not to say you can't have a web app. Odo is written in Python and Odo is a web app. And like you said, there's all sorts of different Python libraries, especially for data science like NumPy and all sorts of more libraries. And inside your Python script, you'll just say import NumPy and then you have those functions available to execute. And perhaps there's even a lot available for what we're doing, which is ETL, grabbing acumatica data and performing analytics on this data. So I have kind of knocked on the door of Python multiple times. I bought a couple of books over the years, multiple times, never got through them. So I kind of like knocked on the door, even opened the door, sort of poked my head in. But never fully walked into the Python room. It was just too heavy of a lift. Like I got into power BI, I got into the DAX language, I got into Power Query. But Python, like I couldn't get myself all the way in until earlier this year. So it wasn't until January that I finally sproaked down and started looking at the AI LLM stuff. Like I just sort of tried to ignore it for the last few years. But just in January, there were a couple things happen where I'm like, all right, I'll finally start to look at it. And this is just my experience. Then you're going to like 10X, you know, with your experience. But I'm just sort of trying to watch, trying to catch up to where you're at a little bit. What I found is a chat GBT, clawed, those are the two that I would probably reach for if you're writing Python. They're really good at writing code in general. And they're really, really good at writing Python specifically. And my understanding, just from interacting with chat GBT itself, just asking it, like, why are you really good at Python? Like the response is just because there's so much examples out there on the internet. And so it's been able to learn Python really, really well compared to other languages. So it's really good at writing Python code. And so I've actually experienced this just in the last five months, where I'll sit down and I don't even type on my keyboard anymore. I use an app called Whisper. I just dictate and I'll just describe what I, you know, want it to do with the data. Can you, can you, you know, move this? And it does a really good job of writing Python. And so I haven't, I actually have an idea for the ETL side of things, but I'm going to save that for a little bit later. Is that been your experience? Like, are you sitting down actually writing Python? Or are you just having something, some LLM right up for you? A little bit about so I went to school. I started in computer science and information science. And then I moved to business and finance. So I kind of occupy multiple spaces of technology and business together. So I did learn Python in school. I got Python certified. I know Python. I can write it. But also, as you said, AI writes Python. And even some of the top software engineers, they're saying, I don't even write code anymore. The top software engineers, they, they know code like no one else. And they're still using AI. But the useful thing about learning code, learning Python is that you know what your AI is doing. When it, when it gives you the Python code, you can see exactly what functions it's calling, what variables it's defining, what loops are running, how the recursions work. You don't have to write it, but you can read it very well. And especially, I don't know if you've had the chance to use Claude's new model release, Fable 5. It was their mythos model that they said, this is too dangerous to release to the world. But did it release? But did it release? I think just like five days ago or something. I don't know. I didn't realize it was real in the world yet. Okay. The dangerous mythos model, that's that's too dangerous to release to the world. They released it as Fable 5 and they softened the parts. Because it's because the name has such a bad rap. And they originally released Mithos to the big company so that they could protect their cybersecurity because of how powerful Mithos is. But now in the Fable 5 they just released, you can't do anything related to cybersecurity or biology. The moment you mentioned cybersecurity or biology to Fable 5, it says, sorry, we need to switch to Opus 4.8 now. We can't talk about that. I did not know that so they dunged it down a little bit. Interesting. So that's your preferred, I mean, this is changing on a weekly basis, but that'd be the one you would reach for when building something in ETL with Python code. Yeah. I would do is Fable 5, Clawed, even Chat GBT. It's interesting. I'm curious the future of how Chat GBT even competes with Fable 5. It's completely passing the intelligence benchmarks of like GBT 5.5, like Fable 5 is killing it. Interesting. So that's your preferred L11 of choice right now, Fable 5 in the end, Clawed World from Anthropic. So you'd sit down, you would, talking about how you would, would you start with the AI and then you would maybe modify some of the Python or talking about the process when you're going to build an ETL piece in Python. Yeah, so I have my own understanding of the architecture we want, the roadmap, how we're getting from point A to point B. And then along the way, you might ask it to do specific things for you. Build that Python script to do the ETL. Here's what exactly we want it to do. You can prompt engineer it to give you the exact script you need. And so do you feel like for the specific task of, hey, go look at acuomatic O data and build some way to do a differential refresh off of the last modified date time field and put it into this table over here in SQL for that task itself. Is that a pretty standard prompt? I mean, or do you do much tweaking to it? I wouldn't call it a standard prompt. There's definitely a lot of tweaking and it might even be different client by client because everyone needs different things. We're building different generic inquiry. Maybe we're trying to pull data about inventory, quantity and MRP or maybe we're trying to pull project accounting data. It really matters our intent. How the business is operating and what we're trying to get from it. Maybe we're trying to get this to run in Power BI. And I know the problem originally started that Microsoft changed pricing for I think it's data flow and then what's the solution for this change? Like if you're using data flow, what can you do instead? That's where we had the exchange. It was on that post on LinkedIn. Got you. Yeah. Remember that. Right. And so we have the problem. What's the solution to this problem? You don't actually need data flow. Data flow is useful, of course. But as I said, there's many ways to scan the CAD. We can still use Power BI. We could still use Excel without data flow. We just build that staging table, that data warehouse and build our ETL. So I want to dig into a little more when you say it's not a standard prompt. I'm still skeptical. So if I'm just, I'm super narrowing it. I know you can do a ton of stuff with Python. But if I'm super narrowing into, I have an O data source out of Achaumatico. I know that there's going to be different column names, generic inquiry by generic inquiry. There's going to be different data types for the columns. But prompt wise, can I have the AI just go look at the O data and know what those types are? And then even, I mean, I would even want it to be as simple as, hey, just go look at the O data URL, look at the data types and everything. Write the SQL to create the table in the database and design the Python script to do the differential refresh. Isn't it, it's not as standard as that in your experience? You bring up some good points that are part of it, but we can't just tell the LLM model, go look at my O data, go look at my generic inquiry, unless we provide it to the model. We can't just go into chat, you would see and say, go look at my O data, unless there's like a connector or an MCP of some kind, it doesn't have access to that. But if we, you can export generic inquiries to XML and that is readable by the AI. So when you tell the AI, here's the generic inquiry I built. This generic inquiry is taking our inventory quantity. And here is my O data URL. It's the the acumatica site slash O data. And then built, that's when you tell it your problem, build me what I'm looking for after you give it the context or else it doesn't know what it's coming from, what it is or how to do it. And that's where theoretically, couldn't it be a standard problem? I don't even need to describe the generic inquiry. I go to the URL in my web browser. It shows this really big ugly looking XML file that's the O data protocol. I just copy and paste that in and say, build me both the SQL create table statement and the Python script that's going to do the differential refresh. Couldn't it be a standard? Is that it? Yeah. And it's a great point. You can't standardize those things. And I actually hadn't thought of that until we're having the conversation because like I've multiple times tried to do it and just been like, this is just too hard and I gave up. But now, yeah, maybe it is that standard of 'em. Because I can tell you're very much in the like, hey, I can build whatever you want. And then it's just different styles. I'm very much just like, give me the button or just give me the standard way to do it, right? Yeah. Yeah. But to work on trying to make it standard, right? Yeah, be interesting. Right. And instead of just having a standardized prompt, we're entering into our LLM of choice, we can even have an agent because the agent is running multiple things. It's doing intelligent reactions. Once it knows one thing, it can use the LM to do know what the next step is. So maybe even agents is the way to standardize it. So I think it's a great time check because I could go for another two hours on this topic. But quick time check on your side. What's your time and click like? You guys able to go another 15, 20 minutes or what to look like? 10 minutes for me after the top of the hour. Okay. 10 minutes passed. Lisa, what about on your side? Same. Same? Okay. So I guess I won't dive too much into that then. I think I'll zoom back out then. Let's talk about the Microsoft side. So if you were then in the Azure SQL world and you were going to use SSIS or whatever the latest tool is there, would you do the same approach? You'd reach for cloud to start it and then make some tweaks after that. So yeah, that will be a different situation. And this will apply to many people, especially like in GovCon where you have, you're going to use Azure Government. You need to stay in the Microsoft framework. We can't go program experimental crazy stuff and Linux over here with Python because we need to stay strict where we're compliant. We use Azure Government and we're going to use a Microsoft product like SQL Server Integrated Services. So we're not going to tell the AI prompt, build me a Python script. We need our ETL to work within SSIS, which is more like a flow chart of here's the step to take the data. Here's the step of how we're transforming it and here's how we're loading that data into the new database. So it won't apply like a script as before. But we could certainly ask the AI, what steps do I apply inside of SQL Server Integrated Services? And knowing that this is changing every week, you know, the answer, I'm just curious, what would you reach for there? Would you still reach for the latest thing from from a cloud or now that you're in the Microsoft space, would you use Microsoft's co pilot, but what LLM would you reach for? Yeah, I would stick with co pilot if you're in a strict environment. Always follow your company's IT policy of what you're allowed to and not allowed to do. Turn off the configuration of use our data for training you. We don't want our data to train your models. And co pilot, it'll come with enterprise data protection. So that's what you would want to stick with in some of these more strict environments. Or even if you have a local LLM, like you don't want your data leaving the network, you can set up a local LLM. So there are multiple ways to skin it. What in general, like my theory understanding is that the LLM is better at writing Python, not as good as writing SSIS, what's your experience? Like, is it, do you find you have to battle it a little more when you're doing a CS or is it still pretty much a pleasant experience there? It might be a pleasant experience, especially with Fable 5. Fable 5 changes everything. Like a year ago, whenever I would ask chat, TBT, like, how should I do this generic inquiry? What joints, tables, conditions should I use? Chat GBT would have no idea. It doesn't know the DAX. Oh, you're just talking about building a generic inquiry. Forget about this part. Yeah, tell us about that. Yeah. Well, and then things have improved. Like chat GBT isn't what it was a year ago. AI industry is improving exponentially. Like not just a linear, it's better at a steady rate. It gets insanely smarter. Like even Elon Musk shows like charts about the singularity, but who knows if AI will get smarter than us or not. But anyways, the LOMs are much better at building these things now. You could build a generic inquiry in cloud-fable or much easier than you could a year ago. - When you do that, Lisa, you might have thought some of this as well. I know we're geeking out over here and kind of leaving out in the cold, feel free to chime in. What do you have it do? Do you have it produced the XML file or do you have it just like describe how you're supposed to go graphically build it? What do you guys do? Lisa, you want chime in first and then what? - I do it both ways. So when you're talking to it, you have to tell it build, because again, you're working in the GI form, like I said before, I didn't understand that at first. So you have to figure that out. So I tell it, this is what my form looks like, display your results and your suggestions and a table layout that mimics my generic inquiry form. So and it does great in that. - So you kind of write it up in text and then-- - Yes. - And then I manually input. But I can tell it to give me an XML, clog. - So why would you do one or the other? Why wouldn't you just have it do the other-- - It depends where and what I'm doing. - It's like if you don't want to break something versus building, like is it building from scratch versus making modifications to something like this? - Right, in my own site, I can use Clod in a client site. I need it to tell me manually how to do something. - I do, I do. Okay. - So in my own site, for me figuring out the generic inquiries and what the best, most efficient, I give it all the requirements. I'm using this for OData. I'm doing this, this is what I'm trying to do and it will modify. And I give it a few inquiries, a few XMLs and I say these are the XMLs that I like but they don't do this, this or this. And it gives me a new XML. Just watch out, don't tell it to give it a site map ID because that's problematic. - I have a XML general, so that's not a bad-- - Yeah, no. - No, no, no, no. - Yeah. So that's all I gotta watch out for. - I know, I'm behind you guys. I haven't done much with having it build GI's but that makes a lot of sense. Yeah, makes a lot of sense. - And for me-- - Clod is better like check out. - Clod's the one. - Yeah. - Yeah. - Okay. - Yeah. - Especially Fable 5 and I always have to give the model context first. Like how does it know what the XML of a generic inquiry even should be? It either has to learn from its own training data of whatever however they built it. It either has to scrape the web of how do you even build generic inquiry? How does that allow them to know what to do? But if you provide it and attach an example XML of what even a generic inquiry is, then it knows how to mimic that structure and then just change it. Put in the tables, you need the conditions, you need and the results grid. But it doesn't know what to do unless you give it context. - Awesome. Yeah, love it. Thanks for all that. I love just hearing, because it's changing so fast, it just, I love picking up just how it's working for people. So it's super helpful. So I think since we only have a few minutes left, I wanted to throw an idea at both you to see what your reaction is to it, what your thoughts on. So I was hoping that maybe you would have come up with some more magical easy way to do differential refatch. Sounds like it's still multiple ways to skin the cat. It's getting easier. Maybe we can come up with the standard prompt, but it's still like, it's still Cody a little bit, right? I have had a thought for an idea. So let me see if I can describe it. Ideally for me, when I'm specifically going from Accumatica to Power BI, and if I'm still looking for that, way to do differential refresh with now a third party tool, the simplest way to me is to use the good old fashion CSV file. So I've used, I've done a test on this with like six million records. Just random data and I tried loading the records into an Azure SQL database, and just representing the records in a CSV file and store the CSV file in OneDrive, which I get like tons of storage as part of my Office 365 account in OneDrive. Let's put it to work, right? I tried both approaches. I then pointed Power BI to refresh off of both of those and just timed it. Got out my, you know, actually no, the timer shows on the screen. So you hit it and it'll show you how long it took. The CSV file refreshed twice as fast as the Azure SQL database. And so I'm like, okay, let's not forget about good old CSV files. So now what I really need though is I just need a way to do a differential refresh on the CSV file. And the problem is a CSV file doesn't have a concept of a primary key. So if you pull, you don't have a way to say in a CSV file, absurd. You know, just find the record and update it or insert it at the bottom. It doesn't have that concept. But through having multiple interactions with chat to BT on this topic, together we came up with the idea of why not load the CSV file into memory temporarily. I think it was recommending pandas, I think. I think of it as like a temporary database. Add the primary key in that memory piece. Then do the O data differential refresh into that pandas object or whatever it is. And right back the result over the top of the existing CSV. Do it all on Python code. Then I discovered you can run it on Google functions. I think it is. I looked into Azure and you have to like open Visual Studio to publish a serverless Python script into Azure. And Visual Studio has always been like this thing that I don't want to open. So I went with the Google instead. You get just a simple text window and you paste the Python script that you want and you can just say run every night. In theory, I think I have like a serverless CSV file way to do a differential refresh. Where now the differential refresh data lives in a CSV file on one drive instead of a SQL database thoughts on that crazy idea if I should pursue it or not. - That sounds like a lot of fun. And you've almost made pandas be your staging table in a way like instead of having SQL or Postgres or something that staging area is still there. There's just in pandas and you're putting in your flow that way that's kind of funny. - And it only lives during the refresh then it disappears. The CSV file is the only thing that holds data permanently. And then you just need to throw a big chunk of resources at it, which I'm fine with because the things only gonna run in the Google Cloud for five, 10 minutes maybe, right? So I'm okay with giving it a good amount of resources and being more expensive for that few minute run. But I don't know, have a test set scale. I'm not gonna throw it at you if you're gonna think about. Because then you don't have the database site. Cause for me the hard part is matching up, this is a date column, this is a text column, this is a number column, making sure that matches your database. In the CSV world, you don't care about it. When you bring it into Power BI, you deal with the data typing and Power Query. I never have to deal with the data typing in the middle layer. That's what excites me about it. But yeah, that's a very cool end. It's a very fun garage build, so to speak. But I would say in formal businesses, they're much more reluctant for garage builds. When you say like, oh, I have this quirky way we can get around this and do it. They're like, is there a better way? And that's when I try to stick with formal process flows, sticking within the Windows framework because it's not always about capability and what we can do in a garage build, but just comfortability, formal structure because this is a real business. Especially in the GovCon world where you need the compliant architecture, yeah. Yeah. So the GovCon world will be very reluctant when if I try to tell them. So I have this pandas Python script and then it throws a CSV into my Google. They they they I mean, it works. It's cool. It's functional. It's a proof of concept, but it's still a garage build. Makes sense. Makes sense. Yeah, I might put my shoulder into a little bit. The whole goal would be I guess to remove the quirk. To remove the quirkiness from it if I could pull that off. Yeah, that's awesome. I know we're up against it. Any any further stuff you guys want to throw out there before we wrap it up? Well, I appreciate you having me on here. I wish you the best with the ERP seal thing. That's cool. I like to see how it's coming on. And I like how we Duke out some of these concepts in our threads. It's a lot of fun and I appreciate you having me here, Tim. Well, I'm glad you have fun with it too. I do too. Sometimes I'm not sure like are we getting to heated here? But no, I have fun as well. Like sometimes you're just like, oh, different approaches and just leave it at that. But it's still fun to learn from other people. I see the acumenic community is like an iron sharpened iron. Some people know one thing. Maybe someone has a different approach. And that's how we really arrive at the best answers. And that's what's so great about AugForums ERP seal or your Discord server. And I recommend that for anyone. And Lisa, sorry, we didn't get you more involved in this conversation, but thank you for joining us. - That's okay. Thank you very much. I appreciate it. - And let's keep the Hawaiian shirt culture going on the AugForums Discord server. - I will be there with my Hawaiian shirt. (laughing) - That was awesome. All right, thanks again. Thanks everyone. - Bye bye. (upbeat music) [Music]