Name: “Bridging the Gap” Between Legacy ERPs and the new generation of AI-First ERPs
Uploaded: 2025-11-11T16:16:06.151Z
Duration: 1 h 6 min 32 s
Description: “Bridging the Gap” Between Legacy ERPs and the new generation of AI-First ERPs

Transcript for "“Bridging the Gap” Between Legacy ERPs and the new generation of AI-First ERPs": can hear me. Right? Yes. Perfectly. Perfectly. Okay. And we're live, and I can see we've already got some people with us. We'll just wait a couple more, minutes for the rest to join. Welcome to those who have joined already. Great to have you here, and looking forward to this discussion. Do not hesitate, of course, like on any good webinar, to, let us know where you're dialing in from. Keen to see how, global this audience is. Also, if any questions, that you're hoping to get answered from this webinar, put them in now. We have a dedicated q and a section on the right hand side of your screen. Stick them in there. We, of course, want to make, at least ten minutes, and we'll run over, of course, if we if we need to, four questions at the end. So please do pop them in there, as you're there. For those who've joined, still just saying put those questions in the q and a section now, and, of course, as we go, and we will do our best, Dudley and I, to answer those questions at the end as we go. Okay. We'll leave a bit more time for people to keep joining. Yeah. Maybe if I could encourage everyone to to ask a question because sometimes you might leave a webinar and you haven't haven't quite asked the one that's in your in your head and, you carry that with you. It's, you got an opportunity now where where we can pretty much answer most of the questions. And if we can't, we'll get hold of you afterwards and and have a, you know, have have an answer for you if there's something that we can't answer. Okay. We'll just wait maybe thirty seconds more, and then we can get going. Should I turn. my music on that? Yeah. I can't hear you, mate. It's all good, darling. Welcome, Brian, from Johannesburg. B Squared. Yasmin from London. Welcome. Okay. I think, we've got a good amount of people here now, so let's kick off. Welcome, everybody. Today, we're gonna be talking about, the impact of AI on ERP systems. This is a follow on conversation from a conversation that Dudley and I had, I think last month now, at ATCAP days or in person event. For those of you that don't know ATCAP, I will, give you a little, overview of what we do as the host of this webinar. But, obviously, I wanna focus the bulk of the time on the conversation. So first of all, big welcome to Dudley. For those of you that don't know me, I am the head of partnerships at AGCap UK, and Dudley is the CEO of Financeflow dot ai, a company specialized in implementing ERPs, and also, AI solutions to support those ERPs. Terms of AGICAP, for those of you on this call who don't know us, maybe we have some customers, maybe we don't. We are the number one, middle market treasury solution for, the mid market globally. We've got about 8,000 customers in 2025. We've raised, nearly 200,000,000 in VC funding today, and we have just launched our latest office in Austin, Texas, going global. As you can see, we are the market leader, both in cash flow management and also mid market treasury. And what our platform does is to provide a one stop shop for cash management. So we will integrate to ERPs, which is why, I come across this topic so frequently. I'm so interested in it. We will integrate to companies' ERPs, pull all of their purchase orders, sales orders, supply invoices, customer invoices, credit notes, etcetera to create an automated forward view, which we will then use our software to convert into cash timelines rather than invoice due dates. We will then connect directly to our customers' banks to give them a live view of their current cash position, and then we'll complement both this direct feed from the ERP and the direct feed from the banks with some longer term data, perhaps from an FP and A tool, perhaps from a CRM tool, or just a standard budget that we might feed directly from SharePoint or Google Drive. This way, we aggregate kind of three sets of data. So you're kind of you're you're cash position today, and then we aggregate, you know, what's gonna happen tomorrow via that data in your ERP with, okay, what's our long term plan, all in one view. And this gives our customers kind of the core of our platform, which is that top right automated cash flow forecast. Top left, okay, what's our cash position today, and what's it gonna look like next week? And what's the risk of that? Can we move money between different bank accounts to take advantage of short term excess cash or derisk going into our our overdraft? And then with all of that centralized information, we provide, these two bottom modules, which is the ability to directly execute payments from the edge cap tool. So run an end to end supply invoice management, verification workflow and then a payment approval and payment execution workflow. And then also for mainly for our b to b customers to identify their own late paying customers and come up with automated workflows to encourage them to pay more promptly and, of course, use that customer data to inform their cash flow forecast around when particular customers actually pay their invoices. So in this way, we create, a one stop shop for cash where the top half is about visibility of information both today and tomorrow and months or years into the future, and the bottom half is then about taking action of that information. So, to try and kind of piece that together, I always think this can be a helpful, slide, to understand kind of what functions or what roles within finance teams we play. So here, I've kind of split out into, four columns the sort of classic aspects of the finance function, and ATCAP is really playing a role in these six, I think. So cash management, cash flow forecasting, after all, we're a treasury tool, also do a large component of the reconciliations. And then, you know, more operational tasks like credit control and supply payments, those two bottom bits are doing that. And then you can also do much of your budget management and control in AGICAP as well as produce many of the reports, that you need for monthly and year end reporting. But, as I said at the beginning, we're obviously not here to talk about AGCAP. We're here to talk about the impact of AI on ERP. So so that reason, I'm very pleased to welcome Dudley, to have this conversation with me today. Dudley is the veteran of an absolutely staggering 650 ERP implementations over twenty five years, including some of The UK's biggest names like BAE Systems, Jaguar Land Rover. He's also implemented the full range of ERP systems, so all of Sage, SAP, Oracle, and Microsoft. So full coverage of different systems, but also full coverage of enterprise and mid market and SMB. And he has most recently published a book on, how AI unlocks the full potential of of ERP. So we're speaking to a man very experienced in implementing ERPs and also with the the rare knowledge of what AI, will actually do, to ERP systems. And as I said at the beginning, this is kind of really, this came with the back of an initial conversation Dudley and I had, at HDCAD days, a couple of weeks back where we talked about this kind of this AI train that feels like it's coming to ERP. I don't know if it's the same for everyone else on this call, but every other post I see on on LinkedIn is about an AI native system. New AI native ERP raises huge funding rounds. And I think what we talked about in that session was how it feels like you have to kind of invest in these AI systems, you know, because they do have momentum, or at least that's how LinkedIn makes you feel. So you have this kind of idea for known momentum about AI native systems. But at the same time, it's very hard still to decide what the difference is between an AI native ERP and a legacy ERP or a standard ERP like like NetSuite or one that's been around a big brand for twenty years. And so what we aim to do in this talk is really to uncover those mysteries and help you guys, as finance teams, make the right decisions going forward for your business. So we talked about this kind of, the options that you have. You can either hop on this new train of AI or stay on the platform and stick to what you know. So from this webinar, we want you to walk away with, and if we fail to do this, please do tell us at the end or ask the questions. As we said, put them in the in the q and a, a clear understanding of what AI native means, first of all, how that's gonna impact ERP systems, and then also be able to kind of conceptualize practical examples. I always think it's it's very easy to or not very easy, but easier to understand the theory about what AI is gonna do to different systems, but actually conceptualizing the impact can be harder. And so we hope to demonstrate that for you guys today as well. And then also a means of comparing legacy ERPs with AI native ones when it comes to investing in an ERP for your business in the near future or or or in the coming years. So, Dudley, let's kick it off with the question that I think is probably on everyone's mind and certainly took me a long while to understand, before talking with you. What does AI native actually mean? That's a great question. So it is on everyone's minds. So let's let let's do a comparison between what we call legacy and then AI enabled and AI native. So the I I would say there are three three categories, but it's in essence, ERP, which is enterprise resource planning. Most mid market companies and top end companies would be seeing enterprise resource planning as their system of record. So it'll be the the place where their transactions are kept, sort of debits and credits, master data, things like names, addresses, telephone numbers, etcetera, of their of their customers, and then the posting rules. So in other words, when you're generating an invoice, what gets debited, what gets credited, and what kind of ledgers you you're keeping. So it's really the system of record. So when we're looking at at any ERP, whether it's AI enabled or not, that's essentially what it does. But AI enabled or even AI native means that that you've you've added in a chatbot. Most people know what chatbots are, and and also know what a copilot is. So you you've seen this, I'm sure, advertised and spoken about a lot. And in essence, what that is is added a chatbot or a copilot on top of or inside that system. So a chatbot essentially is is helpful, but when it's done and implemented on what we call old plumbing, so the infrastructure behind the the chatbot is is is really old, and it's it's, not really built for an AI enabled in a world where AI native is is designed so that an AI assistant or an AI chatbot or copilot can do its real work and but do it safely. It can do it in a in a relational ledger, so for integrity, so the debits and credit, etcetera. But now you can add a semantic layer, a layer that helps you understand things like documents and messages and emails and conversations, etcetera. And I think the best way, Nat, to to look at this orchestrator or or chatbot or Copilot is is to think of it as an as an orchestrator that puts out calls into the dataset and returns, information and returns calculations. So what we really have is if you're AI native or AI enabled, you can describe the outcome that you're looking for. In the second place, what happens is the system then would go and assemble that information. And then in the third step, it would then give you the the review of that, information. Instead of humanity manually building that on on Excel spreadsheets, you'll be able to get a visual or or a or an actual image of that, or tables or or lines of that information as it's assembled it. Okay. So clear. So if we if we think the difference between AI native versus adding just a chatbot or a Copilot, as you said, is the the AI native is like a a a Copilot or an LLM, designed to work within the datasets, that is fed rather than just added and the difference between the kind of data that you'd be able to extract in an AI native system is, the level of semantic data that you can pick up, so stuff like, you know, from your emails or from your notes. I think some of the examples we talked about before is, you know, if a particular competitor kept cropping up in in certain notes of your of your sales reps or, or or emails, this is the kind of information that, an AI native ERP could quickly extract and pull together into something, whereas it would be very, very hard to pull that together in a traditional ERP. Yes. And and maybe if I can expand on that. There's there's really, two main areas that, that any system works on, and that's on the front end, which is where where you would be working. So you'd be capturing data, you'd be processing, and you'd be running reports, for instance. So that's the front end or the software side, the logic of of a system. But what many people don't realize is that there's a a a really, really important component, which is the underlying database. The underlying database essentially holds all the data, and AI native, really does, updates both of those. So if we look at updating the software and the type of logic you're building, so if it's AI native, it's got a whole lot more AI ability or capability. But the back end database now is able to use both structured or relational data as well as unrelation unrelational data. In other words, unstructured data in the back end. So AI native really updates both sides of those. And your logic layer, it gains sort of an AI orchestrator, level, and the data layer combines tables with the semantic index index. So what you really said there is about emails and and and messages. Those are all unstructured content. So what this means to the user, really, is that that users can do less work and get the same or even better outputs and often a lot quicker too. Okay. And, really, then the difference between, an AI native e r ERP and an AI, Copilot based ERP would be that with in the second example, you're just adding an AI to the front end, whereas in the AI native, you've got it both on the the software layer on top and in the database. Yeah. So. the whole the whole infrastructure is different. Okay. Clear. One of the other things you've often talked about is kind of moving away from doing the manual work when it comes to, constructing databases. You can imagine, you know, filling in a table in a CRM or in an ERP in order to be able to then build out some analysis. With AI's ability to bring together unstructured as well as structured data, you move towards just being able to do review and, as you said, that produce even better outputs. But who's doing all that analysis for us? Is it is there a model in there as well? Well, that is that that is a that is a good question, and some people talk about hallucinations and and numbers being made up, and and you're not sure whether it's it's real or not. So if you if you reframe your thinking in terms of what AI is, it's really an assistant or an orchestrator of information. It's not really a processor of information. So, it it delivers, the the the bridge between what we call deterministic services, like math, like doing its simple calculation, x minus y equals zed, things like tax or forex or allocations of of ledger transactions, for instance, or even things like MRP, which is materials requirements planning, things like managing your your stock, your inventory, and and your stock quantities on hand, and those that have been ordered, and those goods in transit, etcetera. So those are all deterministic elements to it. The assistant is able to to draft evidence and help you, not only view, but you can approve data, whether you, you know, whether when you once you see it, you can understand where it comes from, but you can also then approve it. And after reviewing it, and then only can you make use of it, once you really are 100% comfortable that that information is indeed correct. And and that that's really the difference, I would say, between just generating an Excel report. We know that with Excel, we would generate we would create, formula, and these formula will will will be run on macros, for instance, but they also go wrong. So sometimes you might have a beautiful spreadsheet, but your formulas have all gone wrong. So what AI allows us to do is to is to negate a lot of those errors. Okay. Super clear. Thank you, Dudley. And then I just wanna come back quickly to this idea of layering on top AI versus, being AI native and what the difference is. Because I think the question that you hear quite often is, can't an existing ERP just replicate the same output, of an AI native ERP by simply adding on more and more AI? So, we talked about a very basic example at the beginning, but there's many specialized AI tools that they could keep complementing their sys system with. Is it not just possible for a legacy ERP to continue to do that and produce the same outputs? I wouldn't say the same. I I would say definitely, it will be improved outputs in in the sense that that you'll be able to do a call or or or or try and retrieve data using AI functionality. It'll help you, retrieve the data potentially better. But think of it like this. You got three stages that a lot of companies will be entering. Number one is they hold on to their legacy system and don't use AI. The second level up would be having a what a copilot, injected into a legacy plumbing, if, if you'd like to use that word plumbing, but the engine. And and it's to run it like a a a sidecar. So what you wanna try and do in if you're in the middle, is you have the old legacy software and you have a sidecar or a vector sidecar, which in other words, it is a it's a separate database that is an AI, enabled database on that sits next to your core ERP database, which is the relational the old plumbing, if you like. And that's a lot like taking a jet engine and putting that on a bicycle. Look, the bicycle is gonna go very, very fast. Whether you can control it or not, that's another story. But AI native is built where, it from the ground up, you have things like policy aware retrieval. So it it's incredibly smart in its ability to be aware of the policies that that you've built into into your system. The tool calling can be can be, built with what we call guardrails, so you can make sure that that the output is exactly the way you're expecting it. And what we call end to end provenance. Now that's a that's a big word. So I'm just going to say provenance really is the historical, record of of origin. So provenance, essentially, from an accountant's point of view, is critical. We want to know where did the data come from. And if you can't, with your AI tool, be able to really drill down and and see the the true origin of that data, where it comes from. All your AI is doing, it's doing something over the top, and there's a hidden middle somewhere. And if you can't, as an auditor or an accountant or a financial controller, be able to drill down and and and see how the calculation was created, then you you you'll be, you'll be wary of the results. So bolt ons do help. This is the sidecar layer that that you can do. But remembering that the architecture is is really the underlying, change that makes, work get done and get done properly, and the review is still stays in the hands of of of the, of the user. Okay. Super clear. Thank you, Dudley. And I know now we're gonna come to, one of the other main areas where, certainly, AI native systems talk about their ability to have impact versus, traditional ERPs, which is implementation. But I'd just like us to tackle quickly one of the questions that's come up in the chat, from Brian, out in Joburg. How do you tell the difference, Brian says, between an AI native ERP system and one where the AI has been added to a legacy ERP system? That's, that is that's that that's a really good question. There there's there's normally, four tests. I just wanna go and and and pull up one of my bullet points there. So there are normally four tests that we look at. The number one test is, is it in a two dimensional universe? So if you think of a database, it's a lot like a spreadsheet. A spreadsheet's got rows and columns. And a legacy database has got loads of spreadsheets in the in the database behind. And these data these spreadsheets are connected to each other by relational ID tags. So, really, what you have is just a whole stack of of spreadsheets in two dimensions. So the first thing we look at is the is data. So if you got the legacy system, you're gonna have a two dimensional dataset. If you've got an AI, system, you've got over 3,000 or more data points. So instead of two dimensions, you've got over 3,000 dimensions. That's the that's the database. So that's point number one is to look at what the origin, the actual data, the database looks like. So that's check one. Check two is the security elements. So security is, it's it's very it it's it's impossible to build security, proper security, if you're putting a a call directly into an old, legacy database because it's not built for those additional security layers you need if you are, or sort of putting your fingers into something like poking yourself in in in the eye kind of thing. There's no protection there. So the security layer is not built in. The third thing is tool calling. So and what an AI does incredibly well is calls, tools from different places to do a particular job. When you when you do have the tool calling, it's that's that is what true AI and native or or AI enabled does. It allows for the correct tools being called for a particular job. And then the fourth thing is which comes back to true accounting principles, and that's the provenance. That's making sure that the the the the historical record of origin can still be visible even if you've generated AI output. You need to be able to click a button somewhere or be able to look at the the the the string of text as to how it did its calculations and how it got to the output. So, Brian, I hope that answers your your question. Okay. Let's keep moving as I know we've got a lot more questions, and the implementation one is a recurring theme in all these conv conversations. Can AI really shorten implementation time lines by 60, even 80%? Well, that's, yeah, that's the million million dollar question. So what we still will have for many, many, many years to come is the human condition. And if you look at any ERP implementation, if you've lived through one, and and some people call it suffering through one, is that you still have, the whole discovery process where we're trying to understand the IP, the the acumen, the business intelligence of the entire business. So that's from the old system, the way they do business, the workflows, and so on. So you still gotta do that discovery process. However, now you could potentially record those meetings and then transcribe them, and that can become documentation. So maybe there's a bit of maneuverability there to to shorten the discovery process so that AI can help us do that. Then there's the mapping of the business requirements against the functional requirements. So, again, there could be, again, a bit of a bit of maneuverability, so we can be more measurable measured, sorry, in our in in our ability to deliver. And then testing the system. We can't really let AI test the system on behalf of users because at the end of the day, the users are going to need to see the output, and they need to test whether that output is indeed the correct output. So in in terms of, auto generating, configurations for for a new system, it may be, shorter, maybe getting to 80%. However, the real time, is still comes down to project management, change management, the quality of data that already is in the organization, and also the scope. Because often when you're moving from one system to another, you don't just wanna do a sideways motion. In other words, just exactly the same replica in a new system. You want to go improve. So you've got the scope that you want to now use the ERP implementation as a launchpad to grow and and and lift the ceiling of growth of your business. So I I don't think those things will change. So, yeah, hopefully, that answers the question. Okay. So if I understand what you're saying, it means that, you know, the change management's, aspect of an ERP implementation is not gonna go away. People still need to be trained around the user tool. The project management element of it isn't gonna go away. And I guess also the upfront, you know, what you've called scope there, the planning of, the system and what you need the system to do and what outputs you need to produce is still gonna require a lot of that upfront thinking. So just in terms of pure processes, it's not gonna change that much. But maybe the mapping element or the data cleaning element, these are things that can be improved, and sped up. They can, but there's one thing that that, anybody who's who's lived through an ERP implementation would know that one of the core problems we normally have is the quality of the data that's that's held in existing systems. So you have systems that have been around for ten, fifteen, twenty, thirty years in an organization, and many of those systems would have this legacy data, duplicates. It would have poorly captured and and maintained datasets. And to clean that lot up is an immensely big job because it it needs the oversight of a human. It can't just be done by AI. Of course. Yeah. I think, like in chess, human plus AI always beats AI alone. That's what they say. So, moving on to our next question, we have some real practical examples. I think you've talked already, about the difference between an AI native architecture where you've got so many different dimensions that the system can call upon versus a traditional infrastructure where essentially you have a series of connected, two dimensional databases, or two dimensional databases that you can connect. Let's try and give some practical examples to that. Really, what is the advantage of having so much extra data? Can you give us an example of how, actually, in practice, that makes it that much better a process? Yeah. I think one of the the biggest, bug bears would be, if we look at the AP or accounts payable process. If you're if you're a financial controller or an accountant, you'll know that, you know, AP is a is a module that can have loads of headaches. So for instance, you have a price variance which lives in different places. So the you know, straightforward could be simple in your ERP. It could be a purchase order, and that's quite easy. You convert the purchase order into a, into a supplier invoice and then make the bank payments, etcetera. So all of those are transactional, two dimensional, transactions. However, what what what the ERP doesn't, see are things like delivery notes, receipts, other things that, you know, when when you've picked, packed, and then dispatched an item, was were there delays? Were there breakages? Were there returns? And emails and phone calls and all that. All those things that are unstructured data. So accounts payable can be quite complicated because of all the moving parts. The ERP can show a sliver of data in the real world. But if we take delivery notes and all the other things like email and and other communication that happens in the delivery process, we can take AI native systems or AI enabled systems to read all of that data together and then pick up the the PO, the receipt, and explain the variances and and really help the user understand where things may have gone wrong and and help you draft things like postings and post these entries correctly. So if it's a partially returned order or if it's a PO that wasn't fully delivered, etcetera, you can get the AI to do a lot of those those things for you because it helps check, even through machine learning, the variances that that that can occur. Clear. I think that's a really good example of, something that I see come up in a lot of our my own conversations when I'm talking to payments, or prospects about using Adrigap. This is one of the main, challenges is understanding, you know, why an invoice and a payment don't match. So I mean, and a lot of the time it involves, you know, them having a chat with, someone else in the business to dig up the reason or, explain why they can't make a payment yet. And if the AI could sort of sift through all of that data for them, so much time would be saved for that person. And imagine you could just read the explanation, instead of having to have an interaction or depend on someone to get back to you. It would be, without wanting to sound too enthusiastic about AP, it would be game changing. Let's move on now to, the more change management side of things because, I guess, when you think about, the infrastructure, it can feel like, it's such a big change. But does an AI net native system actually mean different outputs for the end user, or is it just a different way of working? So so, we still need to be able to, as humans, digest the data that that that the AI delivers. So seeing a a chart, a pie chart, for instance, or a table, or or calculations that have been created, or lines in in in a particular document format. So you're still going to get tables, charts, and and even a bit of narrative around an explanation of a particular, transaction. But the change is really how these, items are are assembled. So AI can call on tools to deliver, the output. So for for instance, you if you're looking for a particular outcome, you're looking for, what am I say what's my sales growth or what which which elements of my business are costing me more money than others or which has shown a higher increase this last month than the previous three months, etcetera. What the system does, it it retrieves evidence, it uses calculators, and then it proposes next steps, and it will add in sources of where it got that information from. So you are constantly verifying and double checking and and being sure of the data that's delivered is true data. And that comes back to to that provenance concept that that I was talking about. So making sure that the historical record of origin is always the the true, source of truth in in in a system. Yeah. So it's the the outputs really won't change because, obviously, we still need, as humans, to digest data in the same way, or we will carry on digesting data in the same way. Just just how you get there will be much more communicating to the AI to, summon the data that you need and convert it into the right format. And then your job, rather than doing all that building, will just simply be verifying what the AI has done is correct. And then talk to us quickly about, to Steven's point, I think, the agentic AI layer. How how does it verify, the data? Talk us through that process. So so it's it's it's simple, but but but not easy. So so the the the AI really is grounded at the bottom end. It's grounded at, it sees the data that you're allowed to see. So if you think about, an output, and you're busy you're you've got a particular, user access permission, for instance, and you're allowed to see particular parts of data or process particular transactions. It can only see the data you're allowed to see, so it it can still be ring fenced in terms of what you can do as a user. What AI also does, it sites the sources where the information comes from. And the newer systems that I've seen, and just last week, I saw it, where you you deliver the output, and there's a button you can you can press. And it will give you absolutely every calculation and every source that it got the information from to so that you can double verify on your own to be sure that the output was indeed correct. And it uses things like, schema checks and policy gates. So each cell, if you like, or each coordinate of packet of data has got a security layer built around it. So you got policies and you got schema checks and so on. So you can be quite confident in a proper AI native system or an AI enabled system that the confidence can be very, very high. And if your if if if if your confidence is not there, you can double check the the source. So, you can get an exact explanation of what it did and why. Okay. Yeah. It's like what we see at EdgeCap's in EdgeCap's AI assistant. You can easily find the source data very quickly in an LMM that's so specialized on a on a particular dataset in a particular environment. Okay. I think now is a natural follow on for that other question that most people, are worried about when it comes to AI, which is controlling the LLM and, you know, stopping it, hallucinating it, hallucinating. And, of course, how is all the data secure as well? Can you just walk us through how you keep an ERP's LLM so controlled, how you stop it hallucinating, and how all that data that it has to mine through, you can be sure that it remains secure? Yeah. It it really, it's a it's a really it's a it's a fascinating question, and it does come up so many times, and it comes back to provenance. So it comes back to that reducing hallucinations to becoming a nonissue. What the way that we do it is we ground all the answers in in the data that exists in the, in the dataset. We're not going outside of the system to go and call data that's in the in the, AI universe, if you like, but it's only very specifically focused on the on the data inside your system, using specific tool calls for the math. So we will only allocate we will tell the AI to only use particular tools to do the math calculation calculations, running the validations. And then there's then there's additional layers of of of security and and guide guide rails, which we add in, which are things like segregation of duty. Segregation of duty checks that can be done on the on the data core. And this might get be getting very, very technical, but I just wanna go into just touch on this so we can always have a deeper conversation later. But the segregation of duties already happens within accounts departments and even within teams in the wider organization. AI then helps us put in those SODs, the segregation of duties across even a bigger set of data now, and then requiring the human approval for any material actions. So there's not only retrieving data, but there's actually writing of data to the database. And if you need to go through an approval process, we can have a workflow that then gets, once an action is taken, that there could be an approve a human approving particular actions. So security flows through, and and this is really something that's happened for many, many, many years already in in the legacy systems, is what we call role based access control. So you'll know a user logs in, can only see what they need to see, can only do the actions they need to do, so it's role based access control. And the other one is attribute based access control. So only certain attributes can be available or visible to different users. So attributes are really just fields and functions in in in the system. So these two models for managing user permissions have been used for years and years by ERP companies. So it's very standard practice. Row and column policies, So that's the two dimensional rows and columns. And if you then go into an AI enabled system, your your, your multi dimensions, packets of data, have got policies of retrieval, encryption, etcetera. And then there's also what we call masking, which is personally identifiable information. So anytime you do a call or a or a post, you're going to have personally identifiable tag being added as an audit trail to all of those transactions. So the data never leaves the the boundary within without consent, and it stays within that sort of let's call it a vault of of data, all being tagged and tracked, based on an audit trail. So So if you if you think about, where the current vendors in the marketplace, your Oracles, your Sage, your SAPs, etcetera, and your Microsofts, they have all been doing this for many, many years. And for context, many of them have been around for thirty, forty, fifty years already, and this is just standard practice. Okay. I'm gonna confess, much of that was way too technical for me. But what I took away was that there is definite standard practice that you can impose to keep your data secure and keep your LLM or which is, I guess, at the heart of any AI native system, focused and stop it from hallucinating. Okay. So I think we've talked if I just kind of sum up what we've talked about, we've talked about what an AI native infrastructure, how it's different from a legacy infrastructure, what, you know, the real, implications are in terms of, how that works in practice and then the difference it can make. You know, we talked about that AP example, how suddenly you can go from, having a smooth workflow, but then there still has many blockers around, okay, why do these, payments not match, why can't we pay this invoice, Where an AI based workflow could really give you all the reasoning in one go without you having to to speak to someone else. We've talked about the difference AI can or cannot make to implementation. But if we try really to sum up what the real transformation of AI native ERP is, what would you say, Dudley? So so what yeah. I mean, that that's it. Again, to try and simplify and and and and the the use of technical jargon is is probably is not useful in this instance. I'm gonna be very, very practical. So I'm gonna say that any ERP system should be a closed vault and a large language model all built to only view and extract data from a closed system. But the closed system now, instead of just having two dimensional data points, you now have with an AI enabled system or a native system, you have thousands of data points to to to call on as opposed to just a few tables within an ERP system. So really, at at its core, the real the real transformation is going from a small set of data doing manual, input, and sometimes you can do a little bit of automation. But now with the AI enabled, you can you can retrieve and even input data using voice, using, text, and and and using all kinds of natural language. So if we just conceptualize a little bit, so we say the ERP stays your system of record, your your your area of truth, your your your, your true points of of origin, and then you would use specialized tools like AduCap for cash, for instance. You could plug that in directly into the system, and you can feed, all the correct, systems, all the systems you want, but with the correct data. So the ERP doesn't really change its its role. It just now becomes a lot more powerful. And your transactional records, your documents, your objects, and storage, and all that just stay in the same place, but now you got that semantic lookup where you can look at, unstructured data. Okay. So if we if if we try and sum that up, then the key changes is the architecture, so the number of of data points that you can tackle, the ability to pull that unstructured data as well as structured data, which all ERPs can pull structured data if it's organized and and accessible in these in these tables, and then the way you produce the outputs, based on the natural language processing with which you communicate with an AI native ERP. Okay. Then I think, just conscious of time, Dudley, if we move to kind of the critical question, would someone be mad to invest in a legacy ERP now? Does it still make sense? How can our users on this call, how can they go about comparing the two? How do they choose between whether to go AI native that might not have the functionality there yet or might not be rolled out, versus, tried and tested brands that have been there? Yeah. That's that's that is an another great question. So so I I would, I would not life carries on. So I wouldn't stop. If you're in the procurement process right now, I wouldn't stop. I would, however, be a lot more aware of what to ask. So there are some key areas, of questioning that you can be doing. But in essence, it's now a lot more about planning. And, in the old in the older days, pre AI, we we did exactly the same thing. You go through a discovery process, you understand your business requirements, and then you go out and search for an ERP that would match those requirements. So none of that really changes. So, just to go through the the the steps, so you're going to say, I'm gonna map my requirements, my, my accounting requirements, but now I can actually add unstructured requirements as well. Things like email, documents, SharePoint, other things like that. And then the second step I would do is speak to experts, to actually score the vendors based on those four tells that I was speaking about earlier. So what does the database look like when you are talking? Don't just ask about the features and functions, how you process the transaction, but ask what does the underlying database look like? What does the security look like in the underlying database and also in in the front end? The third is which tools are being called by the AI, if it's got AI, and then also the provenance. Can you go back to look at the point of origin of the data? And if you can see that that, all those four being visible, then I think you're absolutely fine. As long as you're asking those questions, I would, I ever say, that that a good a good thing to do would be to plan before you start engaging with any suppliers. And there are some suppliers that are willing to work with you even over eighteen or twenty four months before you make a decision on a particular ERP and rather plan or what they say measure twice, cut once, like the carpenter's rule. So rather double check, on what your requirements are and make sure everyone's on board before you start looking for an ERP. Same rules as ten to fifteen, twenty years ago. Nothing's changed. But now you just be more aware of the additional tools available out there. Okay. Super clear. And then biggest change, you think, for teams switching to an AI native ERP? So if they switch, it really is is, think of the AI as an assistant, a copilot, an orchestrator of of your data. So instead of doing all the grunt work and and all those sort of trying to churn out management packs or some kind of report, think of it as as being able to generate high quality drafts that are maybe 80 or 90% correct, or even a 100% correct, but that you can then rather review. And the the bridge you're getting really is moving away from that really hard heavy lifting for days to generate a report where you could do this now in a few minutes or half an hour or an hour to get exactly what you want and even looking quite quite pretty as well if you if you know what you're doing. Okay. So I know that more or less brings us to the end here. And I think if I sum up what I've heard, you know, it's clear the change is coming. There's a big change in in architecture. If you use these AI native systems well, you can have a big change in outputs, which can massively increase the productivity of what you're doing. And, also, there's a huge change management element because the tasks that people end up doing are gonna be very different. Dudley, if if you were to give one piece of advice to all of the the CFOs out there or the finance teams being confronted with this decision right now to sort of stick or twist, to hop on or hop off, about which ERP they should buy, what advice would you give them? I would say look internally and and and keep your ledger and your transactional data solid. Think about how you can add additional semantic or unstructured data or semantic brain, and then the guardrails that go with that. And and think about how you could help people move faster. And and it's really just that instead of doing the really tough grind work, is to be able to generate, and review documents or outputs without spending time to get those outputs. I would say that that would be the the the future that we're looking to to to get to. Okay. Thank you, Dudley. Super good, and clear advice. So, I did I think we did have, one question in the q and a. Please do keep adding them. We'll tackle Stevens now. I'll just comment quickly and leave it on the screen. For those of you who are interested in learning more about AduCap at the beginning, reminder, we are the host of this webinar, and we are g two's leading cash management and cash flow forecasting software. And you can book your own personalized demo of Adjutap with me here by this QR code. Steven, let us come to your question here on AI agents. Dudley, Steven asks, how advanced is the AI native foundations, and can these typically be created and built around agentic processes and flows? And are these reliable? So less about the extract instead and move to actionable outcomes from data, structured and unstructured. So I think he means how advanced are, the AI agents within AI native systems today, and how reliable are the workflows that they then carry out. Yeah. That that is that's, that that is a that is okay. It's quite a quite a tough question in in the sense that, sort of how how long is a piece of string kind of thing. If you if you think of, the current AI, tools that are out there, and if you go to any any one of the the weeks over the last two or three years, every week, there's a new tool being announced. Every week, there's a there's a new, LLM that that's being released. But if we come and we focus specifically on the on the ERP systems, It's it's they often use any of the, agentic AI tools that are available and that they have a relationship with. So Microsoft would use their their LLM base and so on. But it but to come to a very specific, answer on what you what you're asking there, how how how powerful is are those agentic AIs? Now it's AI and agentic AI and machine learning, are all things within the same realm. So machine learning has been around in ERP systems for the last ten, fifteen, twenty years, if not going back further. So machine learning is basically using data and and learning the the the the pattern of of the usage of that. So you can see that as agentic to a certain way, but the machine learning, learns the behavior, of the workflow and then can can produce, the, reports on the variances, for instance, on things that have gone wrong. You've got generative AI, which we don't really use, in ERP systems other than to generate, text, and to generate, notes and so on. And then the the agentic AI, that's where we talk a lot more about what we call MCP servers. We talk about tool calling. And if you've done any programming, in AI, what you would find is agentic AI, you can you can start to structure that AI to tell it, the agent, the agent, which tools to call and which tools not to call. So you could very, very much define to a this the most minutest detail which tools that agentic AI can use in particular circumstances. And that's really where the ERP is going, is the ability to to laser focus or do surgical based math, chart generation, table generation, and so on using only very, very high high high potency, tools that it would call on. What you would find it in a, let's say, commercial AI, just a sort of let's call it chat GPT just for for for the sake of this conversation. Every time you ask it a question, it'll go out and find any tools that are available. And today, you might ask it the same question as tomorrow and the next day and the next day, and it'll give you different answers. That's because the toolset has not been regulated specific to your use. In an ERP, what we'll do is we'll say you may only use these tools. You may never use those tools. So AI can, and its agentic ability is incredibly you got to see some of the things that they do. It's amazing. But what really is is amazing to see as well is a lot of the, and I'm just gonna mention this briefly, Nat, if I if I may. And what has astounded me over the, many implementations that I've been involved in, what has astounded me and really impressed me is is the internal knowledge, the intellectual property, the the, the acumen, the business acumen that a business accumulates in its systems, in its people, in its workflows, in its ways of doing business, even brand, etcetera. If you think of of moving to a new system, you want to essentially transport all of that essential IP or acumen from the old system into a new system that will unlock that IP, that acumen, those workflows, the the the people's brains and abilities, and really give that that company a lift, a massive lift. And if you can get correct tools at the agentic AI and machine learning and all all those tools correctly focused and precision tooled, and even with tools like AduCap being plugged into that, you have got a winning formula. Whether it's gonna happen right away right now, I I I have my doubts that you have the perfect AI native system right now. And in theory, you'll only have in the next two to three to five years real true AI native systems that have been well tested and have got a strong reputation being built. One thing you must not forget is that the current, vendors of software, have got thirty, forty, fifty years of intellectual property themselves, acumen in their own systems, that they are rewriting into AI native systems. So just watch the space, and you'll see there'll be a number of of the current vendors rewriting completely rewriting their current, software into brand new, AI enabled, AI native platforms. So I went on a little bit longer than that, but I think that hopefully will give a a better view and that x and y information, etcetera. That's, again, the two dimensional that that that, Stephen mentions. But when you talk about a vector database, you please look it up. Go go on to Google and look look up what a vector database is. You'll see it's got hundreds, if not thousands, of of, coordinates where packets of data are being kept, structured and unstructured data. So there we go. So, hopefully, that's given you a bit of food for thought on your question. Okay. Cool. I think, we have tackled the other questions, as we went, and that brings us to time. Oh, no. There's some questions I have missed, in the q and a, from Jean Michel Richard. Amongst all the legacy ERPs, where do you see the AI tech road map going, Dudley? Amongst all the legacy, ERPs, if you've been around the ERP space for for a while, you'll you'll know that, there's always this constant frenzy of of of, the big name brands buying up smaller brands, smaller products, software products that are out there, and they absorb them into their ecosystems and rebrand them and give them a new, coat of paint. But, essentially, they stay, they they stay those products. So which, which are the the big name brands you're going to see? I think what's going to happen is if you take the list of the top 10 of which take Oracle, for instance, with their they've got two key products, which is the NetSuite and Fusion, for instance. You got Sage, which has got the x three and the Sage Intacct products, for instance. And then you can go to Microsoft, which has got Business Central, SAP has got Business One. You can go through that, all all those those those, software kits. And you'll probably find they'll they'll find maybe one of those and convert that and make that their new flagship AI native product and continue selling legacy systems. And the reason for that is it takes a long time for industry to move, and it'll take us another five to ten, fifteen years before a lot of companies have migrated into, full tech or AI enabled systems. If I look at some of the shipping companies, just to give you an example, they still work on old dot matrix printers in triplicate, in paper, in files, in big rooms, and they still haven't migrated. They're still working in DOS, for instance, and they're still working in COBOL, and they haven't even migrated. Why? Because the change management is so tough, to move from one system to another. I think it's more agile for the smaller companies to move, and I think the uptake will be probably be the mid market companies will be the first movers because they've got money, but they also got the the hunger to grow and to unlock the data that they've they've accumulated over the years. Hopefully, that answers that question. Okay. Another question then from Kiara Vincent. If the AI native system is working from a closed set of data, how does it incorporate information from emails, other data outside, the ERP, like a system like AchiCap, for example, or, just emails that haven't gone into the ERP. I guess some emails would go into the ERP, but some won't. Yeah. So so this in between world, between legacy ERP and a fully AI native ERP, we're going to see the the sidecar. So you're going to have, let's say, an Oracle database or a Microsoft SQL databases as your two dimensional relational database as your ERP where your transactional data sits, and you're gonna have a sidecar of a vector database. So think of these two working in tandem. So you got the the person who rides the the the motorcycle, and you got a sidecar on the motorcycle, if you can imagine that. The sidecar will be the vector database, where you can pull all kinds of data, unstructured data from everywhere, but the main driver is still the the motorbike. The sidecar gets all the vector database. Now the beauty about AI is it can call data from so many different sources into one single output. And you can have true data coming in from your ERP system, your legacy system, and you could you could enhance it by pulling vector database or sidecar. Some people call it a data lake or or a a data warehouse. You can have a sidecar of a vector database using a vector database to store your emails and your transcriptions and your salespeople's telephone calls. And all that data can be stored in the vector while your transactional data can still sit in your, in your Oracle or your Microsoft SQL database. That's gonna be the interim model, I think, for years to come until the new AI native systems are either released fully by the software vendors that you know the brand names or the new up and coming AI native first, VC backed companies. Pretty much like yours, Nat, if you think of AduCap. You guys have been backed heavily by venture capital. And and in that way, you've managed to grow because you've had the bandwidth to do that growth. We're gonna see the same happening with a lot of the new AI native, vendors that'll come on to the market. Has happened. I saw three of them. Just three companies took, it must have been close to $200,000,000 in the space of three months, just three vendors. So, Harriet, I answered your question. If I were you, I would wait and see, personally, see how far you can get with Xero, but keep testing the new solutions when you compare them to functionality. One thing I have noticed with, AI native VRPs is they're very, very happy to demo their tool. So you should be able to see what they look like already and see if they can handle your, data foundations and your intercompany elements, which I think a lot of them target as well. Okay. I think, that's all we have time for. I don't think any new questions have cropped up. Thank you to everyone who stayed. Thank you for joining. Hope you enjoyed this discussion. Huge thank you to Dudley, of course, for sharing his wisdom, and, we look forward to seeing you at a future AGCAT webinar. For those, who are interested and curious to learn more, we will be sending, as part of our follow-up, a survey for you guys to judge whether, you are how to assess your own unique situation on whether you should make that call, which Harriet's asked, on whether to switch to an AI native ERP or perhaps just transition to a core financial system that can allow you to get your data in a really good place that can set you up in two years' time to move to an AI native ERP. So watch out for that, ten minute questionnaire that'll come in the follow-up, from AduCap. Fill that in, and, you can begin to make the call yourself. Of course, otherwise, if you wanna learn more about AduCap, reach out to me. If you wanna learn more about AI native systems, and what to do when tackling this question of ERP, reach out to Dudley, on either LinkedIn, or email. Thank you very much for attending, everyone, and, have a great rest of your day. Hopefully, see you at another EdgeCheck event soon. Thank, you. Ned. Alright. So has not has not gone off the call. Oops.