Name: AI in Finance: Automating AP & AR for Real-World Efficiency
Uploaded: 2025-11-18T19:16:03.700Z
Duration: 1 h 8 min 52 s
Description: AI in Finance: Automating AP & AR for Real-World Efficiency

Transcript for "AI in Finance: Automating AP & AR for Real-World Efficiency": Alright. Thanks, everyone, for joining. We'll give about a minute or so to wait so for some more folks to join. It's always interesting to see where everyone's dialing in from. So Saligo CA, is that a location? Maybe from this legal office. Yes. Alright. Another thirty seconds or so, and we can get going here. Cool. Well, I know we have a lot of content today, so we're really excited to chat today about accounts receivable, accounts payable automation. I know it's a a topic that a lot of prospects and customers have requested to learn more about, and it's actually something that we're doing internally, and using AI to facilitate, those automations. So you're going to get to hear directly from folks on our own finance team and internal AI teams who have operationalized this. We'll start with just a little bit of a quick background on the Soligo platform, how we fit into the AI story. Then we're going to talk about real finance use cases that we've seen in our world at Zaligo, business discussion and on some of the the implications and some of the outcomes that we've been able to realize. We'll have a demo of our AR inquiry agent that's built using Soligo, and then we'll end with some best practice, key learnings, and some q and a. So we will save time at the end to answer your questions. If you have anything that comes up, feel free to drop it in the chat, and we will make sure that we get to it at the end. Now I'm Chris Ferraro, on the product marketing team here at Soligo. We will also have Thomas Neil, who's a VP within our finance department. We'll be talking about some of the the business use cases. And then Sandeep Gautam, who is a director of AI strategy, who's actually responsible for building out AI use cases internally. We'll be giving a demo of some of the agents that we've built in within our finance team. So that'll be really interesting. But first, it's always good to just establish a baseline. I know many of you are familiar with Solito, might be new to some of the folks on the call. But our platform, we're not just transforming how integrations are built, but also the management of those integrations. So a lot of different features like our low code builder, prebuilt marketplace, reusable intelligence, and AI code assistance make it really easy for anyone within the organization, including folks on the line of business, to be successful and build critical automations in Saliga. But IT's mind is at ease because they're able to federate access to those integrations, but do so with the right guardrails in place, the right security and governance where they can essentially federate access of the platform. Line of business can automate processes that they know best, and core processes, standards, and frameworks can be established in central IT. So that's really some of the collaboration that we go for and really what our platform can enable. And now we have all the bells and whistles of an enterprise grade integration platform, and that's SOC two, HIPAA, GDPR, really high uptimes. But we're not just focused on app to app integration, which you see all the way in the top left. That is ultimately where we grew up. But since then, we've expanded into oh, I think we went far into supporting multiple integration patterns and really whatever integration challenge is thrown our way. So b two b manager was rolled out last year. That's helping you connect with your EDI trading partners to facilitate the exchange of documentation. API management if you want to expose different APIs internally to other lines of business or externally to partners and customers via a portal, data ingestion, and now also the ability to build AI agents within our platform. And that's what we're going to be discussing today. Obviously, it's a really hot topic. I'm sure you're tired of hearing the word AI, but we're really gonna be focused on that component of our platform and how you can use Soligo to build agents specifically in the finance department is what we'll be discussing today, but, really, it's across your whole business that, that audit that this type of automation is possible. And now before we go into sort of the AI overview, we wanted to do a quick poll for you because we know everyone's at different stages of their maturation journey. So how familiar are you with agentic AI and its potential business operate and business operations? And this can just be honest. It's okay if you're not familiar at all. It's great if you're very familiar and you're actively implementing or if you're somewhere in between, but we'd love to hear from you just so we can kind of get a baseline for the stage that everyone is at. And we'll give it about thirty seconds to a minute. Okay. So it looks like from the early votes that there's a range between being very familiar, actively implementing, or at least somewhat familiar and understanding the basics and probably ramping up into 2026. That's awesome. So it sounds like, there's at least a baseline level of understanding of agentic AI. You're either actively implementing use cases or gearing up to implement, a critical part of your AI strategy in 2026. So this should be relatively familiar with you. If you if it isn't, feel free to ask any questions that you have in the chat again, and we'll get to it at the end. But this is very common across organizations. Right? You're getting some deliverable or remit from the CEO or someone else in the executive team where they say, hey. So and so competitors leveraging AI. Try and figure out how we can use AI within our business, how it can be operationalized, and you have a few weeks to come back and present a plan to you to me. And it it could be overwhelming at times. There's now an expectation that these agents or these AI workflows need to be live immediately because, again, competitors are starting to embrace AI, and it becomes hard to understand what are the real use cases that we can actually operationalize, and what is just sort of the the noise out there that we should ignore for the time being? Because there's certainly a lot of it. And this is why I don't know if you're familiar at all with the, there was an MIT study done. Again, not sure that the total accuracy that they had at that, like, ninety plus percent of AI pilots that failed to reach production. This study had eighty eight percent. But some of the key reasons behind this, in one respect, it's fragmented data. So data doesn't just reside in in one system. It resides across your whole tech stack, and this could be on prem or in the cloud. And these workflows are inherently disconnected. So you could have you could be leveraging AI within Salesforce, AI within Zendesk. All these point to point tools have their own artificial intelligence that's built in so you can use those tools easier, but you're unable to get the right context. And that's what we'll get into in in this slide. So as AI gets built into all these individual tools, it only really sees the data that falls within those tools boundaries. So, again, it could be ServiceNow with AI, Workday with AI, but they're all making these decisions in isolation, and none of them have the full business context. So if you look at the right side of the slide, that's really where the market sits today. Every platform is building its own flavor of AI to make users more productive within that specific environment, and that's great for localized automation, but it also creates fragmentation. And that's because valuable context is locked away in these other systems, and companies are struggling to really bring this all together. And I know we talk to customers all the time who say, like I mentioned before, they're under pressure from leadership to adopt AI. But most of what's available today is rigid and purpose built for really slim and narrow use cases. And the reality is that every company operates very differently. There's unique data models, approval paths, exception handling, and those nuances might not neatly fit into AI that's embedded in one app. And that's really where connectivity comes in. When you orchestrate data and logic across systems, you can combine deterministic rules. So if that's this, then that, or or maybe human approvals in the loop with AI driven decisions to build processes that are adaptable, contextual, and are really reflecting how your business operates. So how do you start with building that foundation? Here's what it could look like when integration and AI come together in practice. So you can see here so legal connects all the systems that your business runs on, whether that's a CRM, ERP, ecommerce, support system, so AI can now get the full picture. Because AI is only as good, as we as we mentioned, as the data they can reach. Then we can clean, enrich, and structure that data so your workflows are reliable, your data comes in standardized format, and it's AI ready. And then finally, we're putting AI to work. So once your data and systems are connected with Soligo, AI can start driving towards measurable results, whether that's automatically enriching CRM records, resolving tickets faster, surfacing customer insights, the list goes on. And that's what we mean by operationally operationalizing AI. It's taking AI out of isolated pilots and embedding it directly into the systems and processes that run your business. Now as companies are going from experimental AI to really operationalizing it, we're seeing the rise of agentic AI. So that's systems that don't just generate insights, but can actually act on them. And we're gonna see that a little bit in the demo later. But we're really referring to intelligent agents that can perceive what's happening, make decisions, and take actions autonomously, either without waiting for humans every step of the way or getting human input or approval at some stage in the process. So it's that shift that you see here from AI that creates to AI that does. And connectivity is ultimately the foundation of all of that autonomy. Without the seamless flow of data between these different applications, even the best AI agents can't perceive the full picture or take meaningful action. So, without further ado, I'm going to welcome in Thomas Neil on our finance team, who's actually going to talk through how we went about implementing and thinking about AI within our finance organization. So, Thomas, I will let you take things from here. Alright. Thanks for that introduction, Chris, and, yeah. Great topics there you already covered. So, yeah, once again just to kind of reintroduce myself, my name is Thomas Neil. I'm the controller here at Silego and everything we're going to talk about today are things that we are actually building internally. Internally. So I mean I'm really thrilled to be here today talking about agenik.ai. I'll be honest, I've attended my fair share of webinars specifically focused on AI and I usually leave those sessions feeling I mean, I guess smarter, but also thinking, okay. That was cool. But how do I actually do this? So today, I didn't wanna just talk about AI. I wanna show you what we've actually done here at Seligo and how we've brought Agintiq AI into my team's daily workflows. Real life, in production, no theoretical whiteboard stuff. Because while buzzwords are fun, I really dream of a world where balance sheets kind of just balance themselves. And I think that's really where we're going. So let's start with a confession from a controller. Right? My own confession. So a lot of controllers, myself included, once upon a time say that we're adopting AI, but what we're really doing is using AI that someone else has built inside of a product, which is kind of what Chris covered a little bit earlier. So maybe you've got a procurement system with smart procurement agent, you've got an AP platform that codes GL accounts automatically, or an AR tool that predicts cash application, or maybe the timing of cash flow for cash flow forecasting. It's all great stuff and it's extremely useful. But the way I see it, there's kind of three problems with that approach. The first problem is the logic behind that AI is actually owned by someone else. So it's kind of like you're borrowing someone's car, but you can't adjust the mirrors in it. You might look really cool, but is that really gonna work for you? Problem number two, as Chris already covered, it's looking at data inside of that specific application or system. It's not leveraging the data across your full tech stack, and that's really where the magic is gonna happen. The more data you can feed to an AI model, the better outcomes you're gonna have. Now the third problem is one that I think is a very big problem, is that every smart AI application button that at least my vendors are adding to the applications we use, they seem to be behind a paywall. So before you know it, your AI transformation starts to look suspiciously like a subscription renewal nightmare. It's just not something that's going to fit within most budgets, at least not within the budget that I'm given here. So that's when I realized if we really want AI that truly works for us, we need to start building it ourselves. And I'll admit, I was stuck for a while. I was doing exactly that, using application based AI, copying pasting stuff into an LLM to make my emails look much, much better. But I hadn't really figured out how Agenik AI was gonna work. And from the poll, it sounds like a lot of people on the call already have, so that's pretty amazing to hear. So let me just share my light bulb moment. So it really all started with a humble little automation that our own CFO built. He really wanted to figure out Agenik AI, and so he worked with Sandeep, who you're gonna meet later, to create a very basic functionality that connected our we use Gmail, but Gmail to Salego to an LLM just to filter emails for them. If anyone uses Gmail, there's a lot to be desired desired in the filtering function. So it was nothing fancy, but it was ours. We controlled it, and it was something that was fully customizable. And I really watched in awe as this little thing became more and more useful. And that small success completely changed how I saw the future of finance automation. It proved that Sylligo plus AI wasn't just powerful, it was practical, it was scalable, and most important, even I could do it. So today I'm gonna share a few real world examples of how we're using agenik.ai and AR and AP, and how you can start doing the exact same thing very easily. And I promise you're not gonna need a PhD in computer science to make any of these work. So alright. Make sure this slide goes. There we go. So every automation journey starts with a simple question. Right? And it always has to. You have to say, what are you trying to achieve here? So it's always important to kinda define the goals. So as a controller, and maybe we have other controllers on the call, my goals are pretty straightforward. Control cash flows and in inflows and outflows tightly. Right? I I manage all the cash in and out of the business. Keep financial data accurate and compliant so we can have all of our statutory reporting on time and accurate without any material findings. But now in the age of AI, we get to have a new one, right, which is maximize the agent capacity and utilization. So that's something that's definitely new to my function, but something that I think I fully embraced and most people are now fully embracing. Because if no one's using your automation, it's not really innovation, it's essentially just a fancy screensaver that no one's even going to look at. So before we go any further, let's go ahead and take a quick poll. Give everyone a few seconds to answer. Interesting. I have to admit, I was a 100% expecting to see AP and AR at the very top. I I think those are the most common ones that we hear about. I see expense management's creeping up there. I'm very surprised to see reconciliations on reporting this high. So yeah. That's actually one of my kind of pet projects and I'm just gonna digress a little bit here. We're not actually covering it in this presentation as this is mostly about AP and AR. But one of my pet projects right now is exactly that, is trying to figure out how to use AI automations in order to create GL accounting automations. And specifically something that I found that I found very interesting was that you can use an LLM in a very well written and structured prompt to actually automate journal entries and reconciliations. Now it's not quite there yet. I'd say, AI is very enthusiastic, but it's occasionally a little bit too creative when it comes to something that's structured, though it might be my inability to prompt it properly. But I do think as AI evolves, the use cases within GL accounting are gonna become vast. And I think that's really the next big area that I personally want to be focusing on when I think of AI automation. So let's jump right into the APPs. So this is a very simplified AP and, travel and expense automation heat map. Obviously, this is not complete. It's just kind of to give an example of places where we've been focusing on Agintiq AI. And the most important thing here is just to understand your colors. Dark green is full Agintiq AI automation. Hands off. Just let it run. Light green is partial automation. I call it keeping the human in the loop. Sometimes it's truly a human in the loop for review. Sometimes it's that there's pieces of this process that the automation cannot fully handle that need to stay manual. And light gray is things that really stay manual. Most of the light gray is not that you couldn't automate pieces of these, it's that I personally like to have the control there. You can see, for example, reviewing a payment batch, approving a payment batch. I don't really want AI out there approving payments going out. The very interesting one to me is the review AP mailbox and forward AP mail forward emails to the AP system. You could argue that could be dark green. I think it could be dark green. You could totally use Agintiq AI for that. Now the problem that I continue to have with email based AI, especially Agintiq AI, is that I do not allow my AI to click any links or to open any attachments just because of the phishing concerns. I'm sure there's a way around that. I'm sure we'll figure it out sooner or later, but I am always concerned about phishing, so I try to be really careful there. So I'm going to go into three automations we have built that are shown on this heat map, and those three workflows are going to be AP invoice processing, expense report approval, and travel and expense inquiries. So jumping into the first one, which is going to be the AP invoice processing. So all of these slides are going to be set up somewhat the same. Right? You can see at the top, you have your source systems, and these just happen to be the systems within our tech stack. In the middle, you have siligo, which represents the siligo platform, which connects all of this, and that's where your LLM sits as an endpoint within the siligo platform. And then you have your outputs, which are the different systems where output may be available. So for AP invoice processing there's a lot of this that's already kind of built into certain applications, but once AP applications or platforms. But once again you don't control it, you don't know exactly what is being checked, and so I made the decision to go ahead and build on top of that. And I found that one, once again I control it, I know it's being checked, and two, that it actually can be much more accurate because I can continue to tweak it on my needs. So here at Zaligo, we currently use Coupa for our AP and procurement platform. So So you can see the three sources here are going to be Coupa, are going to be Zendesk, and are going to be NetSuite, which is our ERP. So the Zendesk one is, extremely important, and you can kind of see if you look up to the far right, you can see there's a copy, a screenshot, of my, prepaid expense policy. And so Zendesk, we use as a repository to hold all of those documents that are what the AI is pointed to to understand our processes, our FAQs, any documentation as far as procedures, you're gonna find that that's key in all these automations. And there's a really important reason why. Because as a controller, as an accountant, once again I'm not a coder. I can write, I think, a pretty good prompt, but, you know, I'm not an expert at it. I'm not gonna be in there writing prompts and changing code and all that to make this work. And so the design that we came up with is that if you have the LLM pointed to these repositories, I simply can change my policy, I simply can update an FAQ document just like I normally would when a new inquiry comes up or when I change my policy, and when I save those into that repository, the LLM automatically updates. Right? So there's no need to touch the automation. It's like giving your AI a new brain without having to do all that messy coding. So once again, so we use Coupa. So in this automation, Saligo is first going to go into Coupa and it's going to retrieve the first invoice that's ready for processing. Then it's gonna go to the LLM and it's gonna read the details of the invoice. Now this is interesting because whether, you know, the text, the numbers, the line items, all that messy unstructured data that at least I used to fear because most automations couldn't handle it. You know, whether it's OCR, whether it's computer vision, whichever term we wanna use, it always had struggled with unstructured data, which mostly invoices are. AI doesn't have that problem anymore. So AI is able to parse that unstructured data efforts effortlessly, then it goes and checks all those company policies that are stored in the repository, then it goes through past vendor bills from this vendor, whether that's in the AP system or whether that's in NetSuite, and it can find the most common GL coding, it can match the line items, and it can pretty accurately, actually very accurately, figure out the correct GL coding. Because one of the problems is you imagine you change AP systems and all of that memory of where all these bills used to be coded to went away. And most AP systems and their AI automations, while they're connected to your ERP, they don't have AI set up in a way with a connector to your ERP to be able to pull back that level of information. So the system doesn't one note is the system doesn't actually approve invoices. Right? We're just talking about invoice processing. So it processes the invoices, which then triggers the workflow, that goes to, you know, the the buyer's review and the accounting review. Right? So this automatically has that balance between autonomy and control, which is going to be key in any automations that we talk about. And again, unlike built in automations, like, we own the logic on this. We can change it as we need it. There's no mysterious black boxes about what's actually going on in the background. And most of all, I don't have to pay additional subscription fees to any of these applications because I'm not using their built in AI. In fact, I believe that I've been able to build something that's better. So going on to the next one, which is the expense report approval agent. So this one's, very interesting. Now we built this one for one specific use case. Right? Here at Saligo, we have a monthly phone and internet reimbursement. It's simple. It's repeatable. It's very time consuming. Right? Whether you're the manager that has to approve everyone's or whether you're the accounting team that has to approve every single one of these submissions. And it's also very low risk. Right? This is basic. It's gonna be reimbursed. There's not much room for abuse even though the AI component of this does check to make sure that there is an abuse of the policy. But you can see it's kind of the same setup. This one's a little bit more simplified to where you're gonna have Coupa which is our expense system. So it's gonna go into Coupa. It's gonna find it's going to find the actual submissions that relate to that phone and internet internet allowance, and then it's gonna go into Zendesk. Once again the repository, find as you can see on the right hand side, the travel and expense policy. So right hand side, the travel and expense policy. So that's important because what happens if, you know, next year I say, hey. I'm gonna give everyone $70 or a $100 back or maybe less. Right? Well, I can just change that in the policy and automatically the AI component knows that the policy's been updated because that's what it's referencing, and it's gonna be able to automatically approve those and they won't get flagged as an insect as an exception. Now there's two outcomes here. Either one, it reviews all that information including all the information in the attached receipt, matching it to all the field level details, and checking to see if there's a duplicate expense transaction already posted in that month. So it does all the same checks my accounting team would do anyways. If it's compliant, this one fully auto approves. Like I said, I believe this to have very low risk, it's internal only, it's not facing my customers, so I allow it to go ahead and be fully auto approved. And then it leaves a note within Coupa letting us know that the AI automation approved this. On the other hand, if it finds something that's out of compliance, it's a duplicate or anything like that, or the invoice doesn't match, or the invoice just doesn't make sense for that transaction, it's going to automatically flag that, leave a note about the what's wrong with it, and then it's going to send it to the manager directly. So then it goes to the normal approval flow which is the manager and which is the accounting team. So we have a fair amount of employees here at Silego, so you can imagine how much time that actually saves, not only for my own team, but for managers around the company. I think that this is, you know, one of the more beloved automations because of the amount of time that it saves. And it's fast, it's consistent, and it's fair. Right? And this will come up in a couple of the automations where when you introduce the human element of my AP team checking these and then pushing back and saying no this is out of policy, sometimes you kind of have some arguments there even though it is obviously out of policy. But there is that human and emotional element there. When it's an automation that's telling people this and it's automatically rejecting it, you don't see that. Right? So it actually takes out time there as well. And there are some expense platforms out there that have auto approval features like this, or they just don't if certain things are considered in policy they don't even send them for approvals. But once again all of the I've talked to many of those you know providers and they will not tell you what's going on in the background and how they decide that this is compliant. Like I mean one example that I've been talking to is Nivan, which I really like Nivan, but theirs is, hey if it's in policy then you don't have to prove it. But there's the in policy is just is the submission have an attachment and are the amounts in policy? That's great, but no one's actually checking the invoice or sorry, the the attached receipt. And so there actually is no check there. It's just assuming that that's gonna match, which I find to be dangerous. So I feel like this is actually more control. Once again we can control it, we can change it as needed, and it gives me better oversight and comfort into the spin that's going out the door. Now to the third one, which is, by far my favorite, which is the T and E Inquiry Agent, or as I like to call it the Thomas Neil Sanity Saver. As a controller I'm always shocked how many people think that I'm the right person to ask about expense inquiries. I get these all the time, and it is a little bit painful. Everyone should know where the expense policy is. It's been published. It's been communicated. But getting people to read it, it's kind of difficult. Right? So we created a Slack based bot that you can then query, and you can ask it any t and e related question. And actually we've now further expanded this to be any finance based question. Right? So you have the same internal questions over and over. Once again, these are internal only, so I'm much more confident in full end to end automation, but can I expense this? Do I really have to book economy? You know? Why does the policy hate funds so much? Right? Well they can ask that to the bot, and the bot can answer those questions once again based on the travel policy travel and expense policy that's saved in a repository. So it's very simple. You can see on the right hand of the screen. This is back when it was part of the HR bot. We've since then split it out. I just need to update my screenshots here. But you can see this question right? Hey I'm flying from San Francisco to The United Kingdom. Can I book a business class ticket? Well the answer is no. And a lot of times when I tell people the answer is no, I end up in a five to 10 message Slack thread having to discuss why the answer is no. This simplifies it. Someone sends me an inquiry, I say hey, remember you can go here, use this bot, it'll answer the question for you. I just point them there. I don't even have to handle it anymore. And as you can see it goes, it quotes what the policy says, it gives you a link to the policy, and then it also gives you the option to open a ticket if you don't agree with what the policy says or if you have further questions or if it did not actually answer your question. So as you can see it's pretty basic. It's Slack, Soligo, and LLM, and I should say any LLM you have. If you have an LLM, you just add that as an endpoint, and that's kind of covered in this Soligo bubble here, and this is going to work for you. And then you can once again point it to a repository so every year when you update your expense policy and new thresholds, maybe changes in the policy, the LLM automatically knows that because you updated that policy. And the output could be two things, either one, a Slack message giving them the answer. If they're not happy or don't feel like that fully answers it, then they have the option to submit a ticket, which would then go through the Zendesk system. So I think you can imagine this goes a lot further than just T and E inquiries. Right? And that's one of my favorite things about this one, whether it's payroll inquiries, equity inquiries, I mean, AR inquiries, procurement inquiries, anything. So what we've done now is we've kind of created or we have created a finance AI bot and we house all of our policies, all of our FAQs, everything there. Right? So any question that you would ever come to someone on my team with, we can push them to that bot. And this has been live for a while, and as we continue to train people to go to that bot, we've seen significant decreases in the amount of inquiries that come to our team because people are able to self serve. Right? Now when you look in that repository across all those policies, it's hundreds of pages of policy, right, and FAQs. I get it. I get why people don't want to read that. Right? But by having the bot do the reading and only answer the relevant question they have, they actually will use it. And so it's actually saving the teams a lot of time. And and I'm really excited about this one and the scalability that this represents, and also just how simple this one is to set up. That's the other beauty of it. So before we go on to AR, I'd like to jump into a second poll, which is what is your top priority when considering AI and finance? Give everyone a few seconds to answer. And while we're waiting for answers to roll in, feel free to throw questions, in the q and a. I am staring at it. If anything pops up, I will do my best to answer it real time. But more than happy to digress where anyone needs more information or wants to talk about something slightly different. Okay. Oh no, I think that's good, right? I know for me this is exactly how I would stack rank them. Reducing manual effort and saving time, right? And I think a lot of people conflate reducing effort and saving manual time with reducing costs and conserving resources. I don't view them as one and the same. I think that one can be, you know, obviously reducing manual effort and saving time can be the outcome. The outcome of that can be reducing resources. I know for my team I'm luckily not in a place where I'm being asked to reduce resources right now even with all the automation that we've, put in place, and a big part of that is because as we have these automations in place doing this kind of basic transactional level work, I upscale my teams and do take on, you know, a broader range of activities to take on a more analytical mindset of big data, and we're doing validation and checks that we never have done before that is now in just increasing the accuracy and compliance. Right? So I think that those two definitely go hand in hand. I think it's always really important to think of two things as you're introducing these automations. You know, one is once this automation is live, what what is the human in the loop concept? Like how do you monitor, the gov or have governance around this automation to make sure that it's doing what it's expected to do, and make sure it's having the outcomes you want, and that it's not going rogue and doing something strange. Right? Someone still has to be there to be able to review it and to be able to understand the outcomes of what's happening from the automation. And then the second one is if it does save the amount of time that you expect it to, and you always should have an expected time savings when you're submitting an idea for an automation, What do those resources now use that time for? And I think what you'll find is there's always more that they can do that's like a little bit higher level that's more of an analytical view, that they can now perform those which give you so much better accuracy and compliance, get you ahead of month end reporting, do soft closes during the month, all these things that make the process all your processes better, you can unlock that time, by focusing on these automations. Okay. So now let's jump into accounts receivable. Right? So same thing, very basic heat map. The only difference here is you can see the introduction of like a dark gray. So that dark gray really just represents, Soligo, some of our native offerings. So these are things that I already had in place with Soligo that we do really well, and so there's really no reason for me to slap AI on top of it because it's already happening. But then the rest is the same, the dark gray and the light gray. And the two we'll specifically look at here is an AR inquiry answer agent and an advanced AR collections agent. So that AR inquiry agent is kind of similar to the one we just looked at for T and E, but on the other hand this one is going to be email based instead of Slack based, and this one is actually going to be customer facing. So that makes it very interesting. And I think anytime you have something that's customer facing, it increases a lot of complexity because you have to be very careful what goes to your customers. But I'll talk about that in a second. So in this one it's actually fun because I actually used AI to come up with this one, and I knew that we were gonna have I knew we were gonna have a lot of repetitive inquiries. I'm in the a core AR inquiries mailbox. I look at it. I kind of see the traffic that flows in and out of there. I don't live in it every day, but I knew there were gonna be some common ones. So I took Gemini because we're Gmail once again. So I took Gemini, connected it to the AR inquiries mailbox, and then I told it go look at the last year of inquiries that we have received from customers, and I told it to exclude some stuff. So it's a little more complex prompt than that, and categorize all the requests that we got. Then I ran that through a few times and kept fine tuning it, and I was actually shocked by this. And and I don't know if it says more about our our processes or our team, but I'm curious if other people would find the same thing. But what I found is that there's essentially four requests that come into AR inquiries that make up 35% of all of that email traffic. And they are very simple requests. The first one is send me a copy of the invoice. Second one is I want a statement of accounts. Third one is I need you need, so we go, we need to update the billing address for this customer because the emails are going to the wrong place. And the fourth one was they want a copy of our w nine because we're not set up in their system yet. Those are very easy, but I don't know how many people here are are you know NetSuite but I think in ERP it's kind of the same. Have you ever tried to go into an ERP so you're working in an email you see an email hey I want a copy of my invoice alright that's a simple request right just a couple seconds. No. You read the email. You write the reply. You go to the ERP. You download the invoice somewhere. Then you attach it to the I mean, it's probably a good ten minutes by the time you do all those actions. I mean, if you're really fast, maybe five. Right? But it actually takes a lot of time. And so if you have a bunch of those questions coming in, that's a lot of lost time that we can get back. Right? And so what we oh, sorry. Yeah. There we go. One more. And so with the AR inquiry answer agent, right, once again a pretty simple setup here. You're just having that AR inquiries mailbox, you know, connected to Soligo, connected to the LLM, and then connected to NetSuite. Right? And why is it connected to NetSuite? Because the second request well, that request, I need to copy the invoice. Those live in NetSuite. So based on the billing email that's coming in, right, you could either use maybe the domain, or you could use the specific sender of that email. Maybe they're set up, as one of the billing or contacts within your ERP system, so you could rely on that. We normally rely on the actual actual, just, domain name itself because that's pretty consistent across the customer accounts and the email domains. So we utilize that, and we go and we find the customer, we find the invoice, we pull the most recent invoice that they're asking for, attach it to the email, and the key here is, and it says it somewhere over there, but we don't send the email, right, because this is customer facing. And maybe after a couple more months of seeing this and seeing no errors, I'll get comfortable with this, but I want someone to review this. Right? So it creates the email, it attaches the invoice or the statement of accounts or the w nine, and it just saves the draft. Right? That still saves a lot of time. So my team comes in and instead of going and going to all those systems email by email by email, they have a folder that has all of the responses already in it, and they just go from response to response to response and check. Is this invoice belong to this customer? Is this what the customer is asking for? Right now that's maybe a minute in email. So let's say generously you could actually do those emails in five minutes. Well you're saving four minutes in email, I think, I can't remember the actual number, but there's like 200 emails in there, right, a day I think? Maybe that's a week. A day sounds too much. But anyways, you can imagine that's a lot of saved time. And that's just the four basic requests. So you can continue to expand that in a meaningful way to whatever your most next most common request is. Right? And you can keep building on this and building on this. And this actually doesn't have it in here. One of the things we've recently switched to is the same concept to where we built a repository. In that repository, we created an FAQ. We created our a well we already had our AR policy, and we point the LLM to those so that it has kinda it can say, Hey. This is the kind of category of question I'm having, and this is the kind of response that I'm giving. Here's kind of the guidelines of the response that I'm allowed to give, and it can build a better email. And I think we all know, I mean, most LLMs can write a better email than any of us can. So you actually end up with a pretty good outcome from this. Once again, I think the human in the loop part of this is still extremely important. And then you have NetSuite connected for the other reason that if there's an update that needs to happen in NetSuite, you can use Soligo to connect to NetSuite so your LLM can go and can update that customer billing address within email address within NetSuite. Right? Once again, I still have it save an email that confirms what it's done, and the main reason for that one, you want to give acknowledgement to the customer, but two, I want my team to go and check-in NetSuite and make sure that that one happened. I was pretty concerned about that, but so far, we've actually seen it be very accurate, and we haven't seen there be any, issues with it. So I'm still having them check it, but there might be a day pretty soon where I feel confident that they don't need to check that and can just send that email as long as it was the correct, you know, response to the inquiry that they had. I'm sorry. I'm reading the q and a real quick. So I I think that's very interesting. So when adopting Eugenic AI so the question is, when it comes to adopting Eugenic AI in finance, what do you see as the biggest challenge? Building the business case, ensuring data quality, driving oh, perfect. We can put it on the screen. Or driving user adoption across the teams, etcetera. So one thing I will say that I think is extremely important, and I think this was a misstep I've seen in every automation program before. In another life many years ago, I ran a RPA automation program myself, when I was more in a finance transformation role, And I took this exact same misstep, is you need you need two things. One, the ideas need to come from the users. Like a lot of people think, oh, I have this idea of what this should be, and and this is a really good automation, and you kind of push it out there without the buy in of the people who will actually use it, it's gonna fail. Always make sure you have those, hackathons, whatever you want to call them, with the users who will actually be using these so that you can discuss what they actually want to see and what they will use. That is how you're gonna drive the best adoption. Then the second practice is it needs to be where they live. Right? You can see all of these and that's the beauty of Agenik AI. Right? All of these are built into responding to something that happens that the team would automatically be responding to. Right? Whether that's an invoice submission, an expense submission, an AR inquiry, a T and E inquiry. Like, the team would already be in there having to respond to that, and the AI is just picking it up from a workflow that already exists, and then it's taking it forward for them. We had a lot of use cases or or ideas that we built where you have to go and manually trigger the AI to do something that you want it to do, we find those don't have as good of adoption. Right? You wanna keep AI, where the people live. Now the other part is, yeah, building a use case. Because in AI, especially in my opinion, in the GL accounting space, ROI is extremely hard to prove out, especially if you're not cutting headcount based on AI, which once again I'm not doing. Right? I'm building for scalability. I'm building for, you know, the the next, you know, 100,000,000 in ARR, keeping my cost contained, and keeping my team the same size. Right? Like, that's my goal with AI. So it's really hard to convince people to put money into that, you know, use case when you're not gonna be able to present a direct ROI. But at the same time, I think that you can show that aspect of it, and it's the scalability aspect of it. And building the business case, you still have to kind of focus in on the time that is going to be saved, with doing this. And I think in all of these automations, which I, you know, we said earlier, with the compliance piece kind of being the second highest ranking, that compliance piece actually comes better, especially manual processes when you're using AI because AI does the same thing every single time. It's exactly what you ask. It does not make data entry errors, like humans will. So you actually get an income increased compliance automatically when you have AI that's a good use case and is getting used well. Okay. Let me jump to my last slide and last use case here, which is my by far my favorite. And I think if you haven't had the light bulb moment yet, I think this one really is that light bulb moment. And that's AR Advanced Collections Agent. Right. So what is Advanced Collections? I think we probably already know. But you know, hey, things are really past due. Right. In this example you can see here, we usually start this about thirty five days past due. This example is fifty. Right. But if an invoice is fifty days past due, you have a problem. Right? You've sent out multiple automated dunnings letters. You've got no response. You've probably already looped in the AE or the AM, the salesperson, the CSM, whoever you can. You can't get a response, like you're gonna have you're almost thrown a hail Mary here to make sure that this customer pays attention and pays this bill. So if you're gonna do advanced collections well, you should look at a lot of different data points. In this example, you can see you have Gmail, which obviously that's just using the mail. You have Gong. Right? So we use Gong. We record customer calls. We have transcripts of customer calls. We know what has been said with this customer. We know the interactions with this customer. Gong has its own AI to summarize that. We just have to go pull that in. Right? We have NetSuite, which has all their historical payment data, has, we we store the AM or the salesperson contacts there, the CSM contacts there. You have and also you have your invoice to sales orders. Salesforce, right, you have all the notes on the Salesforce op. You have the signed contract, and Zendesk, once again we use it as a repository. In this example, we also use it as our ticketing system for our for our support team. So you can go in and see does this customer have any open support tickets, right. And if you're trying to figure out why a customer is not paying, the answer probably is somewhere in there. Right. So in this example, we're following up with a customer. They're fifty days past due. So it puts together a nice email, somewhat nice, a little bit, you know, firm of, hey we're gonna cut you off. Right? Like you haven't paid your invoice. We're continuing your service. You need to pay to avoid disruption. Right? It also notes that, hey we noted, and this is old, right, but that from September 27 you sent us an email telling us that you've sent us payment and it was on its way, but we still haven't been able to find that transaction. If you have, please provide the remittance information. Right? That's the NICE email. And then below that yellow line is the internal notes. Right? A little bit dangerous here, but the team has to delete these. Once again, this gets saved as a draft. You can see it doesn't even have the recipient in it because I don't want someone to accidentally send it without looking at it. And it tells you their payment history that, hey. This customer's had late payments. Right? They've ranged it up to eighty five days late. But all that considered, they still have always pay paid the bill. Right? So there's a little bit lower risk there. It summarizes the email history that it could find that, hey. They told us they were gonna pay. We couldn't find it. Support tickets, no support tickets to be found, and it tells us who their account manager is so that way the team can loop in the account manager on the next escalation if they have not already. So you can see that's pretty much and then it attached the signed quote, the sales order, and the invoice all to the email. Right? So if my team was out there having to respond to this, that's a lot of research to do. I mean I would say that's a twenty to thirty minute task if you're really good at it, automatically. So they come in, this is automatically in their queue, It went and I should have mentioned it first goes to the aging, and then based on certain age per age date parameters, it creates these emails. Right? So it's actually looking at the aging in NetSuite, deciding how aged something is, and then it's creating these emails on behalf of the team, that then they just have to go and review. Once again customer facing, nothing goes out with someone on my team reviewing it, but it still saves them all that time, to do those reviews. And like I said, so that really shows, I mean, how many things you how many applications in your tech stack you can get involved. Complex process like advanced inquiries that you can easily set up. I mean, I say easily, this one's harder to set up, but that you can set up using Soligo. Right? You know, and that's kind of my big takeaway here, right, is that a generic AI isn't some distant future thing. It's here, it's accessible, and you're able to design it and control it. Right? You don't need to be a data scientist. You just need some curiosity. You need a platform like Soligo, an LLM, and the willingness to let go of, hey, this is how we've always done it mindset. You know, it takes a while. Not everything was successful in the first run, but go out there. Start building these things. Start trying these things, and you'll be amazed at what you'll be able to come up with and how fast it'll progress. So now I'm going to hand it over to Sandeep and he's going to walk us through a few demos. Thanks, Thomas. K. Let me share my screen. Okay. So we will demo demoing the AR inquiry flow here. Here we have this our NetSuite instance. We have a, simulated customer, Celio Networks, who is a customer for Celio. And, he has, that customer has few contacts with us, like, AP@silico.net and AP incoming. And we have AP at, silico dot net as a primary billing contact. And we have John Miller as, one account in contact with, the customer who is sending us emails. And then this customer has few transactions with us. If you go into the sales transactions, he has few open invoices, three open invoices with us, and then different sales orders and few other transactions, with Zilligo. So now going into the email. Right? So here we see, a r-demo@silico.com is a mailbox that, where we receive, inbound requests, for the AR inquiries from the customer. And in this screen, we're seeing five emails from, John Miller, the contact at Siligo Networks, and, all five are different. So one, he's asked for bank account details, other for w nine, other for contact details update, other for account statement, and other for, open invoices. So we simulated, different requests different possible requests for the demo from one customer. So I will go ahead and, run the flow, our IVO flow. And then so this is the IVO flow that, responds to the AR inquiries automatically. So this takes few seconds to run. So in the meantime, I'll just quickly give a glimpse of the flow. So here, what we are doing is we are reading the emails, from the arfanddemo@silico.com, and then we are trying to get the email history, complete history till date what happened on the email. And then we're extracting the customer detail information so that, we know who is asking what. And then we go back to NetSuite, our system of record for customers, get the details about the customer. And then we are trying to understand what the user is asking. Is he asking for statements or invoices, or is he asking for bank account? Is he asking for contact dated update? Is he asking for contracts information, or is he asking any general, inquiry related, information. So once we do categories, we have different actions that, are being done here. So, yeah, here we're getting going to NetSuite, getting the invoices, going to get NetSuite, getting the statement information. Here we go into NetSuite, getting the primary account, details, the contacts, and then changing the billing contact details, etcetera. So yeah. And this can be extended to different different things. So here we have NetSuite as a primary thing, but let's say if you're hosting, information in Salesforce, the contract information, so we could go to Salesforce and get that. We can go to Zendesk and drive to get all the policy informations to answer questions. So adding in more tools is is is very flexible with, using Celico, and then it'll be custom built to your business use case as to where the data resides and how do you want to, respond, etcetera. And then you have complete control here because, the draft control is here. You can change the tone of how the message is, going out and all that is, feasible within Silico. And then once all the drafts are created, we move them to a different folder, so which I'll show it, there. So now the flow has completed here. We'll go to the inbox that we will see. So if you go go back to the invoice inbox here so previously, there were five emails in the inbox, but, all of them are moved to the draft status. So any, accounts that you will remember whenever they come into the inbox, So they go to the draft states to see that, okay, whatever the a could answer, it already, processed them and then moved to a different folder. So the AR rep will come into this, draft section, open, the email. So here, the customer is asking for bank account details. So we have a message here to the user, and then we have attached the bank, letterhead that we use, within Silico. Okay? And then if you go to the next email on w nine details, same thing. We fetch the w nine details from the Sellego internal document thing that the AI team uses, and then we try to, give a draft to the attachment. Contact details updates. So here, the customer is asking for, changing the primary address to a p f and incoming at silico dot net, to change the billing address. So here, if we have seen before, we had a p@silico.net as the primary billing contact. So if I refresh the, page here, you can see APF and incoming at silico.net as the primary billing contact, update with this thing. So yeah. So if we see here so the primary billing contact has changed. So based on user request, we are also making a right request to the source of, truth in NetSuite here, system of record, to, accommodate whatever the user is asking. So once the contact update is done so here we are responding saying, please note that we have updated the billing contact details in our records. So now the rep comes here, and then just looks at the draft and, sends it out to a customer. Same here for account statement also. So if you see this, the customer was asking for a consolidated account statement till doubt till date. So if you open the statement, document so this is the NetSuite, statement that is generated manually by going through the system and all. So it's all automated where within the email, we have, the statement, generated and attached as a draft for the rep to just send it over. So yeah. Same for, open invoice also. So here, the customer was asking for all the open invoices. So the agent, read the email, understood the intent, recognized who is asking the, request, and then it went to NetSuite, opened the, customer, and then here this customer has, I've shown you. They had, three open invoices. Like, if we go into open status here. So they had three invoices, 57, 56, and 58. So all those, open invoices are attached here as per the customer request. And if you open this, any of these things, so it will open the proper invoice PDF that we see it in NetSuite, kind thing. Yeah. So yeah. So here we have, demoed, five different use cases, but, we can extend the flow to, pretty much whatever, your business, needs are primarily beyond these as well. Okay. Yeah. That's from a demo standpoint. Awesome. Yeah. Thank you, Sandeep. That's that was super helpful, and it's it's great to actually see what it looks like in the inbox automatically creating messages, and I'm sure that's that's saving a lot of time for the teams. I do wanna have at least a few minutes to get into the q and a. We had some questions that rolled in there. So if anyone has any additional questions, please drop them in. First question from Chris is, are there recommended LLMs to use for these scenarios or all about the same for finance related automations? I don't know So I'd say they're they're all about the same. I would really if you already have an LLM in your organization that you're using, continue using it. Use the one that you're comfortable with. Right? I mean, here at Zaligo and and myself in particular, I've been most comfortable with chat g p t, so that's the one that I find I'm using the most just because I've used I have used it the most, and so I'm best at prompting it and understanding kind of the intricacies of how it could work. I have plenty of other people in my network I know that, Swear, Byanthropic, and By Claude. I've used Gemini for some things as well, more internal, not actually within Sligo flows. But any LLM you have, you can connect us to Sligo, and it should function about the same. Yeah. I think that's a great point there is regardless of the LLM that you that you use, maybe you could find that one LLM, like Anthropic, is better for a coding specific use case somewhere else in your organization, and maybe you're using OpenAI within your finance department, you can we have universal connectivity, so we can connect into any of these LLMs. So you have that flexibility versus maybe being pinned down and having to use a particular model that might not be the best fit for that use case. Alright. We'll get to one more question. I'll I'll leave this one to Thomas, and I think we only have, like, thirty seconds for this one. But what are your organization's plans for agentic AI adoption in finance over the next year? So maybe a little bit more about the future road map. So I think one is continuing to expand the ones that we just we just showed. I I think that there's so much more potential here that we just haven't touched yet. We started off with, like, the happy path or, you know, base use case that was kind of easy, and now we can keep branching out into more complex things, which I think is definitely just the next immediate step. And then after that, as I mentioned before, it is GL accounting. I think that as LLMs continue to progress, GL accounting is going to be the next hot spot for automation, and I really do envision a place where we can get down to, honestly, a two day close. Okay. Awesome. And then I think there's one more in here. Okay. Yeah. This is the last one. We've heard a lot about agenda AI lately. From your perspective, what kinds of what types of finance workflows, AP, AR, or otherwise, stand to benefit most from this kind of automation over the next year? I mean, I think a big reason why we focused on AR and AP for this is because it is one of the easiest places to start with AI. So I do think that as most people I mean, I know we had the poll earlier and people seem pretty advanced on this call. Most people that I talk to within the finance community, they're essentially behind on this that, you know, once again, they they believe agentic application based I that the AI that they're using and in some cases that is truly agentic. In other cases, it quite isn't there. But getting those full end to end process workflows within AP and AR is very obtainable, with AI in its current state now. A lot of work to get there, but it's possible. And I think it's gonna get easier and easier as the models continue to progress. So I think that that will be the focus for most people within the coming year. Cool. And I think that that is all of the question. Oh, no. There's one more. Okay. Last one. This is more high level. When it comes to adopting Adjenic AI in finance, what do you see as the biggest challenge? Is it building the business use case or business case, ensuring data data quality, or driving user adoption across teams. So I actually answered that one on the fly, in the middle. So I got that one. I just jumped on it. Okay. Alright. Yeah. You you must have missed that, but okay. It seems like we got to all of the questions then. So we'll give you another 10. And if not, it's been awesome. Thank you again for joining. We really appreciate the ability to present, and, hopefully, you found this worthwhile and learned, learned from it. We're always happy to answer any questions that you may have. So if at any point in, the discussion you wanna learn more about a presenter's slides or what they presented, you can reach out to us. It's firstname.lastname@siligo.com, and we're happy to get back to you. I also believe that Alexia is sharing all of the slides afterwards as well. So, yeah, no more questions. Thanks again, everyone. We're gonna end the webinar. Thank you.