Raw capture — raw/retail-2026-from-ai-pilots-to-execution-ivendnext.md

url: https://gist.github.com/njt/43376e9c89569a1c2f67fd12ca1d6467 date_fetched: 2026-07-05 backfilled: true

You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert

The $1.77 trillion retail loss is a data-decision problem, not a supply-chain problem. Stock-outs and overstocks, not competition or downturns, caused this loss in 2024. AI is built to solve exactly this.

89% of retailers are already using or testing AI, but 80% of AI projects fail to deliver expected value. The failure is not the technology—it’s fragmented data, integration challenges, and the inability to act on AI’s suggestions. It’s an execution gap rooted in operational architecture.

AI needs a unified data foundation, not a patchwork of best-of-breed systems. When AI reads from four or five different systems, insights contradict each other, teams lose confidence, and projects stall. One consistent data pool—one version of truth—is the prerequisite.

The baseline for AI in retail is four things: unified data, real-time visibility, at least 6–12 months of clean transactional history, and actionable outputs that connect directly to workflows (e.g., a reorder recommendation becomes a purchase order, not a slide deck suggestion).

iVendNext positions itself as an integrated platform, not a POS with add-ons or an ERP with a retail skin. Retail operations, financial control, CRM, and AI all share one data model. No reconciliation, no exports, no middleware.

Three AI capabilities were demoed:

AI Inventory Forecasting: Classifies items as fast/slow/non-moving, predicts consumption 30–90 days out, calculates reorder levels with confidence scores, and feeds the same data into revenue and cash-flow forecasts.

AI Assistant: Natural-language queries via an MCP server connected to Claude Desktop. It contextualizes answers, provides analysis and suggestions, and can generate complex HTML dashboards in minutes.

AI Copilot: In-app assistant that answers “how do I” questions and quick data queries (e.g., top-selling items), reducing support tickets and adoption friction.

The window between piloting and execution is closing. Early executors see 88% ROI and 55% higher operational efficiency. Retailers who build the foundation now gain 12–18 months of additional data for model refinement and operational confidence—a compounding advantage that’s hard to catch up.

Pithy and provocative quotes

“$1.77 trillion is what global retailers lost in 2024. Not to competition, not to a downturn, but to simple stock outs and overstocks. … This is not a supply chain problem, it’s a data decision problem.”

“80% of AI projects fail to deliver their expected value. Not because the technology doesn’t work. It’s because of fragmented data integration challenges and the inability to act on what AI suggests and surfaces. … It’s more of an architecture failure, not necessarily a technology failure.”

“When you train AI on such an environment, you get insights that contradict each other, your team loses confidence, the project stalls and eventually we call it a technology failure. It’s more of an architecture failure.”

“You can now move from being reactive to being more proactive. Now, you know, 30 days before the shelf actually goes empty and not the day when the customer tells you that the store stock is not there.”

“The window between piloting and execution is not going to stay open forever. … It’s the compounding returns argument. … Retailers who build the foundation now will have 12 to 18 months more of the data for model refinement, for operational confidence and … to be ahead, one step ahead.”

“Imagine all this coming to you in about five minutes, maybe less. … All this happens using ivindext AI Assistant.”

Tools, practices, and methodologies

AI Inventory Forecasting (iVendNext module): Uses the ARIMA model (and optionally Meta’s Prophet or classical regression) to analyze historical sales/stock movement, classify items by movement type, predict consumption for 30–90 days, and suggest reorder quantities with confidence scores. The same data automatically feeds revenue and cash-flow forecasts. Why it matters: Moves inventory management from reactive to proactive; the speaker says it can reduce stock-outs by 15–30% and inventory carrying costs by 20%.

AI Assistant (iVendNext MCP server + Claude Desktop): Connects to live iVendNext data via an MCP server, allowing natural-language queries in any LLM tool (ChatGPT, Claude, etc.). It returns contextualized answers, analysis, and suggestions, and can generate full HTML dashboards with KPIs, trends, and basket analysis. Why it matters: Democratizes data access—any manager can get complex insights without a data background; the speaker demonstrated generating a multi-chart dashboard in minutes.

AI Copilot (in-app RAG implementation): Embedded chatbot that answers operational “how do I” questions (e.g., “how to create a purchase order”) by pulling from documentation, and answers quick data questions (e.g., “top five selling items this month”). Why it matters: Reduces support tickets and adoption friction for new users; the speaker frames it as a way to make the system “sticky” and build user confidence.

Unified data foundation practice: One platform where retail ops, financials, CRM, and AI share a single data model. No reconciliation, no exports, no integration middleware. Why it matters: The speaker argues this is the only way to avoid the 80% AI failure rate; it ensures AI reads the same inventory positions, sales history, and financial data that operational teams see in real time.

FSN analysis (Fast, Slow, Non-moving): A prerequisite step before dynamic pricing. Use AI forecasting to classify inventory, set reorder levels and replenishment policies (purchase or stock transfer). Why it matters: The speaker advises retailers to master this first, as it’s the harder foundational step; dynamic pricing becomes easier once stock management is AI-driven.

N8N node for iVendNext (mentioned, not demoed): Allows connecting iVendNext to any automation workflow or third-party system. Why it matters: Acts as a bridge for custom integrations without heavy middleware.

AI Agents (mentioned, not demoed): Autonomous agents that execute multi-step retail tasks on your behalf, not just answer questions. Why it matters: The most advanced capability; moves from insight to automated execution.

Obvious next steps and questions

Assess your own data foundation: Do you have unified data, real-time visibility, and at least 6–12 months of clean transactional history? If not, that’s the first investment before any AI pilot.

Request a personalized demo from iVendNext built around your industry, store count, and pain points—not a generic walkthrough.

Have a direct conversation about your current tech stack and where the gaps are, especially if you’re running multiple systems that don’t share a single data model.

Questions the talk raises but doesn’t fully answer:

How does the system handle poor data quality, missing historical data, or data from acquisitions? The speaker says “no training required,” but what about cleaning and normalizing?

What is the actual implementation effort to migrate from a fragmented stack to a single platform? The talk glosses over change management, data migration, and downtime.

How does the AI forecasting account for external factors (seasonality, promotions, economic shifts, competitor actions) beyond historical patterns? The ARIMA model was mentioned, but no discussion of exogenous variables.

What about integration with ERPs other than SAP Business One? The speaker said S/4HANA integration is “in progress” but waiting for SAP’s integration platform to settle. What’s the timeline and depth?

How does the AI Assistant ensure accuracy and avoid hallucination when generating analysis? The speaker said it depends on the LLM’s quality, but no guardrails were described.

What are the specific licensing costs? The speaker described a model based on stores, POS terminals, and back-office users, with AI modules as optional add-ons, but no dollar figures.

Significant omissions

The messy reality of data unification: The talk assumes retailers can simply adopt a single platform, but doesn’t address the pain of migrating from legacy systems, cleaning years of inconsistent data, or the organizational resistance to replacing familiar tools.

Human factors and change management: AI recommendations are useless if store managers don’t trust or act on them. The Copilot addresses “how do I” questions, but there’s no discussion of training staff to interpret AI forecasts, override bad suggestions, or handle exceptions.

Limitations of AI forecasting: No mention of black swan events, sudden demand shifts (e.g., viral trends, supply disruptions), or the model’s inability to predict promotions you haven’t run before. The confidence scores were shown, but what happens when confidence is low?

Competitive landscape and lock-in: The talk is a product pitch; it doesn’t compare iVendNext to other unified commerce platforms (e.g., Shopify, Lightspeed, Oracle Retail) or discuss the risks of betting on a single vendor for the entire retail stack.

Cost and ROI details: The 300% ROI in 18 months and 88% ROI for early executors are cited from unnamed studies. No breakdown of total cost of ownership, implementation services, or how long it takes to see the forecast accuracy jump from 75% to 90%.

Dynamic pricing: Raised by an audience question, but the speaker deferred it entirely, saying retailers should first fix inventory management. The roadmap and vision for AI-driven pricing were not shared, despite it being a hot topic.

Data privacy, security, and compliance: No mention of how customer data is handled, especially when using external LLMs like Claude or ChatGPT via the MCP server. Is data sent off-platform? What about GDPR or regional regulations?

Scalability beyond mid-market: The speaker defined the sweet spot as 10–100+ stores. How does the platform perform for very large enterprises with thousands of SKUs and high transaction volumes? Not addressed.

Offline capability: Mentioned as coming in “two weeks, maybe three” for the POS, but no clarity on whether AI features work offline or rely on cloud connectivity.

Welcome everyone and thank you for being here. My name is Manish. I'm the Global Vice President for Product Marketing at City Access iBanext. And over the next 45 minutes I want to have a very honest conversation about what AI could do to retail one day, but what it is currently doing right now in live retail environments. And I also want to talk about the operational foundation that is actually needed and required to deliver the promise of AI. I have divided this entire session into two halves. First part, I will be taking you through a few slides talking about a few numbers that I just showed you in the second half, towards the end, I will actually show you how Aiven Next is able to deliver the AI promise to retailers today. $1.77 trillion is what global retailers lost in 2024. Not to competition, not to a downturn, but to simple stock outs and overstocks. All combined warehouses holding six months of stock that had to eventually be marked down at a loss. This is not a supply chain problem, it's a data decision problem. And that's exactly the problem that AI in retail is built to solve. The good news is that it is already solving it, but it's solving it for retailers who have stopped piloting and have started executing it. And here is the landscape. 89% of retailers are already using or testing AI. That means we are well past the early adopter phase. The market is 18 and a half billion this year, projected to reach 130 billion by 2033. US alone intends spending about $113 billion in technology in 26. But amongst all this, there is one number that I want you to hold on to. 80% of AI projects fail to deliver their expected value. Not because of the technology doesn't work. It's because of fragmented data integration challenges and the inability to act on what AI suggests and surfaces. And that's more of an execution gap. And it's not a technology problem, it's more of an operational architecture kind of a problem. And that's what we are going to discuss at length as you move along in the presentation today. When the foundation is right, the outcomes are not marginal, they are transformational. Have a look at this slide and the numbers here. 15 to 30% reduction in stock outs. 20% reduction in inventory carrying cost. 4 forecast accuracy jumping from 75 to close to 90%. Inventory turnover increasing by 50%. Add a 300% ROI within about 18 months. Walmart alone is the headline case. 24% growth and 30% stock out reduction from AI powered forecasting alone. Another study suggests across 30 retail firms that the ROI is actually proven. The larger question is that does the retailer have the foundation to access all this? And this is the core of today's presentation. The reason 80% AI projects fail today is not because of the model that they're using, it's what the model they are working with. A typical retail tech stack is a collection of multiple system, a retail system, a point of sale system, an inventor system, CRM and erp. And all of them hold, they're all best in class solution themselves individually and they hold that kind of transaction data. But what is happening is that there isn't one consistent data pool from where they can refer to. So even if you were to put an AI on top of it, AI would be reading data from four or five different systems and trying to collate them together. And when you train AI on such an environment, you get insights that contradict each other, your team loses confidence, the project stalls and eventually we call it a technology failure. It's more of an architecture failure, not necessarily a technology failure. We've all seen that, we've all experienced it in different ways, that the technology is not the problem, it's the way you are using it and the way you are stacking things up is the larger problem. And before we move forward, I want to be precise about what every AI model needs to work with in a retail environment. First and foremost is unified data. One version of truth, not four system, each with their own numbers, real time visibility, the last night's match up, load or align positions every other day, transactional depth, a minimum of 6 to 12 months of clean history. Which is what gives models like advanced machine learning models like arima, ari, ma, the pattern recognition that they need to do the forecasting. And we need actionable outputs, insights that connect directly to the workflows. So a reorder recommendation becomes a purchase order and not a suggestion in a slide deck. That's the baseline that we are aiming for. Once you have this kind of a baseline, that's when you would be in a position to implement AI. And not only just implement, get the results out of AI. As I said, we'll see how it goes. And that's where I want to introduce to you aiven next, an integrated retail management platform. And I want to emphasize on two phrases here. One is the word integrated and the second is the word platform. The application is not a point of sale system. It's not a system with a bunch of add ons. It's not an ERP with a retail skin Bolted on top of it. It's none of that. It's a platform where your retail operations, your financial control and all your AI capabilities share one data foundation, one data model. One source of truth. When ivindnext AI looks at the demand forecast it is reading the same inventory positions your warehouse manager sees in real time. And we'll see this as I'm again repeating is the same sales history your finance team is using for reporting. There is no reconciliation step. There is no export, there's no integration middleware. There is just one platform. One truth. That's what that is the promise of Ivan Next and that's what Ivan Next is all about. Here's what runs on these on this unified foundation. Retail operations, multi store, omnichannel, inventory promotion, loyalty. So I'm looking at retailers who fall in this category. Financial control A full built in ERP which means you're buying decisions and your financial positions are always in sync. A CRM customer data that lives in the same system as a sales history. AI forecasting Demand, inventory, revenue and cash flow projections and predictions draw from the same actual data. The AI assistant. The interesting bit where you access your data using natural language queries. You just imagine a kind of a chatbot which is giving you meaningful data out of meaningful analysis out of your own data. And then AI Copilot which is more an operational tool which is used by. Which will be used by the line managers and people at the leaf level who are, you know, running the system. There are in aiven Next there are about five different AI capabilities or modules that they call them. The first and foremost is the forecasting piece. I call it the. It's called the AI Inventory forecasting and it not only does inventory forecasting but also financial, cash flow, revenue, expense kind of forecasting. We will see that the second is ivindext AI Assistant which is more of a natural language querying system essentially using ivindext has an MCP server which is connected to which is used using Claude Desktop in our case but it could be used by ChatGPT as well. Any LLM tool would be able to connect. Ivindext AI Assistant is more of a, as I said, a natural language querying tool. Then we have an AI Copilot which is, as I said will be used by the operational people. There are two more that we have. One is the N8N node for iwind next. What it does is it allows you to connect I win next to any automation, workflow or any third party system. It's more of A bridge between I win next and any third party solution. And lastly is I wind next AI agents. So imagine that's the most advanced capability that we have. Autonomous agents that can execute multi step retail task on your behalf. Not just answers question but actually do the execution. In today's webinar we are going to talk about only three of them. Not for any other reason except for the fact that you know, constraints of the time. We'll talk about Aiven next AI forecasting. We'll talk about Ivan next AI assistant. We'll talk about i1 next AI copilot. So that's the three capabilities. The total five capabilities were the three that I'm going to to talk about today. The first and foremost is the foresight, the crystal ball, right? And the first skip as I said is the forecasting. And this is where the margin impact is most. Direct AI forecasting engine analyzes your historical and sales stock movement data. Every item in your item master catalog gets classified into fast, slow or non moving items or critical for that matter. The system then predicts consumption for the next 30 to 90 days depending upon the horizon that you want to have, calculates the reorder level and suggests a suggested order quantity. And this suggestion has an element of a confidence score on every prediction. So how confident is the forecast when it is giving you any kind of a prediction? The same data then feeds into your revenue forecast and your cash flow projection. So you're not looking at inventory in one screen and revenue forecasting in the other. It's all the same data depending upon who's looking at it. The operational shift is absolutely fundamental. You can now move from being reactive to being more proactive. Now, you know, 30 days before the shelf actually goes empty and not the day when the customer tells you that the store stock is not there. That's the first capability. The second capability is decision intelligence. It's ivindnext AI assistant. It connects to a live data through an MCP server and answers questions in plain English. So any manager, not just someone from a data background, can ask a question and get an answer from your actual transaction history, your livestock positions and your customer data. The system doesn't run, just return a number, it contextualizes it and returns not only the results, but suggestions and analysis also. And that's a great the bigger shift. So what that means is you don't have to be, you know, how do I say, experienced enough to read through the report. You don't have to to analyze the numbers and contextualize them. There is already Someone who's done that for you, you just have to act on the recommendation or not act on that for that matter. But you have someone who has done the analysis on your behalf. The third capability that I'm going to talk about today in today's case is simple AI Copilot. This is an in app copilot and that does a couple of things. It answers quick questions, for example, my top five items in the last month or in this month. And more importantly, it allows you to ask questions like how do I create a purchase order? So all those how do I kind of questions are answered in almost real time, which ensures that the entire support ticketing is automatically taken care of. Simple stuff is already handled by the system. What it does is it reduces the friction that quietly kills adoption of a new system. And this friction kind of compounds the team that understand the platform are the ones who use AI more confidently. So this is a very interesting tool that not only gives contextual answers but also is once, should I say, it helps a new user to start adopting the system that will go faster. Just summarizing this, this is just a summarization of what I spoke about. The three capabilities, the foresight or the forecasting, business intelligence and Copilot. The problem it solves is, you know, stockouts is forecasting insights buried in reports is from business intelligence and quick how do I questions using the AI copilot. That's the play as far as the three AI capabilities are concerned. What I'm going to do now is I'm going to actually show you the application and take it through three different should I say use cases or capabilities and then open the forum for any questions that you may have. Alright, so I'm starting with. So what you see on the screen is my inventory forecast. What I've done is for the purpose of this presentation is I've selected a few products so that's easy for me to walk you through. And it is visually that much more palatable. What the system does is the entire AI forecast. What it does is it goes back in time at least 90 days and then based on the consumption pattern, it then gives you two things. One, it tells you the movement type, the fast move, whether it's been fast moving, slow moving or non moving. There are statistical algorithms written which determine what is fast and slow and non moving. What it also does is it gives you the predicted consumption in the next 30 days. It says based on the historical data you will be consuming 28 of these 31 and so on and so forth. If I click any one of these, okay, I get into the details of that particular product. It says the movement type is fast moving. The prediction confidence 85%. I did it a while ago. And if it keeps scrolling down, this is where the best part of the forecasting is. It says a basic forecast analysis. It tells you historical period, last 30, 90 days, total number of transactions, consumption events, the consumption pattern in so many units a day, average per transaction forecasted result is so many units within 85% confidence, inventory recommendation, current stock reorder levels, etc. It gives you all this information. Now imagine a situation where you have this information 30 days in advance and if the lead time to get to procure the product, let's say about 20 days, you have enough and more time on hand to try and order it in time, receive it and put it back on the shelf. All of a sudden what will happen is the entire inventory holding cost of the entire organization will come significantly down, if nothing else, at least by 15, 20%. And that's a significant number in terms of the kind of investment that a retailer has to do. That's as far as inventory forecast is concerned. And it keep and it, you know, for example, this one has been slow moving. If I go here, gives me similar analysis, okay, at the bottom and tells me that this is what has been the consumption per day and what should be the stock levels, etc. So the, the entire forecasting is. You can do the forecasting by a product and a store combination or for individual stores that you can do by product and at the organization level also normally forecasting and predictably is done by product and warehouse combination. Now running the risk of converting this into a training session. Let me very fast, very quickly move to the second part where I talk about the financial forecasting. That's where it gets very interesting. So imagine this is my inventory forecasting. Someone did this, okay? And certain someone is doing my is doing his sales forecasting. So I could do multiple different forecasts. I can add multiple different forecasts for different kind of accounts, sales, stocks, expenses, etc. So let me pick up the first one. What it is doing here is something very interesting. First and the foremost, it is using the most advanced machine learning algorithm model called the ARIMA model. So the way this really works is it goes back in time, picks up a pattern based on the sales history that you have and then uses that pattern and extrapolates that for the next, for the next 90 days. Okay, so that's a classical machine learning. Should I say use case that we have in this case. What I've done is I've said this is my company. I want to have a forecast by revenue. This is my revenue account. This is the 90 days period that I have. Okay, I'm asking you to do a 70% confidence level. What it does is it gives me a predicted revenue forecast of X number of dollars with a 61% confidence. Now this number, okay, is it's a great number for someone who is doing revenue forecasting. What it had also done is that it has auto automatically synced the inventory forecasts. Okay. And here you can see that my integration is inventorsync enabled. The point I was trying to make here is that the guys who are doing sales forecasting don't have to bother about inventory forecasting. Someone has already done that. The sales forecast just consumes it. And that is again a very big shift from your forecasting standpoint. Yes, it's a crystal ball and you can look into the crystal ball. But what is extremely important is that your foundation has to be very clear and very solid. You should have systems which are able to talk to each other. Okay, When I say talk to each other, I mean the data model has to be exactly the same so that you don't have to join wires or do integration pipelines and things like that. So that's as far as the forecasting bit is concerned. You can see here at the bottom right, a small little copilot chatbot. We're enhancing it and you'll see a very different user interface in the coming days. But this is where life becomes that much more easy for this person who is using. Let's say I were to be on a, for sake of an argument, purchase order list. This is my awesome bar. It's like really, really awesome. I am on my purchase order list and I don't really know how to do that. I could type how to create a. So the moment I do that, what the system does is it understands the question, contextualizes it, goes into my standard documentation repository, gives me an overview as to how this has to be done and also gives me a link which I click and takes me to the official documentation site that makes life that much more convenient and easy for someone who is just started using the system and rather than asking around, he can actually read through the document as to how this is typically done. Let me ask another question which have been my top five selling items in April 2026. Now this is for a. A quick answer. What you will get is a. Is a. Is a list of four or five products with their. The sales numbers. Let's wait for it to come up. It takes a while, but there you are. So it gives me. These are my. These are the products which have sold really fast in the. In the previous month. And if you notice, okay, this has been classified as a fast moving item here and if you. My inventory forecast also classified this as a fast moving item item. The reason why this is happening is, is that everything is running on the same data pool. You don't have to be going around the data that comes in one from one place is pretty much the same as what comes from the other. And more importantly, you have some suggested, should I say next steps. This ensures that you know that element of stickiness to the. To the user as to, you know, you. I'm prompting him to start asking another question. What are the total sales for each item? Which item and the highest increase sales and so on and so forth. So this is the second level of should I say AI that we have in the application a copilot which actually runs along with you don't have to switch applications or tabs. The third one is the i1. I have an AI assistant. What it is essentially doing is as I said, it's an MCP server and I'm able to do some contextual analysis. So what I've done is rather than trying to run it right in runtime, simple questions top five selling items in the alpha store and this will get the same answer. I could do something more complex. For example, I said what is my average transaction value and upt values in units per transaction in 26. So what it does is it just analyzes that so many transactions and then it gives you the Observations Delta Store ATV the store manages 29% higher than Alphas and UPT. Both are virtually identical Transaction mix Alpha consists of so many transactions but so much in debt. This information, this information is really, really critical. Getting, I mean for starters getting ATV and UPT themselves is a very complex process. But having done that, even if you were to generate this to able to read and understand this data and contextualize into a meaningful action item is not that easy. Okay. I then went a step further and I said okay, create a dashboard with all important insights and KPIs for this calendar year starting the 1st of January running generate output inline in the chat as well as HTML and a light theme. So what it does, it just goes and does the analysis. I'm not doing it runtime. I just did it moments ago just to avoid that Awkward pause in the conversation. So what it does, it does KPIs and all that and then I have the option of opening in Google Chrome. This is what it looks like. So imagine all this coming to you in about five minutes, maybe less. I get total revenue, monthly run rate, total transactions, average transaction value, best month, alpha store, delta, all the stores, revenue by month store performance, okay, top categories, category, category momentum and so on and so forth. It gives me transaction distribution, the basket size analysis, everything and then it gives the entire basket analysis, then the different stock values, key insights and signals. All this, all this happens using ivindext AI Assistant. So it could be your question could be as complex or the output could be as complex as this or it could be a very simple question that you're asking on maybe an ATV or UPD or for any kind of promotional analysis that you want to do what has been my promotions, etc. So that's as far as the third capability is concerned. Let me go back to the presentation. So why that was really important. It was important for me to show this to you that there is an application which is actually doing it all single place. The data on the timing is very clear. 4700% year on year growth in AI driven retail traffic forecasting alone is approaching 28% of the entire AI retail market by the year end. Early executors adopting 88% ROI rate and 55% higher operational efficiency. The window between piloting and execution is not going to stay open forever. And this is not a fear driven messaging. That's not what I'm trying to do. What I'm trying to communicate here is that the, it's the compounding returns argument. I mean I'm talking about how you can compound the returns. Small steps can add more values. Okay? Retailers who build the foundation now will have 12 to 18 months more of the data for model refinement, for operational confidence and you know, just to be ahead, one step ahead. And that's the advantage that a retailer would have and that's advantage that it's very difficult to fill the gap for the people who are, for the retailers who are staying behind. As I said, this is not a fear driven message. It's just, and it's not even about fomo. It's all about, you know, realistically thinking that okay, this is time that, you know, it's moved from a pilot stage from a let's do it kind of a stage to actually getting it done. This slide is important, which talks about what is the kind of Retailer that we are looking at that Ivan retailer is a good fit for which kind of retailers? Retailers with multiple stores or channels who are spending meaningful time reconciling data. We're talking about retailers which have inventory accuracy problems, which is based on the 1.77 trillion figure that I showed at the beginning of my presentation. We're talking about retailers who have outgrown their current system looking for a platform that grows with them, but specifically looking for retailers who are not looking to run another AI pilot. They're looking for a production grade foundation where AI delivers the outcome that it promises. See the icons on the right side of the slide? Fashion, apparel, footwear, sports, electronic pharmacies. All these retailers are the kind of retailers who need this AI engine. If I were to again contextualize this in terms of number of stores, we're looking at retailers who are at least 10, 15 stores on this side to about 100 plus on the other. The classical mid market. That's the place where the margin is squeezed the most. That's the place where you would like to fit in and kind of position the value that Ivan next AI brings onto the table so much from my side. As a next step, what I would recommend is that there are three paths forward. You could do a personalized demo, not a generic walkthrough, but a session built around your industry or store count and your specific pain points. You could have a direct conversation with our team about your current tech stack and where the gaps are, or explore at your own pace through our documentation and case studies. A link, a number on the screen. I'm happy to take questions. Let me go here. That's the email that you're looking at. Connectedex.com feel free to write. I open the forum for any questions that you may have. So let me try and take questions one by one. I have few questions here. Question there's a lot of value here, but my concern is with in context answers, how to ensure accuracy and the historical inspection tool you mentioned that was trained generally or customized. So very valid question. Now when we talk about something which is in context. So the way this really works is so you have a single pool of data and I'm talking specifically in the context of let's say artificial context of forecasting. So what we're trying to do is when you try and create a forecast, you will tell the forecasting engine the model that you want to use. Most likely Arima. It could be others as well. It could be the profit that comes from Meta. It could be the classical Regression we are forecasting, but let's say you select a model, then you select what is it that you want to forecast on. So let's. You want to forecast about inventory. So when you want to forecast about inventory, you select the products on which you want to do forecasting. What the system does. Once you've given those parameters and the boundaries, what the system does is it then collates the data from the core underlying data of iVill. Next, imagine creates a data set first, does a kind of a pattern recognition and then extrapolates that pattern to the rest of the data that has been collected. That's the classical machine learning part of it. The other is the regression way of doing it, where you collect entire data and you read through each and every invoice or a transaction and then create a. A pattern out of it. There are two ways of doing it. So to answer your question, there is no training required as such for any kind of forecasting. What you need is more the data, the better it would be. That's what I was suggesting. At least 12 to 18 months of historical data is really required. That's as far as the forecasting is concerned. The second part is when we talk about Ivan Next AI Assistant, essentially the SAP server of Ivan Next. So the way that really works is that MCP server, you have to train anything. There's no training. Imagine you just switch it on. You just switch on the MCP server. It already has the ability to read through the data. And there are two parts to this. One is ability to read the data and the second part is ability to interpret the data. The interpolation of data is a factor of how good your LLM is. It could be ChatGPT or a Claude or anthropic Claude, depending upon which one that you're using. Both of them are equally good. We are using CLAUDE again for sake of convenience, but it is not limited to that alone. So no training of data as such is required. The third that I spoke about was the AIVEN Next AI copilot. That again is more of a RAG implementation where you don't really have to train it. It's more of a natural language querying. It understands the underlying data and it's the ability of the LLM which is running at the back end. It could be an OpenAI API or the cloud API that you'll be using to kind of get the data out of the system. So no training assets required. What certainly is required is the amount of raw data I would say that you would typically need. And I Have a few more questions. Tony wants to know is ivan Next Integrated SEO Business 1? Is it a cloud based solution? Short answer yes, it is integrated with SAP Business One. I WinNext is a classical SaaS based cloud application. The entire thing runs off a browser. And yes there is an integration with SAP Business one that we've already developed. Then I have another question. Can the system be accessed either offline and online? Okay, so let me rephrase the question. Currently the entire thing is all web based and so is the point of sale. But in the next about three weeks or so we are going to release the offline boss as well. That means while it is all browser based you would still be able to use the point of sale which kind of looks like this. Okay, you're able to see my screen. I presume this is again all web based has all the good stuff. Again I don't want to be talking about a point of sale right now here, but has all the bells and whistles fully customizable the layouts, the buttons, what the buttons perform has within a dashboard within the point of sale which gives you information like how long it has been going, some charts by tender, collection by group, etc. So this is currently all online but as I said another two weeks, maybe three at the max we would have an offline capability also available in the application. Let's move on to the next one. Great presentation. Thank you. I would like to hear more about your AI tool and dynamic pricing. Dynamic pricing powered by AI is becoming a major competitive advantage in retail especially as organization look to optimize sell through inventory, aging, margin performance and customer demand in real time. What is Ivy next vision and roadmap regarding AI driven dynamic pricing integration within its retail ecosystem? Yes, it is the latest trend in terms of dynamic pricing as to based on historical sales, based on the promotional data that the promotion that you've run and based on the data that comes from the promotions, you should be able to dynamically price theoretically possible. Okay, but what we are suggesting to all our retailers is that rather than going there first, what is important is to try and manage the stock internally. That means do an FSN analysis first. Fast moving, slow moving, non moving analysis, do an FSN analysis Using that determine your reorder levels and the reorder quantities and the replenishment policy. It could be a purchase or could be stock transfer. If we are able to take that first step which iVindext allows retailers to do today. If that first step is taken the next logical step and far easier step would be to have dynamic pricing. The first one is the slightly difficult bit where you're talking about historical sales trends, determining what should be my reorder levels and the reorder quantities and my replenishment plans, whether it should be fresh procurement or should be stock transfer from the neighboring store or the other stores. But it's a valid point. Thank you so much for bringing it up. This is certainly the place where the world is moving, but I have Ahmed asking Thanks for the presentation. I have a question which is if aiven Next should work as a standalone solution or would it be integrated like SAP Business one and if it is able to integrate, is it fully integrated or some topics cover the integration all right, so I Winnext is a full ERP in itself. It has its own P and l balance sheet, etc. It has all those ERP capabilities. What we've also done is we integrated it with very popular SAP Business one as the erp. So when we talk about that integration, the integration is really deep. People who come from the i1 GTEL background, they would know how deep is that integration. But to give you the answer right now, masters like Items, prices, taxes, etc. All these masters come from SAP business one into Ivan next transactions. Go from i1 next into SAP business one. When we talk about transaction, I'm talking about normal sales transaction, refund transactions, special orders as in orders in the name of customers, fulfillment plans. All these are transferred from iwindext into SAP Business 1. What is also what can also be, what is also transferable is purchase orders created in iVindex can actually become purchase orders created in SAP Business 1. So imagine you run this intelligent engine here, get a recommendation, execute that recommendation. Let's say create a purchase order. You would be able to integrate it all the way back into Business one. Do the procurement, get it back into Ivan Next. Similarly you could do the other way around. You could do the entire central procurement in SAP Business 1 and receive the goods in Ivan next in each and every store. So the depth is fairly deep and the integration is fairly deep. When it comes to SAP Business one integration and let me know, get in touch with us should there be need. We can always have a detailed demo session on the capabilities the integration capabilities of HiberNext into SAP Business 1, though that is slightly counterintuitive. Now one is the SAP Business 1 part of it, the other is the bigger SAP like we're talking about S4 Hana for retail and the others. Okay, we are in the process of developing those integrations. We would eventually have those integrations developed and be made available to the market. But certainly we're just waiting on the dust to settle down in terms of the integration layers. That SAP integration platform, that SAP has quite a few of them. They are also in the process of transition. We're just waiting for that to settle down before we actually create that that integration. But as a proof point, yes, v1 integration exists. That's the easy part. The other integrations we have tremendous amount of experience having integrated our legacy application Ivan GTL into S4 HANA for retail is retail ECC6 Sage X3 Sage 300B1 Dynamics Business Central. So there is a huge amount of legacy and experience that we bring onto the table of integrating into these applications. Integrating aiven next would be much easier. Okay then Hank would like to know thank you for presentation. I would like to better understand partner enablement, regional pricing and a support structure South African region Happy to take this discussion offline with the sales department. Certainly Hank. We have Jan who has recently joined us, so he would be most appropriate person to get in touch. Okay, send an email to connectwildnext.com and I will do the introductions. Possibly you might already be knowing him, but he would be the most appropriate person. Then I have a question on what billing model is used? Rate per use or one off cost? Okay, so Brian, the licensing of the application is fairly simple. All right. We are licensing based on two parameters, actually three. One is one is the core offering of Aivendex, which has financials, buying, selling, stock, etc. Retail. That's the base product. The licensing is based on number of stores and the number of point of sale terminals. So that's the first level of licensing. And the second parameter in licensing is number of back office users because there will be users who will be using the office of the backend, which is the desk. And there'll be point of sale users. So we count both the users, back office users and the point of sale users. And there is a dollar value attached to the optional module. The base product comes with the base product which is buying, selling, accounting and retail. But there are optional modules like loyalty, WhatsApp integration, CRM manufacturing, which will be releasing all these very shortly in the next about two to three weeks. But these are all optional modules. AI is again all the five options of AI that I mentioned, be it the forecasting, AI assistant, be it copilot, N8 and nodes and the AI agents. These are again all optional modules depending upon how you want to use them. So that's as far as the licensing bit is concerned. I would recommend that you get in touch with your local CSM Jan in this case or possibly James from who's based in Nairobi in Kenya. He would be in a better position to explain that. Then Hadija wants to know Ivanx will be integrated with the bank and mobile payment systems. Yes, we are integrated with the few of them. Okay. Specifically in the context of. I don't really know which ones we are integrated with in Africa but apart from the payment processor integration we also have an items integration especially for. Specifically for Kenya but valid question. I'll have this answer the kind of payment processors that we are integrated with in Africa. Then I have another question. Can I include the questions and answers in the items you will send us? Oh certainly, yes. This is the whole session is being recorded, all questions are being recorded. So you'll get to know and I think that's kind of brings me to the end of the Q and A session. And thank you so much for making this session so interactive. I'm very thankful to all of you. Take taking the time out and being present. So so much for now. As I said, that's the email id that you're looking@connectivindex.com you can go to the website, fill up the form if you are. Most of you are already in touch with the local csm. Could be James or Yan or Walid and Ali. You can always get in touch with them and they'll be more than happy to answer. You will certainly can get in touch with me on this email ID So much for now and I will see you on the other side.