Hello everyone. I'm Victor Gabriel, a season supply chain consultant from GP Worldwide. I'm thrilled to be here speaking at Prompt Engineering 2025 conference. Today I'll be sharing a success story on how we have helped a global food manufacturer to uncover about $30 million in cost savings using a Power BI tool with the embedded prompt engineering framework. In the next few minutes, I'll walk you through the challenge that we had on hand and the approach we've taken and the solution framework we have developed, and perhaps we will learn how the solution can be used to implement, to solve similar business problems across other industries. Now with that, looking at the challenge, this particular client is a $15 billion food manufacturer with a huge number of contact manufacturers across North Americas. So like many large organizations today, they do struggled with scattered data across multiple systems, limited visibility into real-time cost information, slow identification of margin, leakages, and difficulty turning data into meaningful de decisions. Now, for them, Excel dashboards weren't enough. They needed something faster, smarter, and easier to use for their employees. Now what we've done was, as part of the solution, we have developed a hybrid analytics approach. Now, this framework was built using a power BI for structure and Microsoft copilot for intelligent prompts. Now, while Power BI gave reliable real-time dashboards, explaining or conveying the data, co-pilot added a prompt driven layer where users could simply ask questions with quick answers. This made analysis, instant decisions that once took hours. Now it just take few minutes or seconds. Diving a little bit into the platform architecture. The architecture rests on three key pillars, power BI Foundation. That's where the realtime dashboards are developed. Presenting the data facts, prompt engineering layer. This is where the scenario modeling is done through the Microsoft copilot. The integration engine, this is where connecting the data connects happens with a multiple across multiple manufacturing systems, supply portals, finance data to form a single source of truth. Now, together with all three pillars, it gives us a single source of growth and a rapid insight generation. Yeah, looking into the prompt engineering in action. So what it actually means. Analysts could talk directly to data. That's what it means. They can ask any question they may want, they can ask a question. Hey, you know why my scrap rate went down in scrap rates were high in Q3. Why? What factors have I actually contributed to the most to the scrap varis and Q3? And the system responds with visuals without within no time, and summarizes instantly. No manual analysis is needed. Prompt engineering made advanced analytics accessible to everyone, not just the data scientists. Now the approach, talking about the approach now that we have the framework here, is the approach for execution. The focus was was what was the focus was kept on two major levers, the margin performance and the cost optimization. Margin performance. That's where, the focus was on conversion cost. The overall yield profitability per product, and the cost optimization. That's where the supplier cost benchmarking was done. Packaging efficiencies the metrics were looked at and the production process improvement initiatives. Now, by targeting these, we could quickly see where the money was being lost and where the improvements can be made. Now talking about the timing now, usually for projects like this, timing is everything. This solution was launched during a major corporate restructuring, key insights, guided leadership decisions telling them which applies to consolidate, which to grow, with which to shift production, directly supported company's new operating model. And these are the results, the transformative impact. The impact was pretty clear. 30 million was achieved in savings across scrap and conversion costs. When I mentioned scrap it's coming from the raw material and packaging scrap from the pro, from the from the products and the conversion costs. Now that's number one. And faster decisions were made with more transparency and the supply performance were improved. And it also helped us with the real time tracking. And the platform paid for itself several times over. It, it was almost a good number of written over written on investment as one of the results. Now here is the real world application. Now, one of the major findings that we saw as the outcomes were the scrap rates were 15% above the industry benchmarks. That have contributed to losing a lot of money wasting millions. Now using the prompt driven insights, we found consistent product recipes across suppliers. Now remember we are talking about the food manufacturer and the product is a is a food. You can say it can be a candy or it can be an energy bar, or it can be any food I, any food item. So now think about. The product recipe now, making it streamlining the product recipes would help now, and that is not easy. So one of the things we were able to fix it by standardizing the formulas food formulas the renegotiating the supply contracts, adding the live performance alerts supply performance alerts, and within 18 months, the food scrap raw material and packaging scrap was at the industry. Best levels now that has given the the client about $30 million in savings. Now, what lessons have we learned? Just in general? In general, from executing this project? Always start the business questions. Start with the business questions, not with tools. What what the tools are important. But starting off with the business questions is primary is most important. And ensuring the data quality early on is important too. Poor data just, kills the good analytics. Work across work cross-functionally. When we say the data, the the, one of the pillars of the framework was the integration engine. It is important that we. We weed the data. We are able to integrate the data, transform the data that comes from it, that comes from finance, that comes from operations, that comes from supply portals. Now, working cross-functional with the cross-functional teams is quite important now, and using the AI wisely now it should enhance and not replace the human judgment. As we are using the copilot to support web, making some of the decisions, data driven decisions, it is not to replace entirely. But it just arguments enhances the human judgment. Those were the, some of the lessons we have learned from executing this project. From insight to action, what was the overall decision cycle that we have observed? Now, a quick reminder that, how insights are turned into action. First is the data integration. This is where the real time updates from all the sources are aggregated now. Then the visual discovery, visualize and spot trends in the dashboards. Then the prompt driven analysis, asking the questions, testing the scenarios with ai in this case, the co-pilot, but you can also use the L lms. Synthesizing the copilot generates the summaries for executives so that the data-driven decisions can help make some of the meaningful business decisions and the implementers obviously making the decisions and acting on them, launching the project. And launching the initiatives, the team are acting on it and monitoring those results. And we go back in the cycle to refine it. So this loops keep kept on the keep the continuously, it kept, it needs to be kept continuously improving from data to decision and decision to action. So that's it's a virtuous cycle now. The practical Brooklyn, if we have to get started on similar projects way to get started in your projects. If you are looking at how can we implement a kind of a similar framework to realize cost savings using a similar solution framework. First, identify three to five high impact areas, then build a simple power BI dashboard, then layer in a copilot gradually. So first is a Power BI dashboard. Make sure that you know the data you have the data and you're able to spot trends. Then on top of it goes of the co-pilot or any LLN and then track the insights that leads into actions. And the fifth one is the scale and standardize. Once it is proven, it's all about proving the value early and then expanding it fast. So that can be a successful blueprint in order to execute for similar projects. Now talking about the future, this is just the beginning of ai, augmented supply chain intelligence. Now next comes the predictive analytics. We just have the prompt engineering layer, and next comes the predictive analytics, automated recommendations and risk forecasting all blended. Blending the structured data with adaptive ai. So whether you are tracking the margins, erosion, or supplier optimization or network complexity, remember Power bi, it just gives a structure and prompt engineering, it gives us the agility. It just gives us that speed. Now together it'll give us a clarity of thought clarity of the business performance with the speed at the, at a good rate of speed. Now I really appreciate, again for this opportunity that was given to me to present this to all of you now, feel free to reach out to me in case you have any questions or to share simply your thoughts. Now, with that, I would like to say thank you for the opportunity once again to be a speaker here at Prompt Engineering 2025 conference. Thank you all.