Multi-Agent Supervision for Supply Chain Intelligence at Scale

Summary

Kohler built a supervisor agent on Databricks that routes cross-functional supply chain questions — covering demand, inventory, and planning — to specialized Genie spaces, enabling unified self-service analytics across previously fragmented data systems.
Belden partnered with Resilinc to build a multi-tier supply chain graph using entity resolution, relationship inference, and confidence scoring, then deployed task-specific agents to respond automatically to supplier disruption events in minutes instead of days.
Both implementations apply a sense-recommend-act pattern where agents detect events, score revenue impact, and operationalize decisions automatically, with performance measured through latency, token consumption, and accuracy monitoring.

Multi-Agent Supervision for Supply Chain Intelligence at Scale

Industrial enterprises manage complex supply chains across fragmented systems, losing days to manual analysis. Kohler and Belden demonstrate how Databricks enables multi-agent supervision and supply chain intelligence. Kohler built a supervisor agent routing cross-functional questions (demand, inventory, planning) to specialized Genie spaces with unified self-service analytics. Belden transformed supplier risk management by building a multi-tier supply chain graph using entity resolution, relationship inference, and confidence scoring, then deploying task-specific agents to automatically respond to disruption events.
Learn how to build supply chain lineage, govern multi-tier supplier data, measure agent performance (latency, tokens, accuracy), and operationalize decisions through agent-driven actions in minutes instead of days.

Chapters

FAQs

How did Kohler use Databricks to improve supply chain analytics?

Kohler built a supervisor agent that routes cross-functional questions about demand, inventory, and planning to specialized Genie spaces, replacing slow dashboard-based reporting with a conversational, self-service analytics approach. The system unifies data from fragmented sources and delivers answers in real time, with the next phase targeting autonomous actions and master data correction.

What is Belden's multi-tier supply chain graph and how was it built?

Belden worked with Resilinc to construct a multi-tier supply chain graph using entity resolution, relationship inference, and confidence scoring to map supplier relationships across multiple tiers. The graph is built on a medallion architecture — bronze, silver, and gold layers — on Databricks and powers risk and revenue exposure scoring for each supplier.

How do Belden's agents respond to supply chain disruption events?

Task-specific agents monitor the supply chain graph and automatically respond to disruption events using a sense-recommend-act pattern, compressing what previously took days of manual analysis into minutes. Decisions are informed by composite risk scores and revenue impact mapping so the recommended actions reflect real business consequences.

What metrics do Kohler and Belden use to measure their agents' performance?

Both companies track latency, token consumption, and query accuracy to measure how well their multi-agent systems are performing. Monitoring these metrics continuously allows teams to detect degradation and optimize the systems as data volumes and query complexity grow over time.

Full transcript

[00:08] Hello. Welcome. My name is Mitchell Calvert and I'm presenting today with Karthik Joshula. Um we'll walk you through the story of um our AI supply chain maturity journey. At Kohler, under a relatively new chief digital officer, we've accelerated the path into new technologies in the last year.
[00:24] So last year I met Karthik on a machine learning project where um in a really complex facility with huge bills of material um helped us determine uh lead times for products when we were out of out of stock. So in that in that time we got to know each other, talked to a
[00:40] lot about AI. And I asked him, you know, I said, "I've got a phone in my pocket that can tell me anything about the whole world. You know, I can ask it about breeds of dogs and different kinds of trees and everything about history. So if I've got a mountain of data, why can't I ask everything about my supply chain?"
[00:56] You know, and he said, "There's this new thing called Genie. Like let's give it a shot, right?" So we'll get into that in a minute. Um but I'll tell you a little bit about Kohler Company. So Kohler Company, as you may know, um kitchen and bath products. You've probably used one
[01:11] today. I've seen them. They're here in Moscone West. Um founded in 1873. We also have um a hospitality division, golf courses, um five diamond resorts where of course we show off all of our kitchen and bath products.
[01:26] Um got a number of different brands. We've also got saunas, cold plunge, everything. So um in supply chain it all starts with the forecast. What will the customer buy? Tell me about the trend of orders. Where are we going to distribute these
[01:42] products? How do we plan the network of supply? What about our production plan? Missing parts, procurement. Um and all of our master data. So we get back to the conversation last year with Karthik. And I'm telling him, you know, I want to
[01:59] surface insights, I want metrics, I want to talk to my data, you know, I I I want to do it in a flash, you know. So, at Kohler in the last 10 years, we've been a huge Power BI company, right? Go to the go to the workspace, go into the the
[02:15] report, go to the page, set about 15 filters, I've got my answer, right? We're we're really trying to pivot that onto the more conversational approach, right? Um so, our goal for this year is to transfer transform into quick insights
[02:30] and more dynamic reporting. So, we're also looking at BI Genie. And so, with the right foundation, we'll be ready for agentic action uh to eliminate the tedious work and opening up people into more strategic behaviors.
[02:47] Um So, in supply chain, you know, we've got a mountain of data. It's about as complex as as you'd expect. And in prior years, uh we kind of struggled with our data architecture. Anybody can put data into anywhere. We've got siloed data, we've got duplicate data. So, this year we're
[03:02] starting over with um with a new supply chain and operations data lake um where we're going to take care of that governance and um build it from the ground up. So, on top of that, we build our intelligence layer with with Unity Catalog, metrics views, Genies, and Agent Supervisor.
[03:19] And finally, we're exploring what's next with the application layer that goes beyond. Pass to Karthik. Thank you. Thank you, Mitch. Uh good morning, everyone. So, how did this journey started, right? Last year, prior to before before last year, we started
[03:35] modernizing our data platform ecosystem. So, we were able to build those common data model where we identified what are all the source systems, had a clear lineage, and we would start developing that common data model. I'll not get into that slide much because there's more for content.
[03:51] So, now once we have got the common data models spin up, we had leveraged Unity Catalog and thanks to Mitch and his team, they were proactive and they started owning, defining the metadata. They went into the Unity Catalog and they had provided all the descriptions and tagging for each and
[04:08] every attribute. And then we identified what are all the key metrics that that are needed that that drive supply chain. And with that, we also started putting more observability rules on the data so that we can continuously measure the data quality.
[04:23] This is the foundation for us to accelerate into the Genie development. But there's more exciting to come on on what we have done on top of Genies, which I will walk you through and Mitch will do a demo. But what we wanted
[04:39] like supply chain is so complex. If if you're all in supply chain, you know like every indicator, key metric is kind of a lagging or a leading in metric. We don't want the supply chain demand team to have have a question in in the demand Genie, then take the insight, go to the
[04:55] inventory Genie, talk to inventory Genie and then go to your planning. It's kind of spreading across like multiple Genies. We want to give a unified experience. That's exactly what was announced in the keynote, Genie 1. We developed a couple of months back called a supervisor
[05:12] agent, which is like a one unified experience any supply chain individual analyst, executive can come to the supervisor agent, ask a question spreading across multiple uh portfolios like demand, inventory, planning. The supervisor agent will
[05:29] route and spin up the correct Genies. So, this is an architecture diagram. This is like four verticals. The first one is user experience where we have defined the personas. We had spin up like multiple authentications and and accesses. The second vertical is where
[05:44] it gets interesting. This is where again I'll double click on the keynote that was called out on control and context. So here the tier one, if you see, that's the knowledge base we have developed, which is kind of the context. It will tell how to do the deterministic
[06:00] routing, and it will try to spin up the more accurate genies. And then the third third vertical is where the actual rag happens. It generates the query. It it hits the database, generates the insights, and converts to a conversation. natural language. And the fourth one is
[06:16] where we wanted to measure the traffic, measure how many tokens we are consuming, what is the latency, what is the feedback looking like, what is the overall sentiment of the users. We did kind of like the the control tower, where we are actively measuring the performance of the genies.
[06:32] Um this is uh I'll go back to the next slide. Before I hand it over to Mitch for the demo, I want to double-click on how we are measuring those performances, right? We had categorized all the key metrics into two buckets, one for IT, so that what we own, and second is for the
[06:47] business, for Mitch Mitch's team, where they understand the overall performance from a business side. So this is kind of key metrics where we would like to monitor like the total request, what is the average latency looking like, what is the error rate, and so on. And then we are actively measuring the
[07:03] tokens, so that if we want to like um understand where our cost uh is is burning, we again we want to convert our AI initiatives from a cost center to a profit center. For that we need to be very clear and precise on what is the cost and tokens that we are consuming.
[07:20] And then spike detection. We also want to see like what where where are the spikes coming up? What is Is there a spike in the latency? Is there a spike in in the inaccuracy? And we're also tracking that. And these slides are more from a business perspective, like how many hours we are saving from from the supply
[07:37] chain, so that they can be more productive, total questions success rate, and so on. Again, performance, more performance deep dive, user engagement, and detailed question explo- exploration. Like, we are also giving an option to give a negative thumbs down if
[07:52] the user is not feeling good about the response where they can key in they can type what is the the negative feedback. We are also monitoring that so that we can work work on it. With that, I will hand it over to Mitch. All right.
[08:08] Thank you, Karthik. So, with our architecture in place, we step into the pillars of the supply chain. And before we get into agent supervision, right, we we set up individual genie spaces. Um and we've been uh doing that as a learning exercise to to just understand
[08:23] how to best curate the genie. Um what's the the right amount of scope? Um what's the right context that we need to answer um to to provide for the genie? Um and then connecting that with domain-specific users. So, production planners, people that work with the
[08:40] master data, uh people that schedule um purchase orders, um demand planners. And and then getting that feedback, right? And they're they're telling us where it's getting it right, where it's getting it wrong, but also how are they going to use it in a day-to-day space, right? And so, with the individual genie
[08:57] spaces, right, that's that's not the vision of how we want to see supply chain users um interact with the data uh cuz people typically ask uh cross-functional questions, right? They ask, "Why am I out of inventory?" We might need to know what the history of the demand forecast was. We might need
[09:14] to know that the safety stock changed last week, right? Um So, you you you need to see some of these patterns and that happens across uh the different genie spaces. So, that's that's where agent supervision comes in. Um so, our data engineer um
[09:30] conversational AI expert, Harry, uh has built this for us and uh honestly, what I love about it is um in the last 6 months the my my phone rings, phone numbers that I don't know, I'm getting flooded with
[09:47] emails, everybody is pitching me end-to-end supply chain orchestration. And Karthik and I took a few pitches. And one of the pitches we saw, we met up afterward and um kind of talked to each other like, was that just Genie that we saw? Like they just were trying to sell us Genie. So,
[10:04] you know what I love about in the keynote they said, you know, talking about getting locked in. And and that's where it's like 6 months ago I thought, we either build this ourselves or we outsource it to a third-party vendor. And especially after this week, I'm I'm realizing like everything's coming
[10:20] online with Databricks. So, really happy with where we are. So, I'm going to get go ahead and get into the demonstration.
[10:36] Should I click the Okay. There we are. All right. So, the users are um met with uh top-line metrics. Um those are dynamic, so we can change those over time. Um they're prompted with, you know, what what you could potentially ask. You see your conversational history. And so,
[10:52] first we'll ask a relatively simple question. Tell me about the demand, supply, and inventory for a product within a plant.
[11:10] So, we can see that the supervisor's doing the routing, right? So, I worked with Hari to to describe when it should go to which Genie. Um it's evaluating which Genies to prompt and then performing some reasoning.
[11:28] All right. So, we get a nice summary, right? And right now we're we're still kind of getting feedback from the end users. So, how should we answer the question, right? Um but I think so far it does a pretty good job. So, then we could show you um a more complex question. This one asking for
[11:43] We'll wait for it to finish. The next one asking for some analysis on the master data planning parameters, right? So, the first one's just telling us about, you know, the status of our inventory and our planning, but this one's going to tell us, you know, is the master data that um
[11:59] that feeds into MRP is how's it how's it doing? Should we consider something else, right? So, it's going to be telling us about maybe min lot sizes, rounding values. And so, we're really pushing um you know, the end users to kind of think
[12:16] more about how they could use this in their day-to-day. Yeah. And this is where they can give the negative feedback. Also positive feedback. Yeah. From IT perspective, we are more interested in the negative feedback.
[12:32] To give to solve that. Oh, yeah. And all the reports are all downloadable. You know, you get that in Genie 1, but also in the agent supervisor I made sure that Harry had a the little download indicator. So, let's see.
[12:48] All right. So, um from here Karthik and I have been pushing the team to see where we go from supervisor agent. So, like I said, we've been hearing pitches from third-party vendors saying we can do end-to-end supply chain orchestration. So, agent supervisor is just one half of this, right? It's
[13:04] surface insights. Um it's, you know, talk to your data. And then the next piece is the agentic piece, right? So, how can we start having people um incorporate this into their day-to-day? Um how can we have it um evaluate master data, make
[13:19] master data corrections, how can we have it flag where there's purchase requisitions missing when where MRP might have failed, and what can we do about that? And so we can see our first approach since we're already such a heavy Power BI company, we
[13:35] thought it made sense to ease people in with something familiar. So we're starting out with metrics, right? And so this year David Kohler announced the 153 plan which is our our grand next vision.
[13:50] It's it's the strategic objectives from the highest level, right? And and he gives those to his C-suite and they get filtered down, right? And they get down to the VPs and the directors and the managers and everybody has, you know, their objectives. And those objectives
[14:06] at the end of the day they tie into what you do every day, right? They tie into the transactions that you do. And so what we're building here is something that ties a direct line from the production planning and the corrections to the master data and the routings on the floor all the
[14:23] way up through the VPs to the CEO. And cuz our our chief supply chain officer was saying, "Well, how do I plan this, right? I'm not going to be asking it about, you know, should I firm this planned order?" But I said, "You know, you can ask it
[14:38] high-level questions and it can tell you how the business is doing because it's tied to all the metrics, right?" Thank you for coming. Yep. Thank you a lot.
[14:54] Hello everyone. Welcome to today's session on systems of intelligence for the supply chain. In this session, we are going to look at a case study from Belden how they went from fragmented supplier visibility to revenue aware decision making. My name is Sachin Gadre. I'm the director of product for
[15:09] for Resilinc. I'm going to take you through half of the slides, but prior to that we're going to look at the transformation that Belden has done. So I have two speakers with me, Shakeel who unfortunately could not make it to this summit, but he has recorded his slides. So, we'll go through that. And then I
[15:25] have Sudeep. He he's going to he's the VP of data at Belden, and he's going to take us through what Belden does. So, let's let's take a minute to walk through what Resilink is. So, over the last 16
[15:42] years, we have helped customers big and small, mostly Fortune 500 customers, through their supply chain risk, resilience, and compliance programs. Now, over the over that period, we have worked with over million
[15:59] plus suppliers, and also millions of parts. Now, that data is part of the Resilink offering. As we know that AI is only good as the data. Now, our leadership has been recognized by Gartner. They have made us the leader in
[16:15] the Gartner supply chain risk and compliance quadrants for the year 2025, as well as 2026. We are Azure uh native, and we are FedRAMP certified. Uh and uh you know, the slide shows some of
[16:30] our well-known customer implementations. So, let me hand it over to Sudeep to take us through uh and explain what Belden does. Sure. Uh good morning, and thanks for the opportunity to talk about Belden and the partnership that uh we we had for
[16:46] last several years, and we are going to have uh a lot of this kind of discussions with the uh things that we saw in Databricks sessions today uh this this week. So, I'm very excited. So, it's a 124-years-old company, as you can see,
[17:02] um 2.7 billion uh uh revenue, and it's in the business of uh connectivity software, mass smart solutions. Uh and it started with a cable business, but then again, cable, copper, fiber, and all kinds of connectivity.
[17:18] Um and one of our goals is this year which I have started is the IT OT convergence. It's a big goal from our CEO coming down from the top. Um the idea is to make sure that not only we build in within Belden the data and data
[17:33] platform and the platform that will do the IT OT convergence between ERP, PLM, all our innovations kind of get connected. Um but then we will also invest in all these other application build that you know Redline is bringing. Over to
[17:50] you. Thanks Sudeep. So now I will hand it over to Shakeel who is going who has provided us his recorded presentation. So let us walk through those slides and I hope that the technology all works. So over to Shakeel.
[18:31] which we are taking decision. So at Belden This slide illustrates how Belden transform fragmented supplier risk and operational data into actionable decisions at scale.
[18:46] Every enterprise has data. At Belden the challenge wasn't lack of data. It was turning millions of disconnected signals into timely decisions. The differentiator is always the velocity and intelligence and together they enable the decision intelligence
[19:03] that you and strengthen the supply chain resilience. Okay. To understand our strategy, it's important to understand the scale of the
[19:19] challenge we are surrounded by. Belden operates across 16 global entities globally, manages relationship with approximately 1,700 plus tier one suppliers, and spends over 1 billion annually with the suppliers.
[19:34] The issue wasn't lack of data for us. We had data everywhere. The challenge was that it was fragmented across systems, suppliers, and risk sources. We had visibility into our tier one suppliers, but limited insight deeper into the supply network.
[19:51] Relied on disconnected tools and manual analysis, and we faced increasing geopolitical and supplier risk and lacked a direct way to quantify how supply chain disruption impacted revenue. So, these challenges ultimately led us to build a unified
[20:06] decision intelligence capability. So, when a disruption occurs, leadership isn't asking for another dashboard. They are asking one question, what revenue is at risk? Answering that question requires tracing a signal through the
[20:22] entire supply chain from the event to the affected supplier and site to the impacted material product, and ultimately the revenue exposure is visible. Historically, this analysis took days of manual efforts because time to
[20:39] awareness is where we lose the majority of time, and then from there the time to action, and then time to recover gets into picture. So, from the days of manual efforts across multiple teams and system, so by combining Databricks with end-to-end
[20:55] supply chain intelligence, we can now quantify revenue exposure and recommend mitigation actions in seconds rather than days. The next pandemic highlighted a critical reality.
[21:10] The most significant risk often reside beyond tier one suppliers. The disruption didn't originate with one of our direct suppliers. It originated several tiers deeper in the supply network, well below our traditional line of sight for the electronics commodity.
[21:27] What made this event significant was not just the disruption itself, but how quickly leadership needed answers. We needed to understand where the dependency originated, which parts and products were exposed, whether contingency plans existed, and ultimately how much revenue was at risk.
[21:44] So, answering those questions exposed the need of a complete supply chain lineage, and became the catalyst for building our decision intelligence platform.
[22:00] One of the key insights from our journey was that a product is no longer just a bill of materials. It is a global dependency network. For example, the Belden industrial ethernet switch depends on components, materials, and suppliers spread across multiple regions, and each exposed to
[22:17] different risks such as labor shortages, geopolitical conflicts, tariffs, forced labor regulations, and raw material constraints. A disruption anywhere in that network can ultimately impact product availability and revenue. The challenge was connecting those upstream
[22:32] dependencies to downstream business impact, which is exactly what our supply chain lineage capability enables us now to do. Once we Once we establish the end-to-end supply chain lineage, the
[22:49] next challenge was operationalizing the decisions at scale. Visibility alone doesn't improve resilience, but better decisions do. We know that. So, to move from insight to action, we partnered with Resilinc to build what we call a supplier decision
[23:05] intelligence. At the core is a composite score that combines three dimensions: spend, revenue, exposure, and risk resiliency. So, instead of evaluating suppliers based solely on spend, we prioritize them based on their potential business impact. That single score
[23:21] automatically drives two decision views: the well-known Kraljic Matrix for category and sourcing strategies, and a McKinsey-style supplier pyramid for governance and engagement. So, most importantly, it made the long tail visible, allowing us to manage suppliers
[23:37] based on risk and revenue impact, not just procurement spend. Ultimately, resiliency is measured by how quickly you can move from awareness to action. Before this transformation, responding
[23:52] to a disruption required days of manual effort across teams, systems, and spreadsheets. Today, as Valdemar Because supplier relationships, product dependencies, risk signals, and revenue impacts are already connected, we can compress that cycle dramatically.
[24:07] Analyses that once took 40 hours for us now take minutes. Manual effort has been reduced by up to 80%. Hundreds of suppliers have been rationalized and prioritized, and revenue at risk question can can be answered during the meeting instead of after it. The outcome
[24:24] is simple. We spend less time understanding the problem and more time mitigating it. If there's one takeaway from our journey, it's that resiliency is fundamentally a data problem before it's an operations problem. The answer
[24:40] leadership needs, what is at risk, how much revenue is exposed, what action should we take, depend on connecting data that traditionally lives in silos. So, Databricks gave us the foundation to unify those relationships at scale.
[24:55] Resilinc provided the supply chain intelligence that we needed, and together they enabled a decision intelligence capability for us as an organization that transforms disruption signals into business action. The result is faster decisions, greater resilience,
[25:11] and better production protection of revenue when a disruption occurs. Okay, so we heard Shakeel. He talked us through the supply chain transformation
[25:27] that Belden went through from having fragmented supplier visibility to revenue-aware decision-making. Now, I'm going to take you through the underlying Resilinc platform that is built on top of Databricks that made it possible for Belden to achieve those business
[25:43] outcomes. Now, when you talk about a question like what revenue is at risk, you cannot find the answer in one ERP table or a dashboard. For that, you need to connect the dots between products, parts, materials,
[25:59] suppliers, sub-tier sites, shipments, geographies, risk and compliance signals, and you need to connect all of that into a living, breathing supply chain graph. And this is the core asset that Resilinc provides.
[26:15] So, Resilinc builds and validates this map. We use Databricks as the scalable data and AI layer that allows us to compute on it, govern it, share it with our customers, and finally embed those
[26:31] that intelligence into customer workflows. So, I want to point out one thing that we are not just providing supply chain data, but we are providing an interconnected graph. So, let me show you how that graph is made. So, Resilinc platform supports the
[26:49] the sense-recommend-act paradigm. So, as part of the sensing, we need to build the network. So, the first thing is we start with the enterprise context. And enterprise context here is every company knows who their tier one suppliers are and what parts they provide. Then we add to it
[27:07] products, maybe purchase orders, EDI documents, automated shipment notifications, as much context as we can get. Now, the initial map that gets created is only till tier one, but we know that uh from experience that most
[27:22] of the supply chain risk and compliance issues occurs in the sub-tier supply chain. So, starting with tier two and below. So, that is where we then triangulate the known enterprise context with third-party data. Now, this is third-party data such as global trade
[27:38] shipment records. It could be community data, it could be open-source data. It includes product certification data including risk and compliance signals. All of that is combined to create the multi-tier supply chain graph. This is a
[27:54] product aware trusted supply chain graph. Let me show you how that is built under the hood. So, we look into the Resilink Intelligence Factory. This is a series We run a series of AI and data processing steps at scale. And what we do is we first start with entity
[28:10] resolution. So, the supplier name is not the same across your internal procurement systems, your global trade shipment data, and third-party data. So, what we do is we normalize the name and make it we standardize it so that we can work
[28:27] through that. Once the name is normalized, the next thing is to identify missing data, infer relationships. Also, we enrich the data with sub-tier site information. And finally, we have built uh an a machine learning AI machine
[28:42] learning model called as material breakdown, which allows us to filter this map and create a a supply chain map that can be trusted and that is tailored for a particular customer organization. So, what we are doing here is
[28:59] we are just not using AI to build the map, but for every relationship we are scoring, we are providing a confidence score. So, this way a user can distinguish between a relationship that is 60% confidence versus 90% confident
[29:15] and that actually triggers a built-in validation loop. So, you can Ultimately, you get a very trusted multi-tier tier N, which could be tier 10 supply chain graph. Now, how is Databricks used and underneath the platform? So, we use the
[29:32] standard medallion architecture. We have bronze, silver, gold layer. At the bronze layer, we we ingest a lot of the raw data. And this is all the known enterprise context, the third-party data, all of that gets sucked in. At the silver layer, we do entity resolution.
[29:48] We identify product material inference. We apply data quality rules. We In In short, we actually create a high-quality supply chain graph. And then at the gold layer, it is all business data ready product. So, it is the graph, but along with it we have revenue at risk
[30:04] roll-ups. We do UDR exposure, uh uh foreign labor uh exposure and things of that nature. So, all of that is provided by Databricks. We use a number of Databricks components starting with
[30:20] uh obviously the the main Databricks uh ins uh Apache Spark for for data processing, Delta Lake for storage and reliability, Unity Catalog for governance and lineage, Databricks SQL Delta Sharing for sharing this data with our customers, and finally MLflow for uh
[30:38] AI/ML life cycle management. Now, uh now that we know that now that we have the multi-tier trusted graph, the next step is to get insights, recommend actions, and then take autonomous actions on this map. So,
[30:54] let's take an example that there is a semiconductor constraint that surfaces at tier three. Now, this constraint could be a geopolitical event, it could be any kind of a disruption event, and that's going to impact some of your suppliers. So, the first step is the is
[31:10] to detect the signal, is to detect the event. The second step is now to identify the suppliers, the sites, the parts, the materials that are impacted by that event. Now, once we identify that, then we have a bunch of recommended actions. Some of the
[31:26] recommended recommended actions could be connecting to the supplier, like sending an email and asking for what's going on. It could be looking at your safety stock and maybe increasing it, or maybe figuring out a dual sourcing strategy for some of your
[31:43] critical materials. Now, going beyond that, we provide an agentic platform, and we have several task-specific agents. So, these agents can then act on those recommended actions, and they can act independently, but more importantly, it is always human
[32:00] in the loop, and they can send emails, they can change your sourcing strategy, they can update executives, and then finally the system is learning. It is constantly updating the graph. Uh we are taking all of the KPIs, like
[32:16] how soon the supplier is responding to our questions. How soon can they recover from this disruption? So, all these KPIs are calculated and are updated in the supplier profile and are used for supplier risk scoring. So, the future of
[32:33] operational AI is not just a chatbot, but it is governed human in the loop actions. And the main thing here is that if you look at step one to step five, previously this used to take days, weeks, months, but now because of AI,
[32:50] this whole thing is getting compressed in minutes. Maybe in hours. And that is the kind of impact that we have had at Belding. So, now that we have this intelligence layer, you know, I want to talk about how customers can use this layer. And
[33:06] broadly speaking, there are three use cases. The first one is the executive resilience genie. So, we have cases where executives, they can go to their chatbot, they can go uh to their Data Bricks genie, they can fire away questions. What revenue is at risk? What
[33:21] events are impacting me? You know, what are my suppliers are at risk? So, you can ask all these questions. The second use case is Delta sharing all this intelligence information. So, we can take all of this intelligence information, we can share it with our
[33:37] customers. We have done that and customers use that information to then build their own internal applications. And then finally, we can have an agentic disruption response. This is where the fact that we have an agentic platform and we have developed task-specific agents allows us
[33:54] to automatically, autonomously work on these actions. And this could include you know, changing your sourcing strategy. It could It could include finding alternate suppliers. It could include uh uh alerting your executives or your
[34:09] suppliers. So, all of that is not the future. It can be done today, and our customers are doing it. So, I just want to wrap up this session with three things. One is Belding was extremely successful in transforming their application uh their
[34:26] operations from fragmented supplier visibility to revenue-aware decision-making. Second thing is Resilinc provides the supply chain intelligence layer that made it possible. And when I say the supply chain intelligence layer, it includes material breakdown, N-tier mapping,
[34:42] confidence scoring, event watch, exposure, all of this stuff. And then finally, this would not be possible without the Databricks platform because it provides that scalable and AI-ready layer. And we use a lot of the
[34:59] capabilities like govern lake houses, compute, we use the the the Genie products, we use analytics from from from Databricks. So, basically, the story started in supply chain, but it is has wide applicability.
[35:15] And one of the things I want to leave you with is that every enterprise has signals. But the winners build systems of intelligence and then act on that signals while there is time. So, I want to leave you with that. If you have any questions, I will be available around here. But thanks for attending the
[35:31] session. Thank you.

Learn more about the Databricks Data and AI platform.

The information provided herein is for general informational purposes only and may not reflect the most current product capabilities or configurations.