TTEC Digital + Microsoft: Agentic AI orchestration platform

Good afternoon, everybody. My name is Chris Bennett, VP of Azure here at TTEC Digital, joined by Puneet J. Singh, who's a principal architect at Microsoft on the EPS team with a background specialty in data and AI. And Chuck Tischner from our team who's an enterprise solution architect will be walking through a demo of TTEC's AgenTeq ARK orchestration platform. What we're here to talk about is not only how to enable AgenTeq AI, which is a, you know, a priority for most organizations, but level set the reality of what needs to be done to enable that to be successful and go through not only Microsoft's pattern for, being able to enable the solutions, but also TTEC's opinionated, means of of of going quickly, with a couple background demos to showcase some of the functionality and some of the value you can get out of the Microsoft ecosystem. So quick agenda run by. We're going to go through building your aid or agents your way, with Puneet cover how Microsoft is looking at the ecosystem. We're gonna go talk doc talk through the transition from apps to agents and what this means for our end users and, you know, really showcasing the value of Agenetic AI solutions in the enterprise going forward. And then we're gonna walk through a demo, of how we've built this out, not the only way, but a an opinionated fashion to be able to get going quickly, as well as the TTEC offering on, how we can help organizations accelerate to get here, quicker. As a quick background, TTEC Digital is part of, TTEC, which is a BPO, multinational organization serving up call center of use cases in various industries all over the world. Our area of that, for the Microsoft business is focused on Azure, data and AI, custom applications, and the integration of AI with enterprise enterprise systems, as well as the the full Microsoft business application suite. So Dynamics three sixty five, sales service, and grounded in a focus on delivering stellar customer experience outcomes both to internal IT users as well as helping organizations enhance their brand externally in in their respective markets. Part of this is our ability to bring the technology side and part of it is also grounded in consulting and years of experience in customer experience. From a data and insights and analytics perspective, which is the foundation for AgenTeq AI, we have a long standing practice with a deep bench of delivery talent, as well as a number of learnings from the inception of AI and ML based solutions in Azure that we've been deploying in large enterprise for complex use cases for years. Enough about us for the moment, I wanna throw it over to Puneet to go through Microsoft's approach to AgenTik AI and give us some background from the Microsoft Lens. Awesome. Well, thank you so much, Chris. So Microsoft views AgenTik AI as the next paradigm of software. Intelligent agents that can reason, plan, and act autonomously and orchestrate multi steps across systems and secure governance with deep enterprise integration. So these agents actually are not just chatbots. They are programmable, they are extendable, and production ready assistants that can automate workflows, analyze data, interact with tools, and collaborate with humans and other agents. And at Microsoft Build twenty twenty five, we introduced the bold vision of what we call it is the Open Agenetic Web. This is the future where AI agents are not just add ons. They are the first class citizens in the digital ecosystem. And these agents operate seamlessly across software, services, and platforms. And they make decisions, and they interact with tools and data, and automate complex workflows end to end. So why does it matter? Because it transforms how businesses and individuals engage with technology. So instead of siloed applications, we are moving towards interconnected intelligent fabric where agents can orchestrate tasks and adapt dynamically, and they can unlock new levels of productivity and innovation. And we are helping partners and customers build agents their own way, right? So this can be done with the breadth of our tool sets. And we have the tools to build the agents across the spectrum from the low code to pro code. So what you are looking at the screen is that you can build the agents in the low code platform. That's our platform called Copilot Studio. So it is our low code graphical platform for creating and customizing AI agents that work across the organisations. So what it does, it lets the makers and the IT teams design, test, and deploy agents using the natural language or visual workflows and connect them to the data sources, tools, and systems. And you can publish them into Microsoft three sixty five Copilot, Teams, or other applications. And then the second one on the right you see, it is that Microsoft integrates agentic AI deeply into Visual Studio Code and also GitHub. So with the GitHub Copilot, agents can automatically handle coding tasks, bug fixes, code generation. And you know that the code editors are evolving, so model selection, debugging support, and deep integration with the version control and CICD. And then also, the developers can build, test and manage custom agents that can connect directly to the enterprise pipelines through GitHub Actions and also with the Azure DevOps. And then the next one is the Azure AI Foundry, which provides the design, deployment, and orchestration tooling for complex multi agent systems. And you can leverage many models in Foundry. So we have tons of models there. We are extending the set of models that we host and as well as we sell. So in addition to the OpenAI family, we also have DeepSake, Mistral, Grok, Meta, Black Forest Lab, and a set of matters that not only provide that unified access, but also enable you to easily switch them in your code. And we are also embracing protocols like MCP, which is enabling agents to securely and dynamically discover and interact with services, apps, and tools. So this makes the AgenTeq AI open and extensible rather than locked by the proprietary APIs. And last but the least, nothing but the least, the main thing is that today we are going to also showcase Fabric Data Agent, and you will see the TTEC team give you a demo as well. So Fabric Data Agent is the new AI powered features in Microsoft Fabric, which lets users interact with their enterprise data using natural language questions without having to write code manually. Next slide, please. So this is a high level slide. And from Microsoft, we are also providing guidance on how to select the right technology platform for each of your potential agent use cases. So we have the AI agent decision tree guides. And you can check that out. And I have also provided a link here as well. You can get this link in Microsoft Learn, too. The question is whether you are using the SAS agent. Is it meeting your functional needs or not? Or which agent you should use for your particular use case? So take a look at that. This document is very helpful. Next slide, please. So today, as you guys are going to see a demo towards the end on the FABRIC and FABRIC data agents, I'm going to, at a very high level, talk about that and talk about what basically a FABRIC data agent is. So fabric data agents are the conversational agent that are built using generative AI that help you get your insights over your data that lives in Fabric OneLake. So you can think of them as kind of like the virtual analysts that help you answer questions and get insights over that data. So Fabric Data Agents, what they are doing is they are providing that seamless connectivity with any type of the data, whether the data lives in Fabric Van Lake, whether it's living in lake house, warehouse, Power BI semantic models, or KQL databases, or even the mirror databases. So there is an AI creator who would basically configure and customize this conversational agent and provide instructions as in any agentic framework to basically allow AI to learn and understand the unique language and nuances of your business. And then, so once basically that AI creator has built this agent, then they would be able to consume it both in Fabric and also outside of Fabric. And from Azure AI Foundry to Microsoft Copilot Studio, Teams, and within Fabric itself. And on the right hand side, you can see some of the latest and greatest things that we have announced during Ignite. First of all, the very first one is that you can start consuming Fabric Data Agent in Microsoft three sixty five Copilot. And the second one is that you are able to use Fabric Data Agent as a remote MCP server, starting with Versus Code and then extending it to all other AI systems. And Fabric Data Agent also supports unstructured data through Azure AI Search. And finally, we have Fabric IQ. That is something new that we have launched during Ignite. And when you are adding the ontology from Fabric IQ as a knowledge source in the data agents, that data agents are gaining their semantic understanding of the business. So I think you will see more all of this in a demo. So now I'll give it back to Chris. Thanks, Puneet. Ton of functionality there, quite a bit of capabilities. So it's no wonder that we've been seeing a desire to start peeling away certain business applications, at least functionality from them to be able to join with enterprise data. The challenge is data is everywhere. And one of the key things that we know solves that problem is being able to aggregate and build out both customer enterprise data platforms. People have been doing that for years. But the big thing is we've got an opportunity with Fabric, with other Microsoft technologies to unify existing investments into a purpose built AI data platform with the intent of not only bringing MCP capabilities for multiple data sources, but also being able to be extensible to external data as well. So I think being able to join external cloud data sources natively to fabric than to expose from an authentic workflow perspective. Quite a bit of flexibility with this, which allows us to start rethinking the application of of some of our business applications within the enterprise and gives us a robust toolset to be able to build something net new instead of potentially buying something off the shelf that you have to fit into how the product works. You're now able to build custom tailored to your needs. How this looks, and and one of the one of the the focuses of our our coming demo is the ability to harmonize all a myriad of different data sources. Now there could be an existing, Databricks implementation that's been heavily invested in via, the the organization. We can use that. The the the pattern fits. We could be potentially unifying external in a myriad of different data sources using fabric, bringing that into OneLake to then, you know, again, exposed to the agentic frameworks that Puneet went over previously. The idea here is we centralize and then we are then able to imagine and map out business processes and outcomes that we're looking to wrap technology around and having access to the data required to take action and the ability to build autonomous agents to truly automate, it's a very powerful platform. What we've seen with this in a couple of the industries where we've heavily applied it to is the ability not necessarily to replace humans, but to augment and make them incredibly efficient so that they're able to do more and also be more effective. So think being able to surface relevant client data while on a call with them, being able to search through enterprise knowledge source to be able to get to an answer quickly. It's down to being able to automate portions of documentation creation, RFP response, things that are repetitive are low hanging fruit, but with the implementation of Identical Gas Solutions, some of the more complex tasks because of our ability to use multiple agents are also now much, much lower hanging fruit. And we've seen pieces of traditional, business systems such as CRM start to fall into this pattern. And the outcomes of it are something that we're gonna show you, and I'm gonna turn it over to Chuck, to go through what our demo is going to show, and then actually show that. Chuck? Alright. Great. Thanks, Chris. Chuck Tischner, enterprise solutions architect from TTEC. Let me share my screen here. So on the screen, we can see the overall architecture of the solution, which covers your data sources, SQL Server Dataverse, SAP, or hundreds of others. The processing of that data, the modeling of that data for consumption by your AI agents, then serving up to your users, whether those users are a PowerPlat, Dynamics, Teams, CoPilot Studio, AI Foundry agent integration as well. We're going to be flowing from left to right in this particular demo. We're going be pulling data from SQL Server through Azure Databricks into a data lake, which will be shortcut into the one lake for fabric processing. In fabric, we're going to create a semantic model for consumption by a fabric data agent, which will then be served up in CoPilot Studio. So the first thing we need to do is ingest the data. Here's a small notebook. You could also pull the data in through Azure Data Factory or other Data Factory methods and process it with other notebooks. This particular implementation shows Databricks. We're pulling in the different tables here from our SQL Server into a data lake area, which is hosted in an Azure storage account. And we see we've segregated the data by the different tables that we're pulling in. And this is all hosted in Azure storage account, which is incrementally updated by the notebook. Once we've got the data staged in our storage account, we can then pull that over into Fabric. So this is a Fabric Lakehouse, which has the data and you can see the data shortcut in. Set up a shortcut here connected to our data lake. Azure Data Lake Gen two, pulled in the data, created the lake house. And here you can see the products that are part of our model. This is a product buy sell model. So, we're capturing buy and sell history associated with products from a consigner or retailer. So this is our Data Lake format, but what we need to do once we've got it in the Data Lake, we need to create a semantic model, which is an intelligence data model that will give the Fabric Data Agent the information and the intelligence it needs in order to return results in a high quality manner. If we just pulled the data in and we didn't do any modeling, then you would have a situation where you would still get responses back from your data agent, but they wouldn't be as high quality because you hadn't groomed the data enough to make it a high quality result. So as you can see, what we do is we put together the data that came in into a nice data model with data connections so that the data agent, once we configure it, can understand the relationship with things like products, price ranges, product brands, product categories, and then can relate the product to the buy history of that product. That can be related to stores where the products are sold, employees who are selling it. So having all these relationships and this rich data model is very valuable in terms of training your agent. So once we have the data ingested, the lakehouse defined, and then the model defined, we can then create an agent within the context of Fabric as well. And what we do is the first thing is we pull in a data source amongst the many data sources that we can pick from in our Data Hub. We pull that data in, we select which data we want to be part of it, then we provide it with instructions because we have to tell it, Hey, what does this table mean? What does store mean? What does product mean? What does employee mean? So, we give it some instructions. And then that enables the agent to understand the data sources and give us valuable answers as a result of the data that it's analyzed. And as you can see here, what we've done is we're asking for the particular sales per category this year. And because it has that relationship, it knows the relationship between the category, the product category, and the sales associated with the product. It's able to put this table out. Then typically what users will want to do is they'll want to drill down, Hey, I saw that footwear was pretty high, so what are footwear sales? Are they trending this year? You can ask that question. This is more of an interpretive question, so it's really relying on the agent's capability. You get an interpretive answer. In this case, the sales for footwear have gone up and down. Then you could drill down even further, show the footwear sales by region for year to date, which region has the best footwear sales. As we can see here, we've got a mix here. It's mostly Midwest, but we've got some Pacific Northwest sales that seem to be the highest. And then you can dig into that even further and drill down if you want to and say, Hey, what vendor did the best sales on footwear for this year? So you can really have a rich conversation with that fabric data agent. So once you've pulled in the data, trained the agent with instructions, done some testing on the right hand side, then you can publish the data, the data agent, Then that data agent is ready for consumption within the context of other systems such as CoPilot Studio. Over here, we see CoPilot Studio consuming that agent. Once again, I've included that agent in here and I give it instructions. I tell it, Hey, under these conditions, these are the conditions in which you can use this agent. Because very importantly, CoPilot Studio can use multiple agents and orchestrate multiple agents. So, you need to tell it when to use this agent and when not to use this agent. And then you can have conversations with that agent in the same way you did with Fabric Data Agent. Now note that the CoPilot Studio adds its own flare. It adds some highlighting here in the headers and some highlighting of responses, but it's still information coming from your fabric data agent and being orchestrated by Copilot Studio. So we saw the full path all the way from a SQL data source, all the way through Databricks into a fabric data agent and then through consumption and Copilot Studio. Back to you, Chris. All right, back live. Thanks Chuck for walking through what the back end of the sauces making process looks like. Obviously, we've got some different data models into some different industries that we've done this in, but at end of the day, the pattern's the same. The centralized data leverage the existing sunk costs in terms of investment and where opportunistic and where necessary, do some ETL, do some restructuring, move data around into a place so we can start actually getting some action out of it. And one thing that wasn't explicitly mentioned is all of this is governed and secure. With security by default, through everything that we do, we've been born out of regulated industries. So that is a default stance for us with the ability to integrate downstream into any tools needed. Because we've done this a number of times since the As the technologies evolved, we've shifted some of the tooling, but we've been delivering these types of solutions for years. We have a fairly prescriptive way to start this. One of the areas that has been the lowest hanging fruit is around CRM because of the typical separation of customer data platforms and enterprise data platforms and having a nice place to to bring in the AgenTic accelerator to be able to provide immediate value and communication from both. But this is something that will be in the Azure marketplace for review and we can send over some additional materials prior to the webinar outcomes as well. What that looks like is a code based delivery of the solution after upfront architecture design and blueprint. The big thing that we found both through implementation as well as through our AI consulting practices over the years is human centered design and the ability to bring in or to get buy in rather at the onset of the process and build something people are actually going to use drives adoption and causes AI solutions to actually be successful. A lot of things die in the vine because that process isn't done correctly, which is a big part of our upfront. The core platform, we have that fairly prescriptive at this point, there's only a few different variations. So that is something that tends to be pretty quick. And then really driving value to the overall implementation is where we spend the vast majority of our time. I'd like to thank everybody for joining. Hopefully the brief demo was insightful and the background for Puneet on Microsoft's approach to the overall AgenTeq AI paradigm and us being able to show you how you can build agents your way with very easy to use tools. If there's any questions or any follow-up needed, please reach out to us and thank you everybody for your time.