Name: Agentic Workflows: Learn how to build trustworthy support automations in under 30 minutes
Uploaded: 2025-11-05T19:01:13.483Z
Duration: 46 min 44 s
Description: Agentic Workflows: Learn how to build trustworthy support automations in under 30 minutes

Transcript for "Agentic Workflows: Learn how to build trustworthy support automations in under 30 minutes": Alrighty. I think it is time. Great. Hello. Hello, everyone. Happy Wednesday, and thank you all so much for joining us to for today's session all about AgenTic workflows. We really appreciate you taking the time to be with us. My name is Cassandra Stumer. I work on product marketing here at Assembled, and I am joined today by Sid Shetty from our solutions engineering team. And if we've got a couple of folks in, please sound off in the chat. I wanna hear where you're coming in from. We got a chat box to the right. I'd love to see some engagement there. We are here in what was rainy San Francisco. Let me know where you're chatting in from. Yeah. I always say that from sunny San Francisco, but today is not one of those days. It is definitely not. The other thing we were just chatting about, is in our last webinar, we had an icebreaker around, our favorite AI tools at the moment. Like, what's your favorite AI unlock, AI tool, platform, whatever you wanna call it that you're using that is amazing? This time, we wanna start with a pain point. Like, what is your AI challenge you are facing right now? If you have one, let us know in the chat. I'm really curious. If you don't wanna tell us where you're calling in from. I know mine recently has been, every AI tool claims to be able to make slides, you know, for in this presentation. I have yet to find, a good AI tool or AI solution that can actually make visuals for a deck like this. But you said you probably have a more technical one. I finally do have a very technical one. I feel like I get these barrage of questions in an RFP, like, 300 questions. And I keep noticing when I have, like, AI fill these out for me and do a first cut. If it's 10 questions, I get each answer, let's say, is 10 lines long. And if I have 300 questions, I get a much shorter answer. I'm unable to find the sweet spot just yet in terms of how to parse the input and get an output that is appropriate for the size of the questionnaire. So, yeah, not the most exciting and interesting one, but something that is taking away a lot of my time lately. It's a real use case and a real challenge. It's true. And it's a good one. Yeah. Let us know too, throughout this, presentation. If you have commentary, questions, pop them in the chat. Let us know. And I'm just gonna get right here into the agenda now that we are at a couple minutes here past two. So in this session, as we've noted, we're gonna take you through a little bit of the what and why behind AgenTic workflows. And then Sajid's gonna jump right into the product, into Assembled, and we're gonna run through a trustworthy automation and, hopefully, in record time. I know we're saying thirty minutes, but we are sharing a bit at the end there, about how we're seeing some folks use agentic workflows, and we are gonna save a bit of time at the end for, some q and a. So if you do have questions, throw them in the chat now. And if we don't get to them live in the moment, we will certainly get to them at the end, pop them right into the side of your screen there. First, I just wanna give you a little bit of background, to ground the session here, for those who may be unfamiliar with us at Assembled. Assembled, we've actually been in the support game since 2018, which seems like so long ago now. Our founders, started Assembled after building workforce management tools at Stripe all the way back when that was called just machine learning. And, today, we do have about a 120 employees with offices in San Francisco, New York, London, constantly growing. Let us know if you have people that would like to work at Assembled. And what exactly do we do here? So, we are not just another AI agent provider. Assembled is really the orchestration layer that makes AI and humans work together seamlessly. And I know a lot of folks are saying that, but it's really true here at assembled because you can actually plan, staff, and run both your human and AI agents on one platform, and that's not just a copilot. This was historically two teams, two separate worlds, your workforce management tools for humans, and then you also have your isolated AI vendors for automation. And assembled really brings those two things together uniquely so you can see and also manage your full operation in one place. And we've really been architected for today's multichannel CX environment. So we plug right into your existing stack, all the good stuff, the contact platform, Zendesk, Salesforce, all your tools, Shopify returns, custom APIs. We do it. We work with a ton of fast growing companies, enterprise, of every size to Canva, Patreon, Etsy, Ramp, just to name a few. But without further ado, let's just get into today's topic at hand. So for a bit of background here, I just wanna share with you all this quote from Andrew Ng. And, he's saying here that, the greatest opportunity for businesses today is in agentic workflows and not just scaling AI. And, if you're not familiar, Andrew Ng, he's really widely recognized as shaping the trajectory of modern AI and, academia, but also cross industries. He's the cofounder of Google Brain, and this is a recent quote, from a talk he did at Snowflake Summit. And he went on to say that, really, if we just focus on traditional AI at this stage, we're just gonna have to increase continue to increase the size, or data intake of models to make this traditional AI useful for real business cases. So that paradigm already over giving away a new era. Right? And that's driven by sequencing together multiple agents capable of multistep reasoning, planning, tool use. It's what we're getting into today. That's what he calls and what we call an assembled agentic workflows. An agentic, AI automation, AI agents, there's a lot of terminology floating around here and everywhere. So just wanna make some distinctions here, by diving into a little detail about the differences, and we're gonna take it all the way back to simple automation. And, workflows have been around forever. Simple automated workflows, nothing new. We've been using these sort of static rule based workflows for years. No LLMs needed. As we all know, we've had classic sort of if this if this then then then that chat box. Right? I can follow rules, but they can't adapt. It's been around for ages. And you do have to think of every scenario, every question. That has to be predetermined. When we, of course, got large language models, now you can apply, apply these large language models, apply AI models to single step tasks, and that prompts the system to generate some output like a document, answer a question, all in one go. Great unlock. And while our language models, they still require the right access to your knowledge, they really don't need the precise instructions and, that previous automations needed, and they can just handle a wider set of tasks even without specific training. This is stuff we're all familiar with. Today, agentic workflows. So agentic workflows are great because they really differ from the traditional approaches by chaining together the multiple specialized AI agents as, Andrew Ng had mentioned. And they're training them together in a structured process that focuses on coordinating those agents, humans, and then all your distributed tools and systems into this structured process. And each agent plays a different role, so the workflow is really capable of breaking down complex projects and then iterating dynamically in real time. So it's nondeterministic. So for example, instead of a single prompt generating an answer from a chatbot, an agentic workflow, is the type of system that can look up your customer's order, your order status via API, determine and gather all that relevant information about the customer, and, then they can dynamically escalate the issue to the right human agent based on that information. And then they can even update your contact platform with a summary of the interaction for the call. So it's doing things end to end. It can solve more complex cases end to end because of that architecture. What's also great is when there is multiple specialized agents, there's less likely to be hallucinations. Each agent has one task to do, and it's not a thousand tasks per one generalized single do everything AI agent. So just less likely to, just less, like, less prone to error and hallucinations. And you can also enforce the policies at each of these steps. So just like with old school rules based automation, you're putting multiple AI agents in this more deterministic flow. So at their core, really, agentic workflows are more reliable, but they're also more adaptable. So they follow deterministic logic that you can really trust, but you're also gonna be able to lay out, layer in AI reasoning to handle all your messy and unpredictable customer conversations. So when to use both of these, just because agentic workflows and the term agentic seems to be the new hot thing, that doesn't mean that, we should never use single AI agents. You can often ask your yourself the question, like, do I even need a workflow for this? It's really about, solving your business case rather than thinking you're just gonna use AI. Right? Just helpful frameworks here for you to think about how to use each, each modality. You can think of your single AI agent as one really smart specialist, so great for ticket classification, drafting responses, high volume, quick inquiries, sort of your, like, what time do you guys close today? Great for one AI agent. It doesn't need to connect to all your other systems and take multiple steps. The other hand, your agentic workflows are gonna be better for multistep, multisystem, and, of course, because it's more accurate, maybe more high stakes cases, things like, account verification, refund approvals, billing disputes, stuff that is going to coordinate lots of customer information, PII, has anything to do with money, places where accuracy and control really matter. So to wrap, really, single AI agent, great at reasoning inside a single moment. Workflows are what let you connect those moments with other systems and logic. So lastly, before I hand off to Sid here, I really wanna briefly just mention how, agentic workflows are here at Assembled, how we approach agentic workflows agentic AI broadly. So we do take a hybrid approach. Again, it's all about finding the best AI solution for your specific support use case. So we can combine all three types of automation, single agents, and agentic workflows, or all of those in one single agentic workflow for, the best mix of predictability and possibility, I like to say. But here's a few other key pieces I wanna call out that make, our workflows our authentic workflows a little different. One, they are omnichannel by design. So whether that's chat, email, copilot, workflows on Assembled are easily created and then deployed on any channel just with the click of a button, which is exciting. Also, our philosophy at Assembled is that agentic workflows and AI general should be as easy to build and manage as possible for frontline CX teams CX and support teams. We don't want anyone to be blocked by engineering backlogs, by something that's too technical, and there are certainly no professional services needed here. And I'll let Sid get a little bit more into the show and not tell of that piece, of our urgent workflows, but super easy to build. Folks like myself who are not engineers can definitely make them. And finally, what really sets Assembled apart here, as I mentioned, is our unified orchestration platform. And what that means is where we can view and manage humans and AI in one place. We can also easily orchestrate workflows across your tech stack, as I mentioned, Zendesk, Stripe, all your internal systems, but you can tie all those things back to Assembled workforce layer. And you can really imagine all the workflows, all the agentic systems that that unlocks. It's super exciting. You can think you have workflows that can dynamically adjust, for example, in real time based on the capacity of your human agents. That's a feature we actually have live today. It's called dynamic handoffs. So it's where, you can adjust AI to human handoffs so that if your team is really swamped, for example, AI can focus on containment. When agents are available, handoffs happen more quickly so you can optimize for a better customer experience. You can even set up sensitivity levels. I mean, based on containment, standard containment levels, be cautious, focus on CSAT, whatever you'd like, and create dynamic rules and just adjust based on real time staffing. So really exciting possibilities with, the Assembled platform connecting to your human workforce. And without further ado, I'm just gonna let Sid get right into showing you how to start with workflows, how to get started with them, and then he'll jump in the product. Perfect. I think before we start making a workflow, there are a few very important questions you have to answer and identify what makes the most sense in terms of solving a certain user journey or SOP. And I think the the questions kind of range across four accesses. The first one is identifying what kind of volume or value you're trying to solve for over here. Does this necessitate a workflow is a question of the amount of volume that's going towards a certain category or a certain case. The next thing that is important is, can I solve it using an agentic workflow, or can my standard knowledge based q and a be able to solve it? And I think that boils down to the point of, is it a multi step, multi system sort of a flow, or is it just one question, one answer, and you're done with it? And I think that sort of boils into the part of designing workflows can get very complicated. At the same time, they can be very scalable. But you have to make sure you design with scale in mind, which means simplicity is at the key of it. And I want to illustrate some of those things as well in terms of how we design workflows and what are the key considerations to have while you're designing them. So let me share my screen and talk about what does this really mean, and let's do this. Here is kind of a visual view of what, agentic flow looks like just to talk about, hey. It's looking at multiple systems. It's fetching information. Is it reading and writing from one system to another? That classified as a very good use case for defining a workflow and designing something around it. So, think about it in terms of the agentic workflow itself is just the orchestration layer, and you have a bunch of areas underneath it. You have your rule based engines, which we are very eager to get rid of, but they do have their their time and place. And I'm gonna show you where that's very important. You have your knowledge, which you're using currently to train your humans. You can leverage the same thing in terms of training your LLMs as well. And then the last part is the humans itself and where do they have a part to play in this. So let's build a workflow over here, and let's take a use case, a very classic use case actually, which is for all of you retail folks out there, you must be having to deal with returns and exchanges. And that's something that we can automate, but there are a few considerations around it. So let's go step by step and talk about building this out. In any system you have, and while we are showing you Assembled, I wanna talk about some of the best practices and guidance that can be used across any system and outside of support use cases as well. So in any agentic flow, I think the first thing is what is the trigger or what is the intake criteria for it? So for example, from a service perspective, you can have that tied to a initial customer interaction. If this is flowing through a ticketing platform, you can attribute to a ticket tag or a type of category or some other value that makes the most sense to you. Or in the most easiest way, you can just write down what the scope of a certain workflow is going to be. So for example, I can just say, use this workflow when a customer is reaching out for the return or an exchange of product and is dissatisfied with the product. You can give the agent certain trigger conditions. These might be intake sentences that are used to train the LLM. And this is where you have the flexibility of how deterministic you want to be versus how probabilistic you want to get. So for instance, if I want to be fast and loose, I can just put down a couple of sentences to say I want to return product. I want a refund. I want an exchange. Now LLMs are smart enough to take these three, four lines and really use that as a baseline and expand upon it. You don't have to mention every single sentence that a customer might potentially say. Now if you're on the other side of the frame and you want to be really dialed in, like, say you come from a use case or an industry that is very compliance driven or heavy on regulations, And you tell me, you know what, sir? I'm not a very good prompt engineer. What do I do? Well, the good thing is you can always use AI to help AI. For example, over here, you can get expand this. Open. Trigger. You can always use something like this to improve your prompts. It's a very good way to get started where if you're not sure if the prompt is very effective or not, you can prompt a prompt and get a better prompt. And it really works. For example, over here, it's telling me exactly what's going on. I'm missing on certain details over here with respect to the amount of characters or not covering every single scope, and it's going to give me certain recommendations and revisions that I can add over here. I can choose to accept these or I can continue with what makes the most sense to me. Again, you have the flexibility, and that's a key area of designing and refining agent technology. So let me move on from here for a second, and then let's just update this step. Now in terms of a journey, you have an SOP. You have step one, step two, step three. In a similar fashion, you will define and design your steps over here. And you have a lot of options and flexibility in terms of defining what steps make the most sense to you. In assembled, at least, when you think about these steps, also have a key consideration of what's the modality in which this is going to be experienced. Now what I mean by that is if you're designing something for voice, you may want to have some conversational elements accounted for. Versus in chat, you can be very specific in the kind of inputs you get because people are typing it out versus they are voicing it over to you. Now how does that translate into what sort of steps you need to pick? Let me talk through a couple of steps, and then we can get to the next one. So for example, at the base level, you might want a step that collects information, and you can define certain fields in terms of what information needs to be collected, and you can just place it over here. That's one way of going around things. If your first step is just information capture, capture order details, you could use a collect information step. But maybe you realize, you know what? I think I want to start by drafting a response to a customer in terms of, hey. You know what? We're sorry this happened to you. What can we do to make it better? Because that's critical in my workflow where they're very customer focused and they're very customer centric. So it's important for us to have this. The kind of message you have can also be dialed in between the same access of being deterministic versus probabilistic. Let's say this was a health care use case. You might want to provide a very fixed response for something, or you could give the LLM in Assembled case, we call it CALP, certain directions on how it should be speaking, and you have the flexibility available to you. Now the other steps are more, I would say, level two in terms of reading and writing from different systems. You want to create something in Shopify or a case. Maybe you are not a technical person, so you don't have APIs available to you. You could mock your system integrations using a Google Sheet, and those are some of the more advanced things. A particular step that I like the most, which combines a lot of these capabilities, is what we call a guide step. A guide step essentially takes the analyze case, the reply case, the collect information case, all of it into one big text box. So think about it as a blank canvas, but with a blank canvas comes certain things you have to be mindful of in terms of setting it up as well. So let's call this tech return and exchange intent capture. And where in the world of support, what do we start our general support process with? We either apologize to a customer or we empathize with them. So what I want to start with over here is apologize and be empathetic to the user. My fingers do it. But item not working. The next thing I want to do is ask them what is it that they're looking to return, and I want to set their intent is to get a refund, or are they looking to exchange their product? Let's say they ordered a shoe. It's not the right size. Those are things you want to capture within your flow itself. So I'm gonna add over here, ask the user what product they want, and what is their intent? Is it to exchange, or is it to return and get a refund? Now let's go back to the thing we spoke about earlier, which is there are certain things that have their place, which is Google based engines in this regard where when you want to capture their product, you want to store that with a certain label because a jeans is a jeans, but a trench coat is a trench coat and a jacket is a jacket. You might have different product codes. You might have different journeys or SOPs associated with it, or it could be based on certain types of articles are not refund or return eligible. In this case, the way we handle that is by adding something what we call info or smart fields. So what we want to do is anything that's going to be part of an integration or certain area that you need validations that are very specific, you can create a smart field to say product type. You can add details over here in plain text, which an LLM can understand. But as a best practice, let's say, if this is part of integration, you can set this up as a drop down and add specific values. So when Assembled is calling a third party system and making an API call, we are exactly passing the kind of inputs that system needs. So what this translates into is you don't have to make changes to your source systems. The orchestration layer is smart enough to adapt and wrap around what you already have in place, which makes it easy to set up agentic flows on the current infrastructure that you have. Now with that said, let's say you design certain smart fields. You also want the ability to connect those to systems that you have in place. Let's say I have an order management system over here. I can very quickly come in here and configure an endpoint, and that becomes part of my verification process, which I will use over here. So for example, in this API, which I did configure beforehand, I just mentioned the order ID coming in, and I can quickly test it to make sure it runs fine. What is happening over here is I also have a sample database, which I prepared before the call, which has a set of products, their order IDs, and the return eligibility against them in terms of the delivery date and what is their condition, etcetera. So certain criteria and contours around what is applicable versus what is not applicable. That's the work that I did before. So everything looks good over here. Now let's go back to our flow and add those as steps over here. So instead of having this captured as product, I'm gonna press at, And this is where I get options in terms of the flexibility and the power that these guide steps or agentic flows have for you. You can pick exactly what field makes the most sense over here. So I might say, pick the product that they want to return. And instead of intent, I want to capture that as a return or an exchange. So this captures my field exactly the way I want it. The next step, obviously, is to authenticate the customer. So I'm gonna ask them for the order numbers. Ask the user to provide the order number, and and instead of using the order number like we spoke before, I'm gonna capture this as an input field and pass this number to an API call, which we configured beforehand, and I place it over here. So in just three lines, we're doing very powerful steps. One, we're acknowledging advertising with the customer. By capturing their intent, and the next step is we're checking if they are eligible for a return or a refund or not. So let's quickly save this over here. Publish our changes. And another best practice when designing flows is always start small, add a few steps, quickly test those out to make sure everything is working the way you intend to work it. It's much easier to diagnose if there is something going wrong at small steps versus you make the entire brush stroke and then trying to figure out what's going on. So for instance, over here, let's try doing a preview on voice. Hi. I want to return a jacket. I want to return it and get a refund. Yeah. The order number is 10892. Alright. Seems like it's working so far. I haven't given any more instructions or things to do afterwards. So I can very clearly see an audit over here and figure out, okay. In the previous step, I provided the order ID. It made an API call. It got details about my order, which was a leather jacket. So far so good. So that's one way in terms of By the way, sorry to interrupt you. I was gonna say, it sounds like your computer wasn't picking up the audio, but that was we could all see the transcript of the voice call there. All good. It looks like it worked really well. You guys can all imagine that the that was our voice AI agent communicating. It transcribes exactly what the, the voice agent said. So, and it looked great. So no worries about that. We also did have a question at at what point in the flow will I communicate the return and exchange policy? Looks like that's happening next. That's a great question, Ryan and Fernando, and that is what we're about to do next. And no good webinar is complete without a technical issue, are we? So let us talk about This is what happens when you do it live, baby. We don't fake it here. So now let's let's give us a set of instructions over here. Right? Let's let's go down to say, at this point, if, so let's say if the product is return eligible, then proceed to let them know that we can return or refund it. Got it. That's one way of doing it. There are multiple ways in which we could do this. For example, if it wasn't return eligible, I might add a sub step over here, which would say if there's an error, an error would mean it's not return eligible or the return the order number was not present. I can let the user user know that it isn't eligible and provide the number again. So you can add different steps like this. Another very important point I want to talk about over here is two things. So let me just save this step for now, publish this, add a new step over here, and I'm gonna call it return eligible. I'm gonna add a couple of steps over here. So what I'm gonna do is if the customer is return eligible, I'm gonna ask them to provide what was their defect reason, ask them for their issue, and store it as a defect reason. So you wanna capture the defect reason over there. And in that case, if I could go different step and say, you know what? Ask them to provide an image of the product as well. Match the defect against the product. If it is correct, then move by it. So this is the typical journey I might add over here in terms of being return eligible. Another important thing I would add over here is if the order number doesn't match two times, then move to I will reference a new step, and this becomes very important. Many of times when we create workflows, we think about 100% automation being the goal. The important part is the automation will be an eventuality. Always keep great customer service at the forefront in terms of the North Star metric you're trying to optimize for. So always when you're starting with, have a negative sentiment handoff step in place. So at any point in time, if your customers are spinning cycles with an AI agent, after a certain point, that is not a good customer experience. So what you want to do is at such junctions, you want to make sure you are capturing the sentiment of the customer and handing them off to a human agent. Start with humans in the loop. As you see your automation rates going up, that's when you should come back and decide if you really still need a human in this place or not. So I'm just doing a quick time check. And just like any good cooking show, this is my initial bowl. I keep this for twenty four hours, and we come back. This is more of a finished version of how things would look like in terms of you see my steps, and I will walk you through it in terms of the question of when do we talk about the return and exchange policy. So over here, for example, I did the same thing of double checking with the customer if they have an order number. If not, having them getting escalated to a human agent. If the product is return eligible, let's move to the eligible part. If it is not return eligible, then have a next step journey for what to do if the reply if the product is ineligible. And a few more things that I've accounted for over here that depending on the channel and the medium, you should also account for certain steps. For example and this is totally out of it. But if channel was phone and if I was capturing the customer's name, I would ask them to spell the name out. And that's a very important step because on phone versus chat, you have that medium difference, and you can account that when you design your workflows itself. So now what I did over here is I had those fleshed out steps for somebody who's return eligible. I'm asking them what the delivery date and what the defect reason was, and I'm asking them to upload an image. Because, generally, again, in a return flow, you want to validate against the picture of the product if what they're saying adds up or not. This is a very powerful step because this is where traditionally a human will be involved, but you can replace this with an agent as a level one check. If both of them match, proceed further. If it doesn't match, you can explain based on the return and exchange policy guide. So just to take a small turn from here, I have a little article over here which explains the knowledge, which has a knowledge around what the return and exchange policy is. Why that is important is because we can take customer context, which is the personal information, marry that with your company context, and then deliver very contextually relevant experiences. And that is really unlocking the power of these flows in terms of the journeys you can create. Now let's walk through a demo and example of this. So I want to return let's do a table for this example. And now if I see for my table, my number is 11245. So, again, the agent starts by apologizing. It's empathetic with me. I want to get a refund for it since I wasn't very clear in terms of just the return as a exchange or as a dough. And if you see in Assembled and generally with any good agentic flow, system that you work on, it's very important to know what's happening behind the scenes. So you want that visibility of every step and what's the rationale and the chain of reasoning behind it. So asking me for an order number over here. My order number was 11245. It's detecting that my sentiment is fine. I am not a malicious entity. And now it's making an API call to a check. Is the product number correct? As you can see over here, these are the various steps that it's running through. And seems like it found a table, and it is eligible. So it's asking me to move to the next step in terms of provide me the delivery date for the order. The delivery date was October 29, and the issue was that I got a table with three legs instead of four. So it's pretty damaged and defective or maybe a different product. Alright. So now let's try uploading a picture over here. Yes. So you can actually send a picture in. You can have your customer send in an image, which is great, to see if it's actually defective. Oh, I really like and I mean, it has alluded to too, but we've heard from a lot of folks that, you can't see into the decisions that were taken that the AI reasoning, it's often like a black box as to why AI made a certain decision or a workflow turned out the way that it did. It's great with Assembled, you can QA every single interaction and see each step that the agent took. So for example, you you can see it's clearly seeing the image and realizing maybe I'm trying to scheme the system. I put up a picture of a table with all four legs. Alright. You shall not pass. Now let me pick a table that actually has three legs. Let's say, this is the one. And while it processes state, I'm gonna just switch over and look at some of the questions we have in our chat. Can you go over a flow that requires a digestion of large amounts of data? Basically, looking for how to chunk and preprocess the information from disparate sources. Alright. That's a great question. I was late to this, but this doesn't look like sales course. That's an interesting one. Okay. So, Ryan, let me answer your question in just a second. The answer is yes. It is possible, and you can create sequential flows in terms of look at one disparate system, fetch the ID, take the ID, pass it to another system, and fetch something else. So just give me one second, and I can show you an example that speaks to it. But just to close the loop over here, as you can see, the flow goes through the different steps. It makes another call to the system that has details on the product amount, a separate table, and it process the refund. And I put a very specific channel instruction earlier, which is because we are on chat, provide the return link as well. And you can close this over here. Once you are happy with your workflow and your Assembled journey that you created, you can choose to enable it for a different channel and see which one you want to enable it for. So that's kind of a high level view in terms of how do you create an agentic journey. This was again, a very simple one that we can cover in twenty minutes, but I want to talk about, I think, three key areas that are important. I I think the first one is it's very easy to build and deploy. You don't need engineering resources. You just need the starting point for at least endpoints. But for the most part, we've seen across our customers where their teams are able to create these for themselves and have frontline teams with more business context set up their workflows. The second part is it's built on transparency and trust. So you need to know what's happening behind the scenes to improve and objectively make it better. And even simple flows can be really powerful. So that's, that's that's one part that goes into it. Now let's look at questions. Give me one second. Pulling up a couple of things. So, Ryan, this might be, he's not exactly speaking to what you're asking in terms of looking for how to chunk the information. But in terms of taking information from different sources, this is like an appointment, reschedule flow where we can pick based on what the user is asking for, if their appointment exists or not. So we make a check over here. And then once we get that information, we're able to check for available slots from a different table and then confirm that appointment. So it's kind of three different data sources. One with the user details, one with the appointment information, one with the availability slot of your providers, and combine all three into one contextual conversation. Is that kind of a direction of your question, or was it more to do with your systems itself have a very large quantum? Can an agent be used to split those up? I see. Might be helpful for maybe us to just talk a little bit offline on this. It's a very interesting use case, in terms of, like, to be able to split data, and then I want to also talk to you about what's the action actionality, like, what's the action that you can take with it. Would be helpful to talk offline about it. Yeah. Let's definitely follow-up with you, Ryan. Mhmm. Yeah. Let's follow-up. Especially in the interest of time, do you wanna head back to chat a little bit about, some of our workflows in production, Sid? Anything else to add here? Absolutely. Great. I think I covered go ahead. Definitely. Pricing. Yes, Maggie. We can follow-up with you here. Let me just share my screen. Oh, I think, Sid, you just gotta stop sharing, and I can add your perfect. Thank you so much. Amazing. I think a lot of these demos definitely can be, a little hand wavy. You can make Assembled workflows pretty quickly, and I wanna just give some, some info on these current workflows that are actually handling cases in production at a pretty large scale as well, across industries. I mean, we know, Honeylove, if you're familiar, a d two c apparel brand, super obsessed with customer service. We are resolving 18,000 cases a month for Honeylove automatically. They're actually a voice and chat customer. Please, if you're interested in our voice product, check out just call Honeylove today. It's a great voice agent. You can see it in the real world. Canva, most folks are familiar with, very high growth and exciting company. We are, automating their workflows. Oh, and I forgot to mention, Honeylove, we are handling workflows for returns and exchanges, among others, but that was their most complex high volume flow. Canva, they have a ton of, resolutions that they need, actually automated for their copilot their agent teams. They, still wanna keep folks in the loop, humans in the loop. So, we are automating a ton of, flows for Copilot, a lot about refunds, print quality issues, order tracking. Then finally, Rappi, if you're unfamiliar, this is the DoorDash of Latin America, huge super app over there. We are handling nearly a million tickets a month total for Rappi. And, really exciting for for Rappi. We're automating a 150 k resolutions a month. That's just for one workflow for their suspicious charge tickets. So really exciting. A lot of their agents have been totally freed up, from a lot of these manual and admin tasks. Finally, what's sort of the business impact of this? Of course, we're automating a lot. People care a lot about that, but how do automations actually transfer your business? What's exciting is agentic workflows unassembled definitely translate to real impact very quickly, into operational impact. Honeylove right now seeing 930 k in savings. Actually, a six x ROI since beating their investment with us, also maintaining their 95% CSAT scores. Canva, automating twice as efficiently. They were handling 30 to 40 tickets a day. Now it's 50 to 60. Rappi, story is also about ROI. In a quick turn, they're saving hundreds of thousand, almost a seven x ROI, and they achieved that in just five months. And you can actually stand at workflows like Canva piloted with us in just three weeks, and then they scaled to hundreds of agents, for Copilot in just four weeks. So we get things done fast here at Assembled. I know we had some time for questions here, pricing. Maggie, let me also follow-up with you about pricing, so we can make sure it's worth it to explore further because it does depend if you're interested in chat, email, Copilot, which channel you're interested in. We'll make sure to follow-up with you after the call. I think we tried our best to get into thirty minutes, questions included throughout. I think we did, almost to thirty, but I think we got it all in there. Right, Sid? Anyone else? We will follow-up. Great. We'll follow-up with any other questions via email. Again, thank you so much for, taking the time out of your day to sit here with us and chat a little bit about Assembled workflows, about agentic workflows in general, leaving you here with a quote from one of our customers, Leslie Ong at Flex Car. And, also, just to say thanks again for spending time with us. You can always, check out Assembled.com to learn more. We've got a page about agentic workflows there. We also have another webinar I'd like to plug for you coming up November 19, which we are, hosting in partnership with the, SWPP group, the society for workforce management professionals. And then we'll have a a voice agent deep dive, if you are interested in our voice product. So, again, thank you all so much for coming. We hope you have an amazing rest of your week, and we will follow-up. And please do visit us, and happy to chat to you anytime about anything. Thank you all. Thanks, everyone.