Name: Demo: Assembled AI Chat Agents
Uploaded: 2025-07-17T21:16:05.618Z
Duration: 1 h 8 s
Description: Demo: Assembled AI Chat Agents

Transcript for "Demo: Assembled AI Chat Agents": Hello. Hello, everyone. Good morning or good afternoon. We will wait for a few folks to join us here, but for anyone that is here early, good morning, and welcome to our live demo session. Thank you for being here. Please feel free to sound off in the chat about, where you might be joining us from. That's always fun. We are here in San Francisco this morning on an unusually nice and warm June day. Wait a few more minutes. Your hopes to join. Few seconds. Great. Greetings from New York. Amazing. Oh, you know what? I'm just gonna get right into it. My name is Cassandra Stumer. I'm a product marketer here at Assembled and also serving as your emcee today. Today, I'm also joined by Brian Yeh, resident Assembled expert, who is gonna take us through a live demo of Assembled's AI chat agent. By way of agenda, we'll provide a brief intro, to Assembl for those who may be unfamiliar. We will also talk a bit about what we're hearing from current customers. Brian will then just jump right into the product. He'll run through how to set up your chat agent, building workflows, maintaining quality, and measuring results. Finally, we are excited to close out the session with a bit of audience q and a. So if you do have questions throughout the demo, go ahead and pop them in the q and a chat on your right. We will make sure to get to them after the demo. So a bit of background on us, if you are unfamiliar. I know we have a few current customers in the crowd, but, Assembled is the only platform that unifies your in house, outsourced, and AI agents all in one place. Our customizable AI agents and workforce management tools provide everything you need to resolve issues faster, operate smarter, and deliver next level customer support. We've been at SportDAM since 2018 and have about a 120 employees with offices in San Francisco, New York, and we actually just recently opened an office in London. We partner with 300 of the world's most interesting and innovative companies to help them deliver world class customer support. Just a few of them here. And with that, I will hand it off to Brian. Hello, and good morning and good afternoon. I saw, a Berkeley shout out in the chat. I love Berkeley. My wife and I both met there, so very near and dear to my heart. But the rest of you folks, it's great to chat. My name is Brian. I'm also based in San Francisco, and I'll be walking us through a demo of our AI chat agents. And for those of you who have joined, I am sure that some of you are, early on in your journey, mid of your journey, or some of you have actually developed a lot of, you know, quote unquote chat bots in your time, and you're trying to see what's out there and new. Always a little bit of a challenge to span all of those different sort of stages of your journey. So what I'll try to do throughout this demo is call out, here are the things that we think are table stakes, right, got albums, and then we'll sort of evolve into things that I think make the next generation of AI chat agents really special. And along the way, I'll also try to call out things that I think we uniquely do here at Assembled, but also things that are just broadly part of what we think is the next generation of AI chat agents. So, it's gonna be hard to do everything for everybody, but I will certainly try to call out those places where hopefully, your ears in particular, might be perked up. I always like to start here, and this is now a little bit of a soapbox moment. I apologize in advance, but one of the things that having spent a lot of time working with support leaders that is really frustrating to me is that because now AI has come into the picture, all of the sudden, everybody's an expert in support. It drives me absolutely bonkers. I hope for the those of you in the room who, are in support, you you understand where I'm coming from now. But I always like to start with just because it is an AI technology does not mean that people who know that AI technology understand what it means to deliver a great customer experience. And you see it all the time in the headlines. It's all sorts of versions of, hey. You know, we can completely stop investing and increasing our headcount for support because AI is gonna do all of it for us. And then you also get the flip side if you see really great companies out there, really successful ones who say, yeah. We did it. We got rid of 700 agents. And then literally within the man you know, within a month, the next headline comes back and says, actually, you know, what we actually needed to to hire a lot of those folks back. And then you also see this sort of headline that goes, well, actually, no. No. It's actually really good. Maybe we can do it again. And it's a little bit of whiplash. And my personal take is that the folks who have lived and breathed support really are getting it. AI is going to be another amazing tool in your toolkit, but it is not going to fundamentally change what the ultimate goal is, which is how do we create a great customer experience, you know, at efficiently as cost as possible, but without sacrificing that sort of goal number one. That's, I think, and I hope why a lot of us got it to support in the first place. How does this relate to what we're gonna talk about today? Well, I I do wanna sort of call out one of my very favorite customers. I work with a woman named Leslie, over at Flex Car, which is an awesome service. They're based out of Boston. And I think, you know, Leslie is one of those leaders that I just so, so respect because she has been in support for so long having spent time at FAIR and other great companies that you've heard of. And one of the things that she has talked about, and she's been working with us for a long time on the workforce management side of the house. By the way, as Cassandra mentioned, we also help companies figure out forecast, schedule all their human agents. But one of the things that she's talked about ad nauseam is, Brian, I'm never going to compromise, my customer experience just because I want to use AI. I wanna make sure I'm using it and deploying it in a way, that is additive, and I wanna be really thoughtful about it. And I love the way that she rolled out a FlexChar with Assembled. So starting with using our workforce management tooling, she part of, like, really had a data driven approach to understand, okay. Well, where am I potentially understaffed? I happen to see that I am understaffed in our sort of, quote, unquote, twenty four seven kind of support. For those of you who don't know, FlexCar does field queries from folks that have used their service. They have a car, and they're having an automotive issue late at night, midnight, 1AM. It's a great service, actually. Well, it's really hard staff for that. And she used it. We've had to have that insight to go, okay. Now here's a place where I know that I can thoughtfully deploy a chat agent or actually they're also exploring phone AI agents. And she stepped it in and basically sort of segmented out her universe and said, hey. You know what? There are customers that are soon to be customers. There are folks that are already customers. We need to treat them differently. I'm looking for an AI solution which is smart enough and flexible enough to handle these two kinds of people differently while also taking on actual agent actions. So for instance, can I take in a set of, you know, descriptors of what happened with my car and not only say something nice or here's a support article, which I think we're very familiar with with in the world of deflection? But, actually, great. Now that you've given me this information, Cassandra, I can go ahead and update your profile and even then go ahead and set off another ticket to someone else who can actually call a tow truck to come get you. I just love this because I think she has such a good handle on where do I deploy it, what are the great use cases, and how do I find the right AI partner who can actually help me take actions. And the results sort of speak for themselves. She actually actually give me a little sneak peek of the presentation that she gave to her board of directors who's always asking, what are you doing with AI? And I just love the way that she had a really data driven approach tying it all together and really is an example to me of a support leader who's leveraging AI versus, someone who wants the AI as a hand you know, use AI as a hammer and and find any sort of mail. If you have any questions, please let me know in the chat, but it's just a wonderful, wonderful use case. With that said, I am sort of going sonic speed because we have relatively limited time, but my colleagues will be helping me check the chat. So if I am going too quickly or I've skipped over something that is interesting to you, hopefully, this can feel more like a one on one conversation. Please leverage the chat. We can go there. Okay. I'm going to start pulling into our own demo instance here. So, again, because we are a support operations company, a lot of the verbiage that I'll use will or when I talk with our chat AI agent about, will be, regarding workforce management and support writ large. So first and foremost, I'm going to start with, as I promised you, what I think should be table stakes things. Okay? If you're going to invest in an AI chat agent, there are just, like, gotta have them where if the vendor that you're thinking about or if your internal team is, like, creating their own tool because, again, everyone thinks support is so simple, we're gonna start with things that are, like, nonnegotiable, and then we're gonna build up from sort of the art of the possible into what's actually possible today. So first and foremost, this is a quick intro to Assembled. We are in what's called our chat testing environment. So it's essentially like a sandbox where you can try a bunch of different settings. You can play with your AI agent and just see what would it be like if you actually did release this out onto your website. Or for some of our customers, they're actually deploying these AI agents through their own app directly. So they're using Assembled as the brains, but they're not using our chat widget, because they have an amazing in app experience. We can actually push everything that you're seeing through that in app experience as well. Okay. So gotta have them. Well, let's just have a chat with our AI agent, for the simple things, and we'll keep going from there. So first and foremost, any chat agent worth its salt, so to speak, should be able to answer what we call knowledge questions. These are things that, like, what are your opening hours? You know, can you help me understand exactly what you do? So for us, in our workforce management world, we might say, hey. You know what? I am new to workforce management, and I'm having a hard time understanding adherence. And adherence is a metric in workforce management which describes, for your scheduled human agents, if they're supposed to be taking phone calls from ten to eleven, whether or not they're actually taking phone calls from ten to eleven or if they step away from their desk, their adherence number would go down. And, again, I would say this is a straightforward question. Any good chat AI agent, you should be able to load up and get these questions correctly because we actually have great knowledge sources. We have a page called schedule adherence. So this would be the closest to your older school sort of deflection chat. Now we're gonna start to take, like, the next couple steps deeper. So I could ask follow-up questions to ask the LLM or our AI chat agent I named him or we named him Cal. We can ask questions about Sernacle. You know, I'm still getting a little turnaround. I think you need to give me the exact formula for adherence, please. And, of course, we also have this information in the knowledge. So it'll go ahead, and, you know, if you've ever played around with chat GBT, something that you would get that looks something like this. Great. We have a formula that looks right. Now that next step deeper. What about something that we don't have a specific help article for? So we can say something like, how is this different from utilization? Aren't they the same thing? I would be curious to know if we have any folks who sort of spend the time and days doing this, but they are quite confusing for someone who is outside of support. Like, what is the difference in these two things? We are sort of getting into a world now where we're taking, like, a little bit more complex steps. Adherence is different than utilization, but there is not a specific article called adherence versus utilization. In the old world of chatbots, this is where immediately we would start to see some breakage. If you haven't literally put in a note that says describe the difference between utilization and appearance, you're probably using some sort of keywords. You're actually not going to get a good example out of this. And we can keep going down this rabbit hole. We could use things that describe it here as an art keywords. I could put in typos that will totally understand my intent and get those questions asked. But before we get there, I want to quickly call out a couple of things here. And this is now getting a little bit away from the table stakes. When you are using AI for your chat agent, in my opinion, the number one scariest thing is not having visibility into the reasoning why our AI agent actually gave us this answer. And for this specific one, again, we're calling it a knowledge based question. We're sort of, and I'll show you exactly how we build out this logic in the back end. But Assembled is checking to see, okay, are they looking to be escalated to a human agent? And it turns out, no. And here's why. And if you read this, this is actually assembled itself describing why I'm answering this question and not immediately escalating it to an agent. We're gonna talk a lot about escalations today. But my point here is we want ultimate transparency. This is going to be even more important when we start to talk about, quote, unquote, agentic things, things where the AI agent has to do more reasoning and take actions. Again, the number one thing I would be fearful of when introducing an AI agent is if we don't have that amazing level of transparency, if I get a weird answer at the end, I can't go back and debug and figure out, oh, is that a knowledge thing? Is it my workflow that's unusual? Did I ask it a weird question? Everything in Assembled, when it comes to AI agents responding to you, will have full transparency as to why, which is the key to making sure that you feel confident when you roll these out into the wild in front of your customers. I'm gonna put a pin in this one right here because I wanna show you an example of how we get to much more complex issues. But I also wanna start with, again, what I think should be table stakes for your AI agent, the one that you're choosing. And we're gonna start with two things. One of we're gonna start with knowledge. And before, you know, I can already hear it in my brain. The very first thing that we hear whenever we introduce AI agents is, you know, are we ready for this? My knowledge is a mess. I've got a six month project to bring it over from Confluence to Guru or Google Drive to Guru to Confluence. There is not a single support operations leader who says my knowledge is perfect. I don't want us to let perfect be the enemy of good. With a good enough AI agent vendor, and I think we do a really good job of this, you gotta pick a partner who is able to be flexible on how they bring in your knowledge because the truth is your knowledge lives in all sorts of places. With Assembled, you can sort of upload them, bulk actions. We have integrations out of the box with Guru, Confluence, Google Drive, Notion. There's not anything that I have not been able to sort of bring into our knowledge, base. And we're also gonna talk a bit about knowledge mapping. You know, one of the very first things that people will ask, especially if they have multiple brands or multiple skews is, how do I make sure that the agent is picking up the right knowledge to even answer the questions? So, again, your AI agent vendor should have a very clear answer to say, hey. You know, we can bucket out our knowledge to help the AI determine what kind of answers or what kind of knowledge we should be grabbing to get those right answers. Again, just things to check for as you're going through this. You don't need your knowledge to be perfect, but you gotta find a partner that's willing to work and meet you where you are with knowledge and help you enhance that knowledge. We'll talk a little bit about that in in the future. Now we kinda get to the fun stuff. This is one of my very favorite examples. We use something called, a style guide. Your team definitely has one, especially if you use outsourced agents. And the key with style guide is in support. If any of you have spent time at support, you will know that almost as important as getting the right answer is having that answer communicated to your agent in a way that makes them feel great. Or if you're a leader on the call, it's another way to say, how do we make sure our AI agent is showing up for my customers the way I want them to show up? And so our style guide is designed to, number one, be super fast to set up. So we have all these radio buttons. We can make the responses more brief, more elaborative. We have customers that, their support teams are supposed to be really, really fun, and they use emojis, and they have great humor in them. And then we have other customers that are in more regulated industries where, you know what, if you're asking about an insurance claim status, you know, we're maybe not using emojis to get that done. And you may or may not be shocked when you go through using an AI agent. Boy, this thing just doesn't sound like us. Well, you gotta pick a better that can help you get that agent sounding like the way that you want them to sound. And one of the things that we can do here is we can not only use sort of the prebuilt out of the box radio buttons, but we can actually take your favorite agent's own messages. You know, it's that agent that you will love it if your mom or your dad or your aunt or your wife or your husband, if they had trouble with your service, you look like that's the agent you hope they get to. We can take those literal messages and bring them as an example from the AI agent to learn from, and that is a really, really discreet way, to have your AI agent sort of mimic or sort of act or sound like your very favorite agents. And the way that I like to describe this is, like, I don't want you to think that you're bringing on an army of faceless robots. I kinda want you to think, oh, I've got this a really amazing agent. What happens if I just gave her or gave him a million arms and legs? That's what I want. I wanna multiply that really amazing agent. I don't wanna sort of create, you know, sort of this army like sort of faceless, robots. I'm getting ping from my team, that we're running out of time. So I gotta get to, I think, what I think a lot of you are coming in for, which is what about the more complex stuff? Okay. And the way we think about automation is you don't want an AI agent to try to take 20% automation across all of your inbound chats or calls or tickets. Our point of view is that, yes, the knowledge based questions like I've shown you, maybe that represents 40% of your volume. Great. Take it out of the way. We're gonna do a great job with that. The second thing though is, well, a lot of the people chatting in, like with Flex Car, about 8% of those are people trying to create a claim. Well, now I need to teach my AI agent how do I handle this more complex kind of issue, But I wanna automate a 100% of that 8% of my tickets. And that to us is the way to get from, like, that 35%, you know, deflection resolution with just knowledge into now, oh, I can tack on this next workflow, which is gonna give me that next 8% of automation. And then we're gonna find three other workflows, which represent another good chunk of my volume. I'm gonna make workflows for those, and now all of a sudden, I'm sitting at 65% automation. It's not magic helping my agent get, you know, a little bit better and a little bit smarter. It's really your AI agent should be smart enough to get all those knowledge based questions, and now you're gonna teach it to three other kinds of things that make up a big volume. And the way that we do that is we use something called workflows. And you don't need to make a workflow again for everything. But a workflow is for that 7% of tickets like that insurance call that you get or someone calling to make a claim. And for us, what that means is you gotta have a way to, for those use cases, set up steps that your AI agent can easily follow in exactly the same way that you would teach that new college grad who's joining your support team how to do something. And so for us, this looks like, you know, if you're, like, a child of the eighties like me, an old wizard that you would use. And we have these steps. And what was happening is this is called a trigger step. So once assembled, reads an incoming chat message, it's going to look through all that list of workflows that I've just shown you, and it's gonna check these little trigger steps. And this trigger step is a way that you would describe, again, to a human agent. Hey. Select this standard operating procedure when someone's asking about managing their gym membership, whatever it is. And maybe some examples are like, I'd like to cancel my gym membership or I need help with my gym membership. I'm giving it an examples just like you would give real examples to a new support agent about what triggers this standard operating procedure. But instead of actually using that old school, you know, giant decision tree where you're looking for keywords and all of a sudden, you know, you have, like, a thousand nodes at the end, you know, some poor summer interns project, which always breaks as soon as they leave, Really, what we're doing is we're leveraging the power of Gen AI to say, great. Okay. The next general step, the thing that I would actually do as a human, is I would collect information from the person chatting in about their gym membership. So I'm going to use plain English to say, hey. Listen. Assembled. I'm looking to fill out a field called member name, and here's what it is. It's a member's name, and if they spell it out, go with that spelling. And I also need to collect their membership ID. And, Jeez, what is a membership ID? It's three numbers. If they give you less or more, remind them that it should be three. And what we're doing here is we're getting an agent to be smart enough to go ahead and grab that information and fill out these fields. And why do we do that? Well, we may have authentication. In our case, you know, we sort of look it up against a very simple database. This database can be accessed via API. We have Google Sheets lookups. But, also, we can now start to take different actions. So for instance, I'll skip down to one of these sort of discrete decisions that the AI is gonna make. But the AI is basically saying, hey. If they're looking for a cancellation, go ahead and offer them a discount. If they're looking for an upgrade, I can hit what's called an API endpoint to go ahead and process that upgrade, and I can get my GenAI to sort of, flexibly confirm what's happening. And you can sort of take all these steps all the way to the very end where you can actually allow them to take that discount. If they, you know, they decide they don't wanna take the discount, you can go ahead and cancel them anyway. And what's cool to us is that you can preview this workflow. And, you know, I know this is not the point of today's, you know, session, but the whole goal here for us is we're an omnichannel AI agent. You can make this workflow once, and you should be able to have that AI agent run through that workflow over phone, over email, over chat, or actually just have it set as a Copilot next to your human agents. And so for us, let's see what this looks like. I'm gonna reset this conversation. Just another way to sort of play with our AI agent, and I can say something like, I need to talk with someone about my gym membership. And what will happen is Assembled is currently walking through all the different SOPs that could possibly go through, and it's going to pick this one because of that trigger step that I just showed you. And so now we're entering the collect information step, and I can say my name is Brian Yeh. And here's where I think, again, maybe this is table stakes, maybe it's not. Assembly is going to be smart enough to say, hey. You know what? I asked for two things, your membership name I'm sorry. Your member name and your membership ID, and you only gave me one. And I didn't write another step that says, go ahead and check four times to see if I can get the membership ID. I set it one set of instructions like you would to a smart human agent, and it's not letting me pass go until it gets that information. Oh, sorry. It's 4567. And, again, the rule is still set up. Actually, the ID needs to be three numbers, not four. Sorry. It's 456. And it's going through these checks now where I'm actually looking up my membership, ID against my member name, and I'm going to see whether or not I've been authenticated or not. And so in this case, I've been authenticated. I can say something like I need to cancel my account, and it will walk me through this flow. It's offering me a discount as this step says to do, and I'm gonna say, no. I just need to cancel. Thank you. These steps are so quick to set up. My my favorite example is we work with a really large South American company that do deliveries. And the truth is we set them up with one workflow. Basically, that's just part of what we do. We help you get, you know, one up and going so you can see the power. And two weeks later, we did a training with, like, 25 other people. Who are these people? They're actually support agents. And the whole session was designed to say, great. Let's brainstorm workflows that we should be building based on what you agents are seeing, and then a subset of them just went out and built it. And this is unbelievable. This is not some sort of, like, product team that comes in and does, like, a entire team effort. It is the people on the front line who actually understand what your customers need, are able to build out these workflows, test them, validate them before release them out to the wild. It is by far, to me, the most fun example of actually taking, like, some of the customer actual insights and then having a tool that is powerful and flexible enough to let anybody create these workflows yet have the controls to make sure that they don't go sort of wild with it, which is going to be how I spend the last few minutes here. Everything that you see in Assembled, we will go ahead and put into what's called a quality review page for you. And one of my customers, they made this great point. They said, listen. If we release a chat AI agent, I don't need to see a report at the end of the week that says, hey, you know, how do we do? That's not how CX leaders operate. It's because they just operate by saying, I need to know what my people are doing at every single given moment. And so what we can do is we can set up, this quality review page, and every single AI interaction has that amazing breadcrumb trail that I was telling you about so that you can, number one, see what the AI agent is saying, but also understand why it's saying that thing. So that if those support agents I was talking about in South America, they did end up, you know, a little bit sloppily creating a workflow, it's giving a wrong answer. We can actually go ahead and figure out why. And so what this is is this is our quality view page. I've filtered over to the chat agent. And you can see just, you know, a couple minutes ago here, I sent in a query about my gym membership. You can sort of see I answers I gave him. My name is Ryan Wang. I'm +1 234 or 4567. But what's cool is not only can you see what the agent said, we can say why? Why did the agent say that? And so we click down just two more steps here. You can actually see literally step by step. I've collected this information. I matched up Ryan Wang versus the column in that Google Sheet to make sure they're authenticated. You can see I don't know if you caught it before, but there are Boolean decisions you can make, conditionals. And it turns out that, yeah, I'd actually am confirmed. I'm authenticated. That's the step. And you can see this little green check mark. That's the AI saying, yep. Here's what I'm doing. Here's what I'm thinking. Here's why I went down this path all the way down to the user need. And then it go ahead and sent the next set of follow-up messages, and so it goes down further and further and further. We actually developed this based on a customer learning, which was, I think we started our journey in developing these AI agents. I think a lot of the ways that you still see a lot of, frankly, middling AI agents. You basically stick a bunch of instructions into ChatGPT. You say, chat g p t, remember this. And then I give more information. Now tell me what comes back. And it turns out that that might work for your more simple answers. In fact, that's kind of what we do for knowledge based answers. But for these really more complex step by step if thens, do only this if, LMS get confused. And the worst part about this is you set up this incredible prompt because you're so smart. I'm so smart. You got a weird wonky answer. And then you go, jeez. Where in my 50 lines of this paragraph of the prompt did I kinda go wrong? And we just can't have that as a CX leader. We actually have to break up these more complex issues into steps so that we have explainability and ultimately higher confidence in the answers that come out. The last thing I'll say here before we hop into q and a is we also need to chat about something really important, which is if you recall when I spoke about Leslie at FlexCar, there are some things that you still want to have handed off to a human agent. So a very common workflow that a lot of our customers start with is great. I know that if someone calls in, to tell us that they had a car accident, there's something wrong with their car, the first six minutes out of that twelve minute call is gonna be asking, what's your name? Oh, it's Lindsay Merrymount. What car was it? Oh, it's a Kia Sorento or whatever. What we can do is we can have the AI agent do all of that intake for you, but we still need to make sure that we are handing that person off to a live agent when necessary. And the last thing I'll say before I get off my soapbox is that I think a lot of human excuse me. A lot of AI agent providers are not really thinking from the lens of customer support. They're thinking about it like, man, how cool is my AI agent? Well, you know what? That's not how we think. We think, what does my customer want? And sometimes what my customer wants is to be handed off to a freaking human. Okay? So what we do is we spend a lot of time thinking about how do we hand off correctly. And for us, we can hand off these chats and messages through a bunch of different platforms. We can either do a live agent hand off. We can make sort of certain cases only hand off to links or emails or and we work with some companies for whom for certain kinds of tickets, there's nothing a human agent we can do. So you know what? We're just gonna ask our AI agent to politely keep them in the agent experience. You gotta get this right because the number one complaint that we hear from people who are switching off their AI agent is, man, when my wife, when my friend went to my website, it kept them in the called called a doom loop. K? And a doom loop is a bad behavior. What we can do is we can set up these hand off criteria. Containment focused, that's what I would call, like, the doom loop. Right? It's a very euphemistic way to say, hey. You know what? For these kinds of issues, we're gonna try to keep it in the chat. All the way over to really cautious. Meaning, if I start to even detect the slightest hint that you need to get to an agent and I am a concierge type support organization, I wanna tell the AI agent, go ahead and quickly escalate when human assistance would be beneficial. And if we had a little bit more time, I could actually play with this and show you what that looks like when we save the changes. So the very last thing I would say is, and this is where working with us, workers management is so important, we can set up dynamic handoff rules. And I know I'm clicking click quite quickly here, but what I've just done is I set a handoff rule that says, you know what? If my occupancy is over a 120% for my support ops geeks in the room, you'll know that's a very bad day for you. Basically, my people are way overloaded because I brought in my workforce manager metrics. I can tell the AI agent, hey. You know what? Everybody needs to be a little bit more than containment based route because there's just no human agent available to take you. And so we're going to increase that threshold that the AI agent has before they actually go ahead and fire this off to a human agent. To us, human agents will always be a major part of support. It is incumbent on the people in this room to determine what is that balance between the two. And, again, in my humble opinion, this is where support needs to have a loud voice. Because I think we've all seen what happens when the non support AI people try to do their thing without thinking about what you've known for years, which is that, oh, think about handoffs and how we do it is important. Handoffs are different between when we have people staffed and we don't have people staffed. Maybe if it's a twenty four seven service and it's 1AM, we actually need to change our handoff sensitivity to be more containment focused. But if we have people just sort of sitting there, they're not that busy because maybe we over forecast because we didn't have a several workforce manager tool. Great. Now we can see a little bit more aggressive to say, hey. You know what? Not everyone needs to be handled by the AI agent. We actually have Cassandra. She's a human. She's got availability. Now let's go to the handoff. And there's about a minute of the things we should talk about, that we think very, deeply about, which is that interplay between human agents and AI chat agents. But for now, I'm gonna put a pin in it, because I think the rest of the time is is is set aside for q and a. So, again, thank you so much for taking the time. I would love the chance to show this to you deeper if it might be helpful for you or share what we've seen from other folks. But for now, I'll sort of put a little pen in it because I'm I'm sort of getting the the Oscar music from my team backstage. Alright. Awesome demo here. Oh, yeah. Would you like to sort of do the summary here, but anything else to add here or or key takeaways? I think if I can give you some time, people will start to be annoyed. So I think I'll leave it here, as little screenshot if you like. Great. Yeah. Yeah. Happy to happy to answer any questions. Sure. Sure. So, we've got a few questions. I know we're a little quiet in the chat right now, but we've got a few questions we've seen a few times, in this type of demo. I'd I'd like to ask you for the group. So one of the top three, Brian, is, people often ask after seeing this just how quickly can we get this up and running? Are you true to us on that one? Yeah. Good question. The short answer is you can get your knowledge questions up and going in less than a week. That's just the truth. So those what is adherence? How's their utilization? And most of that week, by the way, is not spent, you know, adding knowledge. Most of that week is kind of the fun stuff where you're tweaking the style guide and sort of talking about how your AI agent shows up. That part in a week. There are other complex organizations. We work with a company that serves 35,000,000 customers. They are going to be building workflows probably for the next two, three months. You know, that is the more extreme case just because it's such a wide swath of agent activities that they want to do. But my my coaching is always this. With support ops, and if any of you have lived the life, you'll know, it is not always that easy to get engineering resources to put that, like, API endpoint and just stick it into Assembled. So what I often advocate is don't roll it out as a big shebang. Roll it out in the way where you get the knowledge answers going. You can sort of prove out that we have really high quality answers that sound great. And then you'll have the proof points to go sort of build out more and more workflows, which can get more complex when you actually do need that partnership from product. And that journey can sort of take a little bit longer, but I actually think we gotta get we gotta break through that initial inertia, get people confident with that. This AI agent is pretty dang smart, and then you can really start to ramp up the amount of workflows and and sort of custom actions you can take. You know, that process, you know, it could take a couple more months, but, you know, you shouldn't wait to have it perfect pre roll out. You should get started. Have Assemble be smart enough to say, hey. I can't answer this. I'm gonna hand you off. So it's pretty low risk. But it's like getting that ball rolling is is really, like, the the the first important step. Okay. Great. Thank you. Dylan asked, when can we use LLM prompts within Assembled to create those workflows? Of course, we then wanna review what it creates, but it would seem like a great way to get a rough draft flow. Wonderful question. Okay. Let me go back and share my screen here. So, number one is we've actually put a lot of thought into not only, sort of generating these workflows, like, sort of just writing your answers. But but, really, you can see there's a button that I didn't click which is called get AI suggestions. And what we can start doing is, like, say, you don't need to be, an incredible prompt engineer to leverage LMs. You're actually having tools or if you pick a vendor that doesn't have this, like, you really should think about this. It's sort of, describing in plain English what you want to have happen, and now we're using AI to help create AI prompts, which is probably one of the more fun things that I think we're able to do. And secondly, for those of you who still want to have your sort of traditional, creation of, like, pure prompting and using AI to sort of, be smart enough to separate all those steps. We still have them. They are called guides. This looks and feels a lot more like your prompt, your sort of, like, traditional LLM prompt. And there's a lot of cool features that we can do here. Again, I would say a lot of people start here, but for the more, like, regulated industries where you gotta have a lot more guardrails, I think the the the workflows are kind of a good option there. Okay. Deborah Brewster, how fast can you set up workflows and how many do companies typically have? You can set up workflows really darn quick, you know, just to give you a sense when we onboard people and Assembled one of their first spin up activities, is create a workflow. You know, Deb, here's like Deborah, here's one of the really fast ones you can create. It's like, hey. Create a workflow. Someone asked about my competitor. It's a three step workflow. Check to see that they're asking about a competitor, ask why they're asking, and then you can set up an answer that says, hey. You know what? We're better because we're omnichannel on the other folks' mind. That workflow can get set up in literally ten minutes. And then, you know, I didn't Deborah. Jeez. I'm so sorry. One of my best friends is named Deborah, so I keep saying Deb. Deborah. The other thing that we can do is we can set up more complex workflows. So for instance, I'm working with one of my customers to create a, a volume pricing workflow. They basically sell printed goods. And one of their things that they're trying to automate is instead of having people submit a request for quote, I need a thousand of these books printed or whatever it is. Actually, what our AI agent does, we brought in their price book, We brought in their volume pricing discounts, and we created a step that helps their, customers to input how many books they're looking for, of what type. And then we can actually turn around an initial quote, which is unbelievable because for them, they realize that if it took too long for a quote to get back to their customer, they would actually have a chance of losing that sale. That workflow took about an hour for me to set up mostly because I can't do multiplication that easily. So that was more about sort of testing it out. So that would be like an example of complex workflow that maybe took me an hour to put together. In terms of how many, that is totally you know, three, four, five. And you said extra you know, we also have folks who are building out dozens of these, especially in that example where they sort of federated out the the workflow building. Dan, can the workflows be set up to automatically encrypt data by default? Okay. So how we store data and who owns what is a really, really long conversation. Dan, here's what I would say. There are a lot of different things that your AI agent, vendor should be able to do. Number one is there are tools out there, many of which we use, which can redact certain kinds of information that your customers, for whatever reason, are sending to your chat agent that they shouldn't. Then there are ownership and deletion and retention policies that your AI agent vendor should, like, walk through with you. It's pretty common practice now, although you still get the question all the time. The major model providers, Anthropic, OpenAI, they do not train on whatever data is being used with them. That is just like part and parcel with their services. So there's the question of, like, is my data going to be used to train models? The answer is almost entirely or really should be entirely no, especially not with the ones that you've heard of. The second one's about storage and encrypting. Again, encrypting data is always something you should review with your SaaS provider. They should be able to sort of hide their SOC twos, how they handle GDPR if you're living in Europe. But finally, with storage, this is a question that is actually quite near and dear to us. It is true that some people ask, hey. I don't want you to store any data at all. As soon as you process it, just dump it. Totally. You can do that. The only downside of that, potentially, is that you might wanna allow your vendor to hold up that data just for a limited amount of time. Thirty days, forty five days, sixty days. This is purely for operational purposes. If you want, to sort of look back and do more checking against why your LLM gave you the answer that it did or you wanna do more QA checking, you actually need to hold off that data for, like, a little bit longer. It's nothing to do with, like, the quality of the answers. It's more so, Dan. Your team should wanna go back and be like, great. What kinds of questions do we answer? You know? How do I well, we do on those. If you immediately dump the data, you're actually gonna lose some of that visibility. We've done it for people because we understand there are certain kinds of environments, totally get that, and we, of course, will accommodate it. But it it is a consideration. Great. And it looks like we've answered all the questions here, and we are just at time. So I just wanna say thanks so much, Brian, and thank you all so much for being here with us today. We know you've taken time out of your busy day to join us, so we really truly appreciate it and hope it was time well spent. And, just a bit of housekeeping here, we will send you this recording, in a follow-up email. And if you would like to get a custom demo or just learn more about chat agents at Assembled or the Assembled platform in general, you could do that in a few ways. You could just reply to that email. You can click the link here in the chat that I will send forward. There you go. Or you can just head to our website. And, again, thanks so much all. We just hope you have a fantastic day, and hope to see you all again soon.