Name: The AI voice playbook: from pilot to 50K weekly calls
Uploaded: 2025-12-11T19:06:05.356Z
Duration: 47 min 48 s
Description: The AI voice playbook: from pilot to 50K weekly calls

Transcript for "The AI voice playbook: from pilot to 50K weekly calls": Hi, everyone. Thanks for joining us today. We'll get started in a minute here, but I just wanna give people time to trickle in. So maybe for for those of you that are joining us right at the beginning here, if you wanna just drop into the chat, in this messages tab on the right hand console, where you're calling in from today. I am in our San Francisco headquarters where it is currently freezing cold, in the Bay Area. Georgia. Nice. Oh, a couple on Georgia. Oh, and maybe while I have you here, also, let us know, like, what what is the size of your, your support team? How many agents, are you working with these days? Oh, nice. We got a a full team here joining us today. Oh, alright. We'll give it just a little more time, as people are coming into the room. 34 agents. Nice. Cool. Alright. Why don't we get started here? So thanks again for joining us. Today, we're gonna be talking all about AI voice. So thanks for being here online. So by way of introductions, I'm Brianna. I'm on the product marketing team here at Assembled. And I'm gonna be joined today by Sam, our product manager who is focused on building all of our AI agents. So a few housekeeping items before we jump in here. Make use of the console on the right hand side for, any messages you wanna drop in, any, like, thoughts, feedback. Use that messages tab. Use the documents tab to, follow-up after to see some of our related content. We've thrown in a few links, to our website there with some useful follow on content. And then also make use of that q and a tab. So we will do questions at the end, but make sure to drop in any questions you've got along the way there, and then Sam and I will, stick around to answer them live at the end. So let's jump in. What are we gonna talk about today? So we're gonna start by talking about the state, of AI voice support, and then I'm gonna hand it over to Sam for a live demo of our AI voice agents. And then we're gonna end with some best practices, some examples from our customers, from people who have really been able to scale up their voice support effectively. And then, like I mentioned, we'll stick around for some live q and a at the end. So quick introduction for anyone that might be new to assembled. So hi. We are the orchestration layer that makes AI and human agents work together seamlessly. So through our tool, you can plan, staff, and manage all of your human and AI agents all on the same platform. Historically, a lot of teams have treated these as kind of separate worlds. So they'll think about, like, workforce management just for their humans. They use, isolated AI vendors for automation. But Assemble really marries those worlds and brings them together so you can see and manage everything in one place. Because we just think of them as two facets of the same team, not two completely separate teams. Then we're also connected to your entire tech stack through all of our integrations. So all of your ticketing platforms, telephony providers, your knowledge bases, your productivity tools, your QA tools. And then we've also got an open API across the entire platform for custom connections. So we really wanna be connected to your entire tech stack so you can use Assembl as a single source of truth. You don't have to be logging into all of these different tools to get the full picture of what's going on in your support organization. And so what are the actual products we've got? So we've got our workforce management products. We do AI Copilot for agents, and we and we also have AI agents for every channel. So chat, email, voice, SMS, social media, we cover it all. And all of these products work best together, but also each one can be used individually. So depending on wherever your organization is at, a lot of teams will wanna start by automating one channel at one time and then add on others later. Not everybody wants to immediately automate every single channel at once. And all of our products have this shared data layer. So your workforce management insights can make AI agents more effective, more dynamic, but AI agent data and Copilot performance data can help inform future capacity planning in workforce management. So kinda helps you balance that AI to human ratio across your entire support to save you the most money. But today we're gonna be focusing on this guy in the middle here, so our AI voice agent. So let's start by talking about what's happening across phone support lately because I think this channel is changing a lot faster than a lot of people realize. So one, phone support isn't going away. In fact, it's the opposite. It's growing based on serving our customers. A lot are expecting around a 20% increase in support calls over the next few years, and so they're actively preparing for that. Despite all of the investments teams have been making in more digital channels, chat, email, self-service, Customers are still picking up the phone for those complex, urgent, high stakes issues, so it really isn't going away. You wanna make sure you're prioritizing this channel. Second here, 71% of Gen z believes that live calls are the quickest and easiest way to reach customer care. So everyone kind of tends to assume younger generations only want digital channels, but the studies are showing that Gen z are just not afraid of the phone the way millennials, kind of are, and I am guilty of that. So, like, you know, Gen z grew up with tools like Siri and Alexa, Google. So they're really used to these types of voice interactions, but they also expect voice to be really fast, really natural, actually helpful. They are not fans of this old outdated push button, you know, IVR systems, but, really, who is? And so through this generation, they're really driving a resurgence in phone support here. And third, we say this all the time. Customer expectations have never been higher, and that's just never gonna change. A study came out that said 60% of callers just won't wait on hold for more than a minute. And a minute really isn't a long time. That's barely enough time to go look up an answer, to go do a transfer to a different team. So patience is kind of wearing thin on people. And so where does that put us? Here's the operational reality here. So phone support is going up along with customer expectations. And we know that phone support is just a fundamentally different channel from chat or email. It's more expensive. The stakes are higher, and it's harder to scale. One reason being, you know, one agent can't handle multiple calls at once the way they might be able to handle multiple chats, multiple emails. We also know that a really bad phone experience can hurt, your brand. It damages trust and loyalty much faster than maybe, you know, a delayed email or a slow chat can. So it's a pretty delicate channel. And because, like we mentioned, customers won't want to wait on hold. You just need to make sure, that you are balancing your human and AI capacity so that there's always someone there to reply to your customer and you don't leave your customers waiting. And so this is why AI voice works so well when it's part of a more coordinated system, not treated as a standalone tool because you just never wanna be in a position where you're understaffed with human agents and you have customers waiting around and your customers are dropping calls. They're not getting their issues resolved. But on the other hand, you also don't wanna be overpaying for agents, that are just sitting there without enough work to do if AI resolution rates are high. So you need this, like, orchestrated tool to keep you in that right balance, so you have the right ratio there. And so given, like, what we just covered, what does it actually take to make voice AI work at scale? And so we've been working with a lot of companies that are handling up to 50,000 calls per week and scaled up to that pretty rapidly. And so working alongside these customers, we've learned that there are some key requirements here. So these are gonna be the things that separate a really successful AI deployment from one that kind of stalls in pilot mode and never really gets to where you want it to be. First, you wanna look at intelligent routing. So you need routing that adapts in real time, not just like static rules you set and configure one time and they stay that way forever. Like, thinking about when should AI handle a call versus a human, that answer might actually change minute to minute based on agent capacity, the priority of the customer, the complexity of the issue. So you just need your routing to be able to stay, like, dynamic and agile, the way your team will be. Second, you want quality controls built in, not bolted on. So that means it's like you, you know, you can't have a human manually reviewing every single AI interaction, especially at the scale of, like, 50,000 calls per week. That's just impossible. But on the other hand, you also just can't risk an AI agent not handling a VIP customer correctly. You need to strike that balance. So you're gonna want real time monitoring, sentiment tracking, automatic escalation triggers all built into the system from day one. So you want to have those rules, those safeguards to ensure that you're always delivering on the experience that your customers expect without fail. Third, unified visibility across AI and human performance. So this one is super critical. So you're not managing the things as two separate operations. You wanna treat these as one blended workforce. You want to see your AI metrics and your human metrics all in one place so you can kind of optimize and keep improving the entire operation. You never wanna look at just AI alone, without comparing it with the rest of your workforce. And fourth, frictionless handoff to humans when AI reaches its limits. So whether you might have a case that's too complex for AI, a customer who just insists on speaking to a human, we just know there's no, like, world where AI is actually gonna resolve 100% of cases. So you need to be prepared for those scenarios where AI is gonna be handing off to humans. So that handoff needs to be absolutely seamless. If your AI agent is doing any work, if it's gathering details from a customer, if it's looking up data, anything it's doing, all of that information needs to be passed along in the handoff. So when your live agents are opening up the ticket because it's been handed off from the AI, everything needs to be summarized. The agent should never have to repeat any of the work that the AI did, and the customer should never have to repeat themselves. You just never want the customer to feel like they're starting over. I mean, I think we've all been there when you are talking to a chatbot and then you finally get to a human, you have to start all over again and repeat the same thing. You just want to avoid those scenarios. And fifth, you want one foundation that's gonna work across all channels. So this is where a lot of companies might hit a wall. Like, look at one channel working, and then they realize they have to rebuild everything for the second channel and then third channel, and it just, like, really triples their implementation time. And that's where a lot of projects really slow down. So, like, if you look at a lot of older systems, they're kind of made up a bun of a bunch of tools that have been acquired over time. So, like, chat and voice may actually just live in entirely separate spaces requiring double the work to get both of those things set up. So you're gonna want a system that's natively multichannel where you can configure your rules, your foundation, all your settings, your standard operating procedures all in one place, so you don't have to build separate sets of rules for every single channel and can get you to scale that much more quickly. And all of these things make up our philosophy for our product design and kind of inform the way that we build the way we do here. So I'm gonna pause for a second there, and I'm gonna put a little poll up just to keep us engaged here. So if you look over to the right side, you'll see a polls tab open up, and we just want to know from you all, what's your biggest challenge with phone support today, if you are running phone support? Some of you may not even be there yet, and that is a okay. So feel free to cast your votes. I'll leave this open for a little bit here. Oh, okay. Seeing a lot of, inconsistent quality across agents. Got it. Can't scale without adding more headcount. Not enough capacity. Yeah. All very valid, very common concerns. And I just wanna reassure you that a lot of people are earlier on in their AI journey than you might expect. So you are are not behind in any regard if you are feeling some of these pain points. A lot of customers are. So I'll leave that poll open for a little bit longer. Let's jump back into our content here. And so what has assembled actually built here? Let's talk about the product for a second. So our AI voice agent is helping teams replace their old IVR system, their old push button models, and we're getting people up to 70% automation without requiring any engineering. Super easy, super intuitive to use. You'll be able to build these agentic workflows that can help you personalize every single interaction to a customer, can automate all of your standard operating procedures. All of our agents automatically are enforcing your brand policies. Everything is new brand tone. Everything meets your bar of quality. This may be obvious, but I think needs to be said. We will not send out replies to any of your customers unless they pass specific quality thresholds. So you should never have to worry about your AI hurting your CSAT, and we're gonna be with you during implementation to set all of that up. You're never gonna be alone, figuring that out. Third, our our AI agents escalate to humans with that full context that I just talked about on the last slide. So when they're passing it off into a Copilot, all of that information is there and readily available. And then finally, you'll be able to support global multilingual audiences through our automatic translation. So the beauty of AI is that it really helps reduce language barriers, so matter no matter which languages your customers or your agents speak, you'll be able to handle a lot more customers. And so this is how we've seen some teams be able to expand more quickly, without doing a ton of regional hiring, and they're able to save a bit more money. And so with that, I am going to hand it over to Sam, who will be doing a demo of our product here. Hi, everyone. Alright. So just to give you a bit of preview of what to expect here. First, I'm gonna show you what a live real production call with our voice agent sounds like. So we're gonna call assembled, and ask to update our subscription level. And Cal, our conversational voice agent, is going to take action over the phone and process my request end to end. Keep an eye out for just how natural Cal sounds, how they use conversational cues to keep me on the line while they work behind the scenes, and how I can interrupt them seamlessly throughout the conversation. So with that, I'm gonna give Cal a call. Hi. My name is Cal, a virtual agent. How can I help you today? Hi. I'm calling because I'd like to modify my subscription. I can help you with modifying your subscription. For security reasons, I'll need to confirm your identity before we continue. Could you please share your full name and member ID? And since we're on the phone, would you mind spelling it out so I can make sure I get it right? Yeah. My name is Sam Carlson. It's s a m c a r l s o n. Could you please provide your three digit member ID as well? Oh, yeah. It's 111. Thanks for confirming. Your current account type is basic, which offers core AI agent functionality, basic analytics, and community support for up to five users at no cost. You have the option to upgrade to either the premium or ultimate tier. Which one would you like to upgrade to? Yeah. Could you actually just tell me what's included in the ultimate tier? The ultimate tier includes all premium features plus enterprise SSO, custom AI training, advanced reporting, quarterly business Okay. Actually, you know what? I know I need that, updated reporting or the advanced reporting. So could you just upgrade me to ultimate, please? Just to confirm, you would like to upgrade your account from basic to ultimate, which includes all advanced features on the Just go ahead and do it. Perfect. Your account has been upgraded to the ultimate tier, which includes all advanced features and unlimited users. Is there anything else I can help you with today? Yeah. You know what? I, actually would like to speak to a team member. I had another issue, but I need to speak to a person about it. I'd be happy to help you with that. I'm connecting you to an agent now. No. We support agents. Alright. Cool. So as you can see, Cal basically processed my request end to end. And now let's take a look. First thing I wanna show off is the handoff that Bree was alluding to earlier. So if I can share my screen here. If we look in our Zendesk instance, we should see a recent call where we are able to get a summary of the call that transpired. So we see the user who's there, their member ID, what exactly happened on the call, and then the search of a simple summary and the topic that makes it really easy for a human to pick up the phone. Oh, and I just realized I was sharing the wrong tab there. But, I guess now that we're looking at our workflow builder, let's take a peek behind the curtain and just see how Cal is set up to handle these types of calls. So first thing, everything with our voice agent is super configurable, and it's really easy to set up in a matter of days with no code, no engineering effort. And just generally, we focus on making our platform super self serve for support teams versus something that you need a big team to set up and deploy. The first place we're gonna start is with one of our agentic workflows. So Cal knew that it needed to use this workflow as soon as I said that I needed to update my membership. And you can define the rules for selecting workflows just using simple natural language, which you'll notice is a theme throughout most of our product. After Calab selected a workflow, it used this first procedure to figure out how to authenticate me. So these are just natural language instructions, that Cal could follow to execute the authentication process following sort of fairly complex instructions and also using things like this member information API to actually retrieve my account information and verify it live on the phone. After authenticating me, Cal moved on to this next step where it used these instructions to go ahead and, process my membership change. Again, it could follow these natural language instructions. It could access external help docs about the different types of accounts that we have and use that knowledge live over the phone to respond to my questions as well as another update membership API to actually take this action over the phone. Beyond our workflow builder, we also have detailed logs of every single interaction. So you can see the call we just had here, and you'll get a line by line record of every interaction. So we have this transcript here. And then if we scroll all the way at the bottom, you'll also have a recording making it easy to inspect any given interaction someone has with Cal. If you pop into one of these workflow run steps, you'll see that you get down to the API level detail of how Cal, interacted with the user in order to, address their, need. And then the last piece I'll just show off here in this demo is our voice agent configuration, and this is where you can configure basically everything else about Cal. So we have tools for just simply updating the greeting and handoff messages that Cal would use in order to, interact with a user. You see we have tools that we can use to configure the different voices that you might want CAL to have, as well as, tools for configuring, where CAL will route specific cases if you have multiple queues or destinations in your phone tree, as well as places where we can go to configure the handoff. So when to handoff, what type of information to capture, and things like that. I'm gonna hand it back off to Brie to take us through a few more slides, but definitely please post all your questions in the chat, and we'll answer them one by one. Cool. Thanks, Sam. Let me reshare. Alright. There we go. So let's recap really quickly what you just saw here. So, one, you saw production ready AI voice. So, these are examples of workflows that are live and running in some of our customer accounts. So you're seeing real quality controls, real time monitoring, real customer interactions at scale. Second, you saw our agentic workflow. So you're seeing how AI can take all those steps, looking up data, processing information, making decisions, and taking action, all while it's sticking to your business rules, all while always optimizing for the best outcomes. And all of this is happening in seconds. So you see an interaction that takes thirty seconds, and then you see all the steps it takes, behind the scenes to actually make those results happen. Third, you're seeing one foundation across all channels. So those workflows that you saw, they can work for every single channel. So you're configuring that one time. You're deploying that across every single channel. So this is how you can expand from voice to chat to email without having to rebuild everything from scratch every time. All the rules, the logic, all of that applies no matter which channel initiates that workflow. So the end result should always be the same for the customer, no matter how they want to engage our brands. And then finally, unified visibility. So you got a little bit of a glimpse into how, we manage AI and human agents, as one operation, not as two separate tools, and I'll get into that a little bit more in a second here. But this is what it takes to run AI voice at scale. This is how companies we're seeing, like, some of our biggest AI customers are saving up to 2,000,000 per year. It's really not just about the the fancy AI model, but it's really about the orchestration and building an intelligent operation. And so I'm gonna take a break for another poll really quick here. So opening up this one now. Hopefully, this one will be a little more entertaining. So if you can let's complete the sentence about why your customers call in, the most often. We'll give it a second for people to drop in some of their votes. No. I think we're equal across the board right now. Okay. Nice. Something's broken. They need it fixed now. Yeah. I would say that that's that's pretty common. High urgency usually lends itself to phone support. Alright. I'll leave that poll open for a little bit longer as you continue to drop in some votes, but let's move on here. So well, I did mention models. Let's talk a little bit more, about the architecture. So you get a little bit of a glimpse here. So someone has built with all of these top of the line models, to ensure that the output to customers is the best it can be. We are constantly evaluating all these models against each other, and we're always gonna default to the top performing model at any given time. So we've got a lot running on in the back end to make sure, that we are using the best performing model from hour to hour, day to day, month to month. So when you're building these workflows, building your AI agents, it's never on you to have to choose the right model for the job. Though we do give you the option if you so desire. This is one thing that you shouldn't have to worry about. You should trust that the answers are always gonna be good, that we're using the best model possible. You'll see here that every step of this case's journey, so from node to node here, there are multiple models called out here. And so that's because we never wanna be locked into just one model. We're not biased toward any one model. We just want to ensure that you've got the best possible outcomes, and that's why we're just monitoring and evaluating models in real time, because for us and for you all, it's really all about the end results for the customer. And now I wanna show you how this connects to the rest of the workforce management platform, with our smart escalation routing. So these are the controls you'll be able to set for when and how your AI escalates to humans. So at a very basic level, you can can set rules to escalate certain, maybe, ticket types, certain categories, certain types of users immediately. Well, other cases, you may want the AI just to handle, those types alone. So that's a pretty simple categorization exercise, but where it really gets interesting is the connection to your real time insights that you can only get from a workforce management tool. So if live agent queues are over capacity, you can set up trigger overflow rules to move tickets into backup queues. So like in the example you'll see here, tier two, their queue is maxed out. And because we can see that occupancy and that wait time here pulled in from our workforce management tool. So those incoming tickets now can be rerouted to tier three who is equipped to also handle those types of cases to ease some of that load. And you can set all of these rules based on a lot of metrics within the product. So, like, average handle time, SLA, first response times, and others. So when support volumes are surging, your high priority issues, you can make sure that they're still getting escalated correctly while your AI agents are working to contain, those easier cases. And, of course, you'll always be able to set hard constraints. So, like, specific queues may have rules that are hard and fast. You can't stray from those, maybe based on certain business hours, languages, or skills. You wanna ensure that those always get escalated the same way. You can do that, but you can really tailor when and how escalations should occur for any type of ticket. And once those rules are set, sort of becomes this, like, automated load balancer so your teams are never really overburdened. And you don't have to manually make these changes throughout the day because that's really a pain to have to move these around. And so you can trust that every case is ending up where it can most quickly and most effectively be resolved. So now let's talk through some of the best practices we've learned from some of the the teams we've worked with who are automating just massive call volumes per week per week. So they've kind of learned these lessons so you guys don't have to. So you can take their learnings and apply them immediately. First, you don't want to launch AI on every single call type at once. So you might feel the urge to try to automate a bunch of things, set up all your workflows on day one, and it just kind of creates chaos. Instead, we recommend starting with more of a lower, more of a rollout. So starting with, like, lower risk, high volume tasks where you know AI can do really, really well right out of the gate. So you wanna get those quick wins. Get those, like, really high increments of automatable tasks first. Knock those out, and then you can keep expanding into more specialized workflows. So the teams that do really well on this treat it like a rollout, not just like a big bang moment, and you're gonna continue to improve over time. So you'll keep uncovering more automation opportunities. Your knowledge base is gonna continue to grow and improve, and you're gonna be developing more workflows accordingly throughout time. So don't feel like you need to do everything, right out of the gate. And second, when you're setting up voice support, you don't wanna use generic static responses. So for example, one of our customers was getting maybe 20% resolution rate on one of their workflows. They were having some generic answers like your balance will update soon, but as soon as they switched over to a more dynamic authenticated response with real customer data, so pulling in a lot of that personalization. So for example, saying your balance of $250 will update within seventy two hours based on your last transaction, Their resolution rate tripled. So those highly personalized experiences should just never be discounted. Context, personalization really matter. Third, you never wanna leave your customers guessing when their issues are gonna resolve if you maybe can't do can't resolve them immediately. So we've seen a team who is able to reduce their call volume by 30% just by being proactive, setting really clear expectations upfront. So instead of responding to your customer that we're working on it, might say, oh, within exactly twenty four hours, these changes are gonna appear. Just really specifics, are very important here. So and then because of that, customer stopped calling back repeatedly asking for an update. So just being as proactive as possible, will prevent all these repeat calls that might clog up your phone line and keep, you know, your agents in repetition mode. So fourth, you don't just wanna track volume metrics. Knowing AI handled 10,000 calls, great. But it's kind of meaningless if you don't know how well it handled them or how much it reduced the workload on your human workforce. So high performing teams are monitoring this quality in real time. They're looking at sentiment tracking, looking at keywords, getting alerted to any VIP conversations. So, teams that are able to catch problems before they escalate can have this more proactive approach, because you just never wanna find out about a bad experience after the fact on Twitter or on some review site where it's out in public. And fifth, don't deploy AI as a surprise to the rest of your team. So similar to how we talked about managing your AI and humans together in a tool, you also want to bring your human agents along on the journey of deploying AI, even when it's an AI agent and not just a copilot. So the companies that are seeing a lot of success, highest agent satisfaction are making this kind of a team wide transformational effort. They're training their agents. They're transparent about what's changing and why it's changing. They're creating these feedback loops so their agents are really part of shaping the system. At, many companies we've talked to, the top agents are actually tapped in for QA, and response evaluations. So they become the ones who are telling the AI if it's doing a good job. They're the ones making sure that the generated replies feel realistic, meet their standards, before the tool is set live. So they're really involved. And in that scenario, AI just becomes a tool for them that can remove the tedious tasks. It's not a threat to their job security, and it's something they they can become very comfortable with. So you just don't ever wanna leave your agents kind of in the dark to worry about their future, when you're implementing an AI solution. And finally, here are a few examples, from some of our customers who have seen some major success. So Retention Express, they're a customer experience company really focused on subscription based memberships. They run the vast, vast majority of support through their phone. They're getting around, over 1,300,000 cases through assembled. And so within just a couple months of working with us, they've been able to already automate over 30% of their support, and they're continuing to increase over time. So super high volume team there. Second, FluxCar, they're a car leasing company. AI Voice was able to actually eliminate their overnight staffing costs for them while maintaining quality replies. So they've been able to reach 85% containment as well using, our AI agents, and they've been able to save a lot of money not having to staff humans overnight. And finally, DailyPay, they're a payment platform. Using a combination of our AI agents and workforce management, they've been able to boost their CSAT scores by 22% because they've been able to improve those response times. And they've also been able to save a million dollars annually just because, they are paying for the correct amount of staffing. They're not overpaying for agents, and so they're hitting some really great goals here. And so that said, we are very excited about the future of voice support here. We hope to see all of your logos up here on these slides one day, showing off some massive savings and productivity gains. And with that, we've reached the end of our content. So we can jump into questions, in a bit here. So thank you so much for joining us online, sticking around with us today. Just a few notes here. We will be sending out this recording shortly, so, keep an eye on your inbox for that. If you want to learn more, please visit assemble.com or click on one of the links, in the documents tab here. And you can always stay in the loop for upcoming webinars or if you're interested in more content at assemble.com/webinars. So with that, I'm gonna unshare my screen, and we can jump into, some q and a here. Alright, Sam. I have a feeling a lot of these are gonna be you know, you're gonna be best suited to answer them. Let's look at this here. Any best practices for companies that do not do call handoffs to a live agent? Maybe Sam, you probably worked with more customers doing this, but, I I would say that, like, when when people are thinking about doing that, they they might just, like, send them to a chat or, like, an email agent instead. Some companies just have these sort of, like, after hours, workflows built in where they can divert traffic to another channel, but they can give the customer a response in the meantime about what's happening. So, like, I mentioned about that proactive communication. You call in, you may expect an email back. I mean, I would say probably most customers want that full live resolution on the phone, but if you choose not to do that, there are options to still create a good customer experience. Yeah. Happy to jump in here. I think that we see a couple different flavors of it. So I think, Rita, one point, like, you can sort of follow-up with folks through different formats. So we have people who will follow-up either via usually via email, in, like, an async way, on some share of of of calls that don't get through to a person. But another sort of pattern here is people actually just calling back. And I think a huge part of it is just setting, expectations properly on the live call. So when someone sort of requests speak to a human, you're able to give them the information they need to, like, understand next steps. So whether it's someone will call them or contact them and sort of what the time frame is there, you have the ability to set this up, in the handoff message part of our voice agent, to sort of give the right instructions to the end user in order to set expectations about what will happen next. Great. Okay. What about QA? I think we touched on this, a bit in the demo and slides, but are there companies QA ing every single ticket? I would say definitely not. Early on, you will be doing a lot of QA to get up and running, but as time goes on, you become more confident in your AI agent. You're probably gonna wanna do some spot checking, forever and ever, but you definitely will not need to review every single case there. Yeah. We do have a feature I wanted to talk about here. So we use AI to actually review every single case. And currently today, this will sort of classify cases of whether the issue was resolved or not. But we are doing a big push to expand our tooling here to make sure that every case is tagged with sort of full detail and scoring of what the sort of user sentiment seemed to be, what the quality of the resolution was, how helpful was the agent, etcetera. So we'll use AI to basically QA every ticket for you. As Brie mentioned, it's helpful to have some human calibration of that as well. But, yeah, we we don't want folks to have to review every single case, manually. Alright. Let's see. What is the percentage of contained calls when there are integrations involved? If I'm understanding it correctly, I would say that I like, we're pretty, integration agnostic. Like, I I don't think there's any differences no matter which tools you're integrated with. It's kind of all comes down to, like, the workflows you build and the orchestration, to get that percentage of containment. So, we have teams that pull in a lot of integrations. They might use, you know, several per workflow, but, yeah, I I don't think the the integration necessarily influences the percentage of containment. It depends what system what you mean by integration. But I think one thing that we do see as a huge factor in sort of what resolution rates and containment rates look like is what sort of systems the AI has access to. So if you have an internal admin tool with a bunch of helpful data on users, the more that the AI can get access to that same information, the more effective it will be at resolving cases. So, yeah, whether you're on Zendesk or Salesforce Service Cloud, or some very different telephony providers doesn't make a huge difference, but the level of access you're able to get, in AI to your internal systems can make a really big difference in in how effective the agent can be. Let's see. Any you see over here, Sam, that you wanna take. How do people do CSAT when using Assemble Voice? Yeah. Let's talk about that one. So the most common approach today is when we have, folks call. One, if you have sort of CSAT as part of your, like, live agent process, That can sort of be modified to include, like, the AI interaction and capture feedback about that as well. And then a second one is for any call that never touches a, human being, setting up some automation in your contact platform to send, like, a CSAT follow-up email, is a popular approach, assuming that you're able to tie together sort of phone number and email, in your system. I will also add that this is one of those areas where our sort of AI based auto scoring is gonna be really helpful. Well, it's not like a stand in for a CSAT survey. We are able to detect, signs of customer frustration, etcetera, automatically using AI, and we can give you helpful data about that sort of, in the platform on every single call without having to use, like, sparse CSAT data, to make inferences about customer satisfaction. Maybe a fast follow here. When is the auto scoring roughly scheduled to be released? Yeah. That's gonna come super early next year. So we're actively working on it and building it out. We expect to have that in the next couple months. Read another one. How do you think about the ability of the agent to call tools, and knowledge bases, especially complex policies balancing that with latency? Yeah. This is a great question. So in the workflow that we demoed on our live call there, we did kind of all of the above. We followed some complex policies. We made multiple different tool calls, and we used external knowledge documents, in order to answer questions. The way that we deliver this in a low latency way is actually by building sort of contextual, like, conversational, messages into the call flow. So when I gave my information to Cal, and Cal sort of once it had my member ID and my name and it was ready to look into our back end to sort of verify that information, which was an API call, takes a few seconds, it would just generate a message like, thanks for giving me your information first. And that actually buys us enough time to do all the work on the back end, for AI to follow the policies, make the API calls, etcetera, and still take some time to do that, but keep a conversational, interaction in place. This is really similar, I guess, to how humans operate on the phone. So, while you're on the phone with someone in a support context, you are sort of giving them conversational cues like, let me look that up for you, while you actually go do work on the admin panel or to authenticate or answer someone's question. And our AI works really similar to that. It looks like there's one about FlexCar. How did they eliminate eliminate overnight staffing support when needing to hand off? So, I guess I should preface this by saying if you are getting the types of cases that are going to be urgent and need to be resolved in the moment, you will probably still require some form of human staffing. Flex Car, was able to eliminate their human staffing overnight, and shift all of their people to daytime shifts because they realized all of the tickets coming in, were nonurgent, could be resolved by an AI agent, or, could wait until morning when a human could answer them. So it really depends on the types of cases for your business, and that's gonna vary from company to company. So I don't wanna overpromise that everyone can just eliminate humans completely for certain shift hours. It's really dependent on your specific business. See one here about, how are you handling safety escalations? Do you have guardrails in place? Yes. So this is a great question, and there's a couple of different ways that we think about this. So for one, we have really strong guardrails built into the product by default. So we've invested a lot in sort of tuning and specializing our AI to have strong defaults for sort of handling calls safely, understanding when a good time to escalate is, etcetera. The second piece of this is that this is also configurable. So during implementation, we'll work with customers to tune when we wanna hand off versus not, any kind of sensitive topics that might be specific to your use case in business that might not be, like, obviously included in our default guardrails. And then, also, we constantly monitor every call for any signs of, like, trust and safety issues or the AI, not sort of behaving within our definition of, like, acceptable and good agent behavior. So every call is screened using LLMs, and then we manually review those flagged conversations and make sure that we are eliminating, any kind of misaligned behavior. So we've been I guess, I guess, to summarize, we've been able to deliver, like, effectively zero rates of trust and safety issues, with our AI thanks to this approach. Cool. I think final question here. Are there any stats on trust with Gen z, and AI? So I did see a Forbes article recently that mentioned, over 70% of Gen zers, are using AI tools in general on a day to day basis. Don't have exact stats on phone support specifically, but, I do think AI is something that's pretty deeply embedded, within younger generations. They're all gonna be AI natives growing up with it. So I think they're gonna be much more savvy with AI than the rest of us will be. And with that, I think we are we've hit all of our questions here. So thanks so much everyone who stuck around with us. Thanks, Sam, for running that demo, and we will see you online at the next webinar.