Video: In Focus: Fin Voice | Duration: 2492s | Summary: In Focus: Fin Voice | Chapters: Welcome & Introduction (4.56s), Voice AI Interaction (104.385s), Support Channel Evolution (203.625s), Product Improvements (317.36s), Live Demo Walkthrough (394.96s), Verification Setup Demo (510.47s), Card Replacement Process (552.525s), Inbox Conversation Review (680.985s), Testing and Configuration (761.285s), Configuring FIN (805.54s), Deployment Configuration (884.425s), Deploying Finn (1005.16s), Hybrid Model Architecture (1090.145s), Customer Success Stories (1189.03s), Voice Configuration Details (1262.23s), Deployment & Integration (1351.93s), Getting Started & Results (1405.29s)
Transcript for "In Focus: Fin Voice":
Alright, everybody. Welcome back. We are here for our next bin product webinar. As always, please pop in the chat where you are dialing in from. We have customers all over the globe. We'd love to hear we're dialing in. And, yeah. Pop in the chat, whether it's a GIF, whether it's an emoji, whether it's a hi, hello. I'm from this place. I am beaming in from Chicago. As always, I have some teammates in the backstage area that are here to answer all of your questions, respond to your chats, all of that good stuff. So welcome. Great to have you here. What we're gonna do today is talk about Fin voice. So, many of you have probably seen Fin in action over the Messenger, over chat, over email, over SMS, over Discord, wherever the case may be. But maybe not Finn over voice, Finn over the phone. And that's what we wanna do today. So a little bit of housekeeping. I wanna set the scene. I want to demo... Talk talk a little bit about, like, where we've been in the support journey. I wanna do a live demo for you as well, and talk through basically how FinVoice works, and you'll see it live as well. By the way, the demo page I'm using is linked in the doc section. You can use it at any time or have your team use it as well. Under the hood is me talking about how Fin is configured and architected, and then we're gonna talk about optimizing Fin, which is actually more how it's configured. Under the hood is more how it works from a model standpoint. Then the last is getting started and our recommended next steps. So let's set the scene. First and foremost, quick show of hands. How many of you have interacted with an AI using your voice this week? Pop it in the chat. Tell us which one you've used, whether it's voice mode in chat GPT or Claude, dictation, your car, a smart speaker, etcetera. Many of you, I'm sure, have been using it all week. So, yeah, we'd love to hear how you're interacting with it. And I have a hypothesis. The reason I'm bringing this up is, like, a voice is really becoming the dominant way we interact with technology. The reason why is it's just easier, frankly. First, it's faster. Did you know most of us speak about three times faster than we type, which is, if you haven't tested it, it's... It is largely true. Unless you're an extremely fast typer, you probably speak faster than you type. Secondly, it's easier. You don't have to, like, worry about menus or forms or working out how to phrase it. In fact, I've become like the world's worst speller because now I just talk to Claude Code. Now I just talk to, you know, all these different things, that that are AI now. And last, it's just natural. Like, you can interrupt yourself. You can change your mind. You can add context as you go. It's like talking to a person truly getting really, really cool. Let's talk a little bit about how how phone has come full circle. So if you remember back in the nineteen seventies through about the nineteen nineties, pretty much the only way you could get help was over the phone. Right? There was no Internet. When the Internet came out in the .com era, email became the dominant way. There's no queues, so you didn't have to wait. Like, you didn't have to listen to the crappy hole music anymore, but it was slow. Answers, could take at best hours. At worst, it could take days. Then in the twenty tens, Finn, at that time, Intercom was one of the pioneers at developing live chat, and that became the dominant way that we pushed people to that. It was cheaper. It was easier. It was faster. Quick back and forth. Now, like, phone is coming back full circle, and phone with Finn in particular is instant. It's natural. It fixes the problem on the call. Now there's probably, the the way to think about it is, like, slowly higher stakes. So Messenger is the lowest stakes. Quick questions. People tend to give it more grace than other channels, so that's where many of our customers have started. Next tier up is email. It's more complex. It's multi part. There's more context. Sometimes there's more back and forth for clarifications on those types of things. It's a little bit higher stakes, a little bit more of a paper trail. And then now is phone. Now if you're like me, I don't pick up the phone unless I have to. So phone is often more urgent. It's more emotional. Nothing can be reread. Like, silence costs you the call. If it's too slow, they're gonna hang up. If you can't get the thing fixed, they're gonna hang up really, really high stakes. So if you've tried thin voice before, there are probably some things that you didn't like about it, which we have fixed. So maybe what held you back before is it could answer, but it couldn't act. Now, you can take actions during the call or Fin can take actions rather during the call through procedures. You can verify callers. You can update accounts. You can process refunds. That's now GA for anyone that has access to voice, which is great. Procedures were also hard to build if they were available at all to you in your plan. Now, you can build in plain language. In fact, a lot of people are using operator to build, voice procedures for them. Next model wasn't built for phone. We used a general purpose model, wasn't built for phone. It had high latency. It it wasn't a great, model that we used. Now, we developed our own Apex Flash two point o in house. It's built for phone. Our own model plus there's a real time speech layer, which brings faster answers, fewer fewer hallucinations. Everybody likes that. It handles it handles noise and pronounces your brand terms as well, which we're gonna get into. It's also really hard to see why a call failed. Now we've got insights to improve it, real time topic insights, one click recommendations, full actions, and thinking log on every single call. So I want to demo this for you. So let me pull up this demo. Again, this is linked in the docs page. Feel free to play with it, after the call, but we're gonna run through a couple different scenarios. The first one is software as a service. I'm gonna show you how Fin can cancel a subscription. So let's do it. Hi. You're through to Keltthorn Software customer support. How can I help you today? Yeah. Hi. I wanna cancel my subscription. Understood. You'd like to cancel your subscription. Let me check what steps are involved. Hear the background noise. What is the email address on your account? It's demo@fin.ai. I've got that as demo@fin.ai. Is that right? Yeah. That's right. Hear the background noise? Pretty cool. You can configure that. We'll get into it. I found your account, and you're on the pro plan at $650 a month. Wait. I don't wanna cancel, and I just want a discount. There are a few ways to lower your bill, such as switching to annual billing for a 10% discount or downgrading to a smaller plan. Actually, on second thought, let's cancel. I found your account, and you're on the pro plan at $650 a month, paid up until the August 14. Is that the subscription you wanna cancel? Yeah. One moment. To understand better, what's prompting you to cancel the subscription? I just don't use it. It sounds like you're not finding it useful anymore. Should I go ahead and cancel your your subscription is canceled. You'll keep access until the August 14, and you k. So a couple things to point out. First, I was kinda rude. I cut off Finn a couple times. You can see that Finn handles interruptions pretty seamlessly. Also, you saw a little bit of the capability that Finn has to do things on your behalf now, which is great through a procedure. Now the one thing it didn't show was how to, set up verification. So that's why I wanna do another one. Now you can set a verification for any industry vertical, but the the, the workspaces that we have set up that are doing these demos have a verification for banking, not software as a service, but both are available. So let's do the banking one. Hi. You're through to Quinnmark Bank customer support. How can I help you today? Yeah. I lost my credit card. I'm sorry that happened. I'll check the steps to secure your card now. So you hear the typing now this time. I can help you cancel your card and order a new one. As opposed to the office ambiance. Before any changes, I need to verify your identity. I have sent a six digit code to you by text. Please read it back to me when it arrives. 527968. So that's what you can set up. Thanks. You're now verified. You have two cards, a Visa credit card ending in 6 I don't remember which one. That's no problem. Do you remember a recent purchase you made with the card or perhaps the shop where you last used travel with it? Look. It named me Adam. I didn't set that up. I promise. I see two travel bookings on your Visa credit card. One at Hyatt for $549.33 and another at That's gotta be it. That's the one. See, I keep interrupting Finn. I recommend canceling your Visa credit card and mailing a replacement. Would you like me to go ahead and do that? Yes, please. I'm processing your card replacement now. We'll mail a new one out shortly. The card ending in 6786 is now canceled, and a replacement is on its way. It should arrive within three working days. Is there anything else I can help you with today? No. That's it. Thank you. You're welcome. Goodbye, and take care. Okay. Pretty cool. Right? So we saw the authentication part on that one. Again, I swear I didn't I didn't preprogram it to call me Adam Adam. I'm Adam Temple, not Adam Garcia, but that that was kinda cool. I continue to interrupt Finn, and Finn kept rolling with it and was able to cancel my card even though I couldn't remember. It did that look up and shared, that it could cancel and replace the card for me. So all of that is configurable now via voice procedures. So let's take a little bit of a look at how it works behind the scenes. So I'm gonna pull up the inbox for that conversation. Here it is, Adam Garcia. And just like any other channel now, you can show, conversation events, and you can see kind of how the call goes just like you do on any other channel. So this really enables, like, full omnichannel, enablement here. It's got the preview. It's got the full transcript. It's got Finn, assigning attributes, in this case, language. It shows the procedure of lost or stolen card that, was able to get called up. All of these are in here, and you can go through and analyze the conversation from the inbox level, which is really, really cool. And, it just allows you to have, like, full visibility into the conversation. Additionally, if I wanted to look into that procedure itself, I can jump in, straightaway into that procedure. Here it is. Lost or stolen card. I can jump in. I can see what the triggers are. I can see the instruction blocks, the API calls, all of the good stuff that's in here. I can also test it. So simulations are available for voice procedures. I can make sure that Finn is testing itself before we go live with anything. All of this is right in the product here. Now how else can I configure Finn is probably what you're asking next? So let's talk a little bit about how to configure FIN in general. So under train, I already showed, these procedures, but voice learns your content. It has exposure to all that. You can give it data that's living in your intercom help desk or via data connectors in your external systems. You can give Finvoice guidance. Right? You can select your channel as all, or you can do specific, voice specific guidance. You can train Finn on your different attributes, and it can detect that through the call. The one thing I will point out is you can't do rule based escalation guidance just yet. So if fin detects an attribute, it can't auto escalate like it can in chat or email, but that's coming soon. You can, then deploy it. Right? Actually, before I get into deploy, you can analyze. So if I wanted to look at my voice specific performance, I could add my channel as voice. Right? So I can put that in here. I can choose voice specifically. I don't have voice live on mine, so it's not gonna show any, But you can look at your performance just like you can on either of these others. Recommendations are gonna be there as well. All of that is configurable. In terms of deploy, now there are two different ways to deploy just like normal fin. You can do via simple deploy or you can do via workflows. So let's work through what simple deploy includes. So, you wanna have your office hours in here because simple deploy here at the end, you can set Finlive inside of your offers... Office hours or outside of your office hours. You wanna make sure that that is configured properly before you go. In terms of how Fin introduces itself, you can write out exactly how you want Fin to answer the phone, which is great. You can choose from one of the 30 some odd languages that we have available and are adding more and more. You can pick male, female, or no preference. You can pick the accent that you're looking for within the the language, as a subcategory. And then you can use the same greeting, inside or outside or configure your own that's different based off of office hours. You can also then test and hear what the voice is gonna sound like and play with the different combinations that go along with that. Just like I just showed, Finn will follow your training. So your content, your guidance, and has click throughs to each of those parts of the UI as you go. Finn can also detect attributes, so you can tell it which ones to detect. Again, it can't escalate yet based off of an attribute as soon as it, detects it like it can over chat or email, but that's coming. But it can triage based off of it. So Fin will assign an attribute at the end of a conversation, and then you can route it, that way. You can have it ask for a rating, so your typical CSAT on the keypad. If you enable this, it will say on a scale of one to 10, and you can tell it exactly what to say, or I'm sorry, one to five. If Finn can't resolve the conversation, you can have it, do a number of different things. It can offer a callback. It could transfer. It can stay on. It can hand over to a workflow. It can forward the call or end the call. And those are different, different configuration options within inside or outside office hours. And then you can also tell it, which, team to assign it to in terms of team inbox. Then if the customer hands up... Hangs up as you, go along, you can either have Finn close the conversation or assign, the conversation out. Last, setting Finn live. So like I said, you can set Finn live inside office hours, outside office hours, or both. And within Simple Deploy, you can do a percentage based assignment. So either zero to a 100% with any increment there or zero to a 100%, any increment there, or any combination of increments there. So that's how you deploy, Simple Deploy, Fin over voice or Fin over phone. If you wanna do it via workflows, you can do that as well. So you can set up just like any of your other workflows. You can choose your channel that you want it to... Or sorry, your audience that you want it to go on on the phone channel. You can configure having Finn, show up first, or any combination of... You can have different things happen if Finn has to escalate during office hours or out of office hours, or you can add an attributes, and you can ultimately tell it how to end the conversation. So, yeah, all of this is gonna be very similar to how you configure the rest of Finn. Getting back into the presentation, now that we've gone through the live demo, let's take a peek under the hood. So first and foremost, like, that call was fast. Let's talk a little bit about why it was so fast. So this is how most voice AI works. It works on a cascade architecture. So caller speaks, and there's a speech to text transcription model that works first. So it in... It interprets the speech, turns it in text, then it runs it through an LLM, which will generate the answer. And then that text has to be reconverted into speech for the AI to then speak to the caller and hear it. So that's the typical way that most voice AI is constructed. Ours is different in that it runs on a specialized hybrid model stack. So the caller speaks, and then the model layer runs. So there's the real real time listening API that's happening. And then there's the Apex Flash two point o model running as well, which is, on one hand, speech understanding and response delivery, understanding interruptions like I was showing. And then the Apex Flash two point o is the generation and reasoning fast response time. And then once that answer is formulated into text, then a text to speech synthesis model runs and the caller hears. So that's why it's able to go so much faster with, thin ape... Or not thin, Apex Flash two point o, which we're seeing 24% faster answers, and we're seeing a forty six percent, drop in hallucinations as well. Both are good things. We're we're seeing some really cool customer applications with this. The first one I'll call to your attention is Lovespace. They didn't really have to use procedures. Only 1% of their calls are using procedures, but they're seeing, over the last, several months, 11,000 calls handled by Finn with a 57.1% resolution rate. Let me jump my face over here. Park b is seeing a 70% resolution rate for booking calls, which is fantastic. Solid Core is seeing almost a quarter of their calls fully handled by Fin. And then Topstep was able to enable twenty four seven phone coverage by putting FinVoice on. So don't just take our word for it. Customers are really benefiting from this as well. Teaching Finn to talk, optimization. Ongoing optimization is still a thing. I'm gonna jump my face back over here as... Again, so first and foremost, like, a lot of this is stuff that Finn already knows. Right? You've probably configured a lot of this, whether it's knowledge, content, guidance. You may need to put in specific phone, guidance, procedures and data connectors, audience and targeting, escalation past reporting and insights. So those are all the things that you're already familiar with. Then it just becomes teaching VIN to talk. Right? So there are voice specific things like connecting your phone, intercom phone, call forwarding, or through SIP. Right? A deeper integration, which I'll talk a little bit about. You can pick a voice or a greeting, and greeting like I showed in the back end. You can add specific voice guidance, if applicable. You can look at, different audience rules, and make sure that you were deploying only to certain audiences or inside or outside office hours, with with setting your roll up percentage or... And you can run voice simulations and a test call all within there. So that's the voice specific things, but it all builds on top of what you already have configured with Fin. Now there are some differences in terms of, like, phone specific configurations that I wanna call out. The first is every call is logged now. So the transcript recording and AI summary is right in your inbox, which we saw. You have Fin insights for voice in your performance tab. So the topics explorer, the recommendations are all going to populate, with what you have for phone. Next, you can ask operator, like, hey, operator. Help me build a refund procedure. What are callers asking about this month? And lastly, it's measured a little bit differently. Right? Like, if you configure CSAT, it's gonna be on the keypad versus the, emojis over over, over chat or over email, which which kinda makes sense. Right? In terms of where Fin goes, it sits on the phone system that you already have today. So if you have intercom phone, like, super easy, just pick one of the numbers that you wanna deploy Fin on. Nothing new to get going or buy. Call forwarding, if you wanna forward a call over from your provider to Fin, up and running in under an hour, that's really the ideal way to trial this out. And then the last is that, SIP or SIP integration. I've heard both. Deeper routing and data sync back to your platform, here are a couple of the platforms that we work with on on that type of integration. But those are kind of the three ways to get Fin going over the phone. And then just having a little fun here. Forwarded calls behave just like your direct ones. So phone workflows and IVR branches will still work. So getting started, here's what we, are seeing. So if you already are using Fin on chat and email, you can still try Fin Voice, free of charge. It's a separate line item completely. Talk to your account team. They can get you started on a Fin voice trial for your workspace. Connect your phones. Again, if... Super easy if you're on intercom phone or forward over, some calls from your current provider. And then we recommend going live on a slice. So five to 10%, listen to the first few calls, see how it goes. And, yeah. That's how to get started. Now in terms of what good looks like, there's a couple things that we wanna, call out. So, from our newest FIN customers since March, give or take, we're seeing about a 42 to 44% resolution rate from month one onwards with basically no learning curve, just turning Fin on. What grows is really how many calls you give it, how big of a slice you give it. For about two thirds of the customers, that have deployed FinVoice newly since March, the... Fin tends to be at about 10 points lower of a resolution rate over phone. But a third of the customers since March have actually beat their chat rate. So just, again, wanna, set your expectations properly. One in seven, callers ask for human right away. Like, they see... They they hear that it's AI, and they're like, I'm out. But many are happy to let Finn, have a try. Six out of seven in in fact. And then four out of 10 handoffs never actually reach a person, before Finn. So, either voice mail, hang up, or drop it in the queue, so that'll happen as well. So, again, as you start a trial, here are some things to kinda pay attention to in terms of, what to expect. And then that's really it. So with the time we have left, which we we do have a fair amount, we'd love to answer any questions. Again, I've got some fantastic teammates backstage that are here to answer your questions. We'll hang out for the next twenty minutes or so. And, yeah, please ask all the questions. Again, check out the docs tab where you can, look at some some really interesting information. You can play with that demo site that I have linked in there. I also have, some FinVoice demos that were, in Pioneer. If If if you were there last week, some of that content is in there as well. You'll hear from, our PMs and some different folks as, in in Pioneer. But, yeah, just so grateful to be with you all, interacting. And, yeah, please ask your questions. And, again, thank you for coming, and, we hope you enjoyed the presentation and, get to try out FinVoice.