Video: In Focus: Fin Voice | Duration: 2736s | Summary: In Focus: Fin Voice | Chapters: Subscription Management (775.39s), Low Latency Demonstrations (808.3s), Lost Card Verification (840.29s), Ambient Noise Options (951.55s), Conversation Auditing Tools (1047.975s), Fin Voice Configuration (1144.6s), Deployment Settings (1280.71s), Pronunciation Rules (1453.49s), Deployment Options (1542s), Fin's Architecture (1683.35s), Customer Success Stories (1826.51s), Teaching Finn Voice (1947.03s), Getting Started Guide (2131.5s), Q&A Session (2349.69s), Q&A and Resources (2444.87s), Implementation Best Practices (2522.18s), In-App Navigation Setup (2617.44s), Closing Remarks (2713.995s)
Transcript for "In Focus: Fin Voice": Settings in your workspace. After you cancel, your access keeps working until the end of the period you already paid for. So what if I just wanted a discount? It's kinda getting expensive. I'll check what the guidance says about discounts and pricing options. There is no way to apply a manual discount to an existing subscription. You can reduce cost by switching to annual billing, which saves 10% for example. A pro plan build annually works. Okay. That's enough for that one. So a couple of things that I wanna point out with that. So first, hopefully, you notice, like, the low latency. Right? Quick questions, fillers as well. Like, let me check the guidance on that. Let me see, you know, what we can do about that. So, hopefully, that's an improvement over what you've seen before. The other thing I wanted to point out is the interruption piece. So that was the real time speech layer running at the same time as, Apex Flash two point o. So you can see that you can, like, confidently interrupt it and Fin doesn't lose track of things. So let's do another one. This time I wanna do one with banking. Hi. You're through to Quenmark Bank customer support. How can I help you today? Yeah. I lost my credit card. I'm sorry to hear that I'll quickly check the steps to secure your card. I can help you cancel your card and order a new one for you. Before I make any changes to your account, I just 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. Yeah. It's 521547. Thanks. You're now verified. You have two cards with us. A Visa credit card. I don't know which one. I understand. Do you remember any recent purchases you made with the card or... I think I booked travel with it before I lost it. I see a few travel purchases on your Visa credit card, including one at Holiday Inn and another at Southwest Airlines. Is that the card you've. That's gotta be it. I recommend canceling your Visa credit card now and sending a replacement by mail. Would you like me to go ahead and do that? Yeah. That'd be great. Alright. So hopefully you get the point. On that one, what I wanted to show is you can build in the verification step, to make sure that Fin verifies who it's talking to. I also wanted to point out, and this was subtle, so maybe you didn't hear it, but there are different types of ambient noises you can put behind Fin. So on that first one, maybe you could hear sort of like a busy office sounding area behind just to, like, sort of reassure a customer that it's not dead air. And then the second one, had complete silence while, like, Finn was talking, but when Finn was looking things up, you could hear that, like, typing noise. So those are two of the three options. The third option is to turn it off entirely, so it's complete silence. Now the challenge with that is, often customers think the call is dropped. So just tread lightly with that one, because that's not always, the best user experience. But, some customers were adamant about having, like, a full silence on there. So I will show you on the back end how to set that up. Thanks for bearing with me too on those, technical difficulties. Hopefully, I remembered how to use a computer, and we'll be good to go for the rest of it. So what I wanna do next is chat through... Is this... Not quite. I got two. Is that better? Alright. Here we go. Alright. The magic of the Internet, y'all. Okay. So, the cool thing that I wanna show you now is, after that demo, I wanna show you the actual conversation, here in the workspace. So I was Sarah, when I called in, and similar to, like, any sort of chat or email conversation, you can see the full conversation events, by checking it in the inbox. So you can see, the call summary, which is generated here. You can see the full transcript of the back and forth. You can see, like, how Finn triggered this lost or stolen card procedure. You can see, the attributes that it's picking up. You can see the, the data connectors that it's that it's looking up within the procedure as it goes. So you have, like, full ability to audit, like, what's happening under the hood per conversation. And not just that. Right? So if we wanted to look at that procedure, that trigger in particular, I can go into the procedure UI and have a look. Right? So if I go into procedures here, I can click into this lost or stolen card, and I can see all of the triggers, all of the steps with the data lookups that it did. In here, I can also test this and simulate it before it goes live with customers. All of that is configurable via voice procedures. One thing I will call out, one product limitation that we're working on, in order to fin... For fin to execute a data lookup, it has to be within a procedure right now. You can't just have data connectors that FIN can readily access at any point. You do have to have, FIN looking those things up, on on your behalf, within a procedure. So that's one thing I will point out. Let me hop back over to a different workspace and show a few more things. So in terms of, like, Fin configuration, a lot of the things that you'd set up in chat will carry over to voice with a few key, like, differences that I wanna that I wanna, highlight. So content. So first and foremost, Thin will digest... Ingest your content and be able to read it as it goes. Now one thing to audit is if you have picture heavy text that doesn't have descriptors of, like, what's in the picture, An example would be, like, if you had a picture of your UI with a phrase like click the green button up at the top, that will not come across in a voice medium very well because you're not actually describing what's on it. So you need to have, like, a description of, like, here, click settings and scroll down to this section. You you gotta be able to, like, picture picture, like, anyone that's talking on the phone is, like, maybe away from a screen is a good, like, rule of thumb to think about. So you wanna describe what's in the content. You can't just rely just on pictures with it. So that's one, difference with, FinVoice ingesting content, versus Fin over chat. So that's one thing to, be aware of. Data as well, I already said it, but, right now, data connectors can only be accessed within a procedure that's being worked on that will change. Another thing, guidance. So we do have guidance for for Fin voice, across all channels. So you... When you click into a new guidance, you can choose specific channels for it. If you have voice specific guidance, you're absolutely able to set voice specific guidance. And here's, here's the thing that I said literally is, like, hot off the press, is the the escalation. So now, FIN, can read FIN attributes. FIN voice can read FIN attributes. So if you have, different escalation rules, again, like if you detect this value, escalate immediately, that type of thing, Fin can now read those attributes, which is great. It's a it's a new... As of, like, two days ago, this literally happened. So, yeah, one of the fun things about this job is things change every single day, which I love. So, yeah, in terms of setup, we already took the... Looked at procedures. You can also analyze. So the analyze tab works as well. You can, filter by channel and take a look. I don't actually have any voice set up in my personal workspace, but if you wanna look at the automation rate for voice, if you wanna look at see... If you wanna look at, how many, conversations have been handled, it's all in here, as well. I feel like there was one last thing. Oh, yeah. Deploy. How do we get this thing live? Right? Okay. So two main ways, not unlike chat and email. We have, we have simple deploy and we have advanced deploy with workflows. So simple deploy is the simple one. Let me go through a bit of the UX on that. So one of the things that's unique about voice, is setting of office hours. So many times when customers are testing out FinVoice for the first time or they've made the decision to do this kind of forever is to let Fin handle, certain conversations within or outside of office hours. So that's one thing you can do is make sure that your office hours are defaulted, and I'll jump down here to the bottom to show that you can set Fin live within office hours and assign a percentage of calls that Fin will actually interact with, or you can do outside of hours and same thing, same percentage. You can mix and match. You can have, it on for both or one or the other and, a varied amount of percentages. Generally, when you're testing out, when you're trialing, we recommend, like, a five to 10% slice. Pay really close attention to those conversations, see how they're going, see if you need to adjust any content, any guidance, any procedures. Start small and work from there. But to go through the rest of these, so here's some, really, like, cool newer settings that we have. So language and voice. Finvoice can speak, 29 languages. You can pick not only the language and the the gender of the voice, but you can also... There's a subcategory of preferred accents underneath. So we have a lot of different ways that you can have, Finn come through and sound. You can customize, what Finn says when Finn picks up the phone. You can choose to have different greetings inside or outside of office hours depending on what you wanna set up. And some of the newest stuff that we have are in these voice settings. So couple of cool things with this. First, this is the, ambient sound that we heard on the demo. So you can choose either the ambient sound, so that's that quiet room tone from the moment the call connects. That was that sort of like office y type feel that we had behind us. There's the hold sound, which plays the typing noise when Finn is looking something up. Otherwise, it's silent or complete silence if you want, that. And here, we have the little cautionary tale here. Callers may think that the call has dropped. So we'll leave it there for now. Pronunciation rules is also something that's fairly new. So, had a little fun playing around with this one. You can tell Finn how to pronounce things. So for example, this word, everyone tends to say it a little bit different. So I'm gonna say, say it like this, and then let's let's listen to it. Let's see if this comes through. Datta. Not exactly what I was hoping for. Da... I was thinking more like dada. Da. Maybe that'll work. I don't know. Maybe not. But you can go in and you can configure, like... Especially, like, if you have a unique company name that's said in a particular way or you've launched a new product that, maybe isn't, like, said in an intuitive way, you can spell out, phonetically, or through a couple of different ways, how you want Finn to say that, so that it doesn't sound dumb, basically. So... Yeah. We couldn't hear your browser here on this one either. Nice. Okay. Good stuff. Let's pop over this again. Let's see if this works this time. Alright. Here we go. Did that work? Everybody hear that? Nice. I love it. Okay. Cool. Yeah. So that's that's to make sure that Finn doesn't, doesn't say something weird. Okay. Sweet. Let's jump back and go through the rest of the UX, shall we? So let me go back into FIN. Nope. We're deploying right now. We're gonna go back into voice. Okay. So with simple deploy after you've, configured your settings, Again, you can look at what it has in terms of, help center articles, snippets. You can manage content and guidance, as well, which we already went through. Talked about how we can now detect, FIN attributes, so you can choose which attributes you'd like FIN to detect. And, this has little note that you can triage the call via a phone reusable workflow. So not unlike chat. If it detects an attribute, you'll need it to send to a reusable workflow that then triages it to the right team based off of that. And then a little bit different than, chat as well, like, you can ask for a rating, but people are gonna use their keypad instead, one through five. So, yeah, that's a little bit different there. And then you have a myriad of options for if Fin can't resolve the conversation. So inside and outside office hours, you can offer the callback, you can transfer your teammate, stay on the call, so on and so forth. Lots of different, configuration options for you after the call goes. And then, last is if customer hangs up, you can route the conversation. So either you can close that conversation or you can assign the conversation out to one of your team inboxes. So that's simple deploy. I already skipped... I already went over this, so I won't, belabor the point there. But let's look at advanced deploy, and this is gonna look very similar to how you all have your chat workflows and email workflows set up. So you can just build a workflow. Right? So if I click in here, it's gonna template out a, FinFirst workflow, where we pick the audience. Maybe we just have a specific slice of audience that we wanna test out on this, and then we have FinVoice handling the first step. And after that, just like any other workflow, if Fin has to escalate, you can tell it where to escalate to from as well. So that's your advanced deploy option two. Oh, nice nice gifts. Now I'm I'm catching up on the, the chat. I'm glad I gave somebody a couple chuckles. Okay. Alright. So I believe that's it from a deployment standpoint. It is. So let's jump back in to the slide deck, shall we? Okay. So let's keep going. Let's talk a little bit about what's under the hood, the architecture of Fin. So first and foremost, like, I talked a little bit about this, but, this is so fast because we have our own proprietary model that is built for speed. It's like the Ferrari of voice AIs right now. The other thing is we made a very specific decision to build voice, our voice AI, unlike many, many other voice AI's that are out there right now. Most voice AI's right now run on what's called a cascade architecture. So caller speaks, there's a transcription model that's turning speech to text, that's the STT. And then once it's turned into text, it runs it through a general purpose model, an LLM. And then once that text is generated, the answer is generated, then it has to turn it back into speech. So this all runs one at a time and takes a lot of time and can cause the latency that kills the call. Right? Customers basically have no patience for, AI if it's slow on a call. They do on chat. They do on email because, you know, you're multitasking. You're doing other things. But, again, like, phone is urgent, so it needs to be fast. So how is it fast? So it runs on our specialized hybrid model stack. Right? So the caller speaks, and there's two things happening at once. There's the real time speech layer, which is speech and understanding and response delivery, listening for those interruptions, and then there's the Apex Flash two point o. So that's the answer generating that's super fast. Then once that answer is generated, it turns it into speech, and then the caller hears. So that's how we've been been able to make it a lot faster. There's also within the model itself something called noisy mode, which I just learned is actually what our engineers call it. So if Fin detects that you're in a noisy area, it will actually slow down its response time and, like, increase its acoustic model to make sure that it's, like, really listening to you, which is just fascinating. So you you lose a little bit of time, latency. Like, the latency grows a little bit, but Fin is a little bit more accurate in noisy settings. So, again, this is how we've been able to get it, to where 24% faster answers have been viewed and 46%, fewer hallucinations based on this model architecture. But don't just take it from us. Right? We've got teams that are already doing this and having a lot of success. So Lovespace, I love to talk about because, you'll see down here, 1% of their calls use the procedure. So they're able to achieve a 57% resolution rate, based off of basically content alone, which is great. And it kinda runs counterintuitive to the whole thing I've been talking about is, like, voice procedures are here. You should do all that. There are some companies that have done it, without it just based off of their, query types. Park b, these are, booking calls, so 70% resolution rate for for sort of run of the mill booking calls. Speaking of which, in the docs tab here in Goldcast, there's a link out to the, FinVoice presentation that we did at Pioneer, and there's a real park b call in the midst of that conversation. Obviously, everything is anonymized and that sort of thing. But you can hear park... You can hear Finn helping a customer of park b get out of a garage that was locked. So kinda cool. Definitely check that out in the docs section. I have one other, docs in there as well, and I'm kinda blanking on what it is right now. Let me take a look at what that is really quick. Oh, yeah. The demo page that we use. So if you wanna play with the demo page that I failed to share the audio on, feel free to to jump into that at any point throughout the call. It's kinda fun to play with. Solid core now... So this is like an automation rate stat. So 23%, almost a quarter of their inbound calls are fully resolved by Fin, which has allowed them to focus more on their, like, in studio clients. So, again, it's it's not this is not in a, like, let's replace them kind of a way. It's like, let's focus on the people that are actually in person and, in front of our face, and let let AI do what it does, in terms of, some inbound calls. And then Topstep was able to allow twenty four seven phone coverage for the first time using FinVoice. So, again, couple of customers... Customer testimonials, out there for you, to take a look at. Okay. Next. Teaching Finn to talk. Okay. So this is a bit of a rehash of what I talked about in the UI, but there are a few things that I wanted to make sure and highlight. So Finn essentially already knows what to say. Right? Like, you have configured the knowledge content. You have given it guidance. You have maybe set up a few procedures and data connectors. You have some audiences and targeting rules, escalation paths, reporting and insights. Like, many of you on the call are probably Fin users, right, in other of these areas. So we're not asking you to, like, start from scratch and start over. But what's different is, obviously, like, Fin, on chat works a little bit different in that you can't just, like, fully jump it over to voice and not touch anything, and it'll work perfectly. So there's a couple things that you do need to do. So first, like, the most obvious one is you need to connect your phone. So intercom phone, it works natively, obviously. So if you use our phone service, it'll work, kinda out of the box. There's call forwarding, which is great for, testing in particular, just taking one number, having it forward over from your telephony into Fin, and if Fin Voice has to escalate, it can re, forward back into your telephony. And then sip. So that's like a deeper integration. Generally takes a couple of weeks to set up, whereas call forwarding takes, like, an hour or less, at at... For for for many customers. Picking a voice and greeting, which talked talked a little bit about. You can add the pronunciation rules for brand terms, make it sound smart and not dumb. Voice guidance, you can add specific voice guidance, for it. Checking audience rules, running the simulations, especially if you're setting up a new voice procedure, super important. You can do all that in operator, by the way, as well. So shout out operator, which is making all of our lives easier. And then you can set your rollout percent. So, again, I said this before, but start at a a smaller slice, like five to 10%, of a rollout in hours and out of hours. So that's, that's really just kind of a rehash. Here's a couple of things to make sure that you're doing specifically with phone. So every call is logged. Now we went into the inbox, and we saw that recording, the AI summary in the inbox, and we could see which articles Finn used in each call. Super great to be able to have that visibility now. Fin insights for voice. So, you can look at topics and trends across your calls with one click suggestions. Didn't jump into that in the UI, but that's available for Fin voice as well. Ask operator. If I had a nickel for every time I said operator, I could retire. Operator can do a lot of these things. Right? Bill me a refund procedure. What are callers talking about, this month? And and let it help you take from end... From problem to solution within operator. And then the last thing is, obviously, CSAT is measured a little bit differently. It's collected on the keypad and only when the call isn't handed over. So couple of different things to know about that. The next thing that, I wanna make sure I double click into is the three different ways. Intercom phone works natively. Sign it, some new numbers. Nothing new to set up. Call forwarding. Like I said, you can call forward from any provider, very, very simple to do, and then the SIP integration. So deeper routing, here are a couple of the providers that we've worked with to set up SIP. But if the provider is willing, we are always willing, to set up that deeper integration and can take... Just note that it takes a few weeks to get going and live on that piece. So we do recommend, if you don't use intercom phone, start with call forwarding, for the purpose of a a trial before you look at a full SIP integration while you're giving us a test drive. So, and then here's Torsen having a little bit of fun with my face popping up for the last point that I wanted to make. Forwarded calls behave just like the direct ones. Right? So phone workflows, IVR branches, those all still work. Getting started. Okay. This is... I I think this is my last slide, and then, we're... We've got about, ten minutes for open q and a. Again, I've got some teammates backstage. If you wanna start loading up your questions, they can start answering them there for you. But, we really want you to try this. Right? Even if you're already using Fin on chat and email, you can still try FinVoice in a trial, separate line items. So talk to your account team, say, hey. I wanna try this thing. And they can enable it, trial for your workspace. Connect your phones. Again, like, if you use intercom phone, that's the easiest. But if you're... If you don't, just forward over some calls from your current provider, and then we really recommend going live on a slice. Don't, you know, don't give it everything at once just like you wouldn't for chat or email. Give it a quick, quick slice, see how it does, with some run of the mill questions and, and build from there. Oh, I do have one more slide. Okay. So what what a good start looks like. Just note that for our newest FIN customers that started in March, there was about a 42 to 44% resolution rate for them on average. Right? That's across all of the customers that started since March. And in general, they lag about 10 points behind in terms of resolution rate. Reason being, people are tougher on phone. Right? It's more complex situations, all the things that I've been talking about. But I will say a third of those customers actually beat their chat rate. So, you know, you could be part of the two thirds. You could be part of the one third. It probably has a lot to do with your customer, your query taxonomy, all that good stuff. Also, just note that one in seven customers will ask for a human right away. So that means, optimistically, six out of seven will let Finn have a try. But just know that, you know, and many of us are like this too. We say agent as soon as we can or representative. Many people are in that boat. And then four in 10 handoffs never actually reached a person before Finn. So it either went to voice mail, by hang ups, dropped in the queue. Some of this was by design, and some of this is just people hanging up. So here's just some things to sort of anchor on and make sure that you have, the right, sort of grounding for what success looks like. Alright. Now I'm not lying. We're done. Questions. We've got about seven minutes. So please answer your questions. It's really great to be with all of you again, obviously. And, yeah, let's, I got I got a lot of smart people here in the background with me, so let's, let's dive right in. Okay. It's great to see if invoice improves. Alright. Let's answer this one. Okay. Still very basic phone system compared to others like RingCentral and Dialpad. We'd love to see either... See these tools built in the intercom phone or a better connector to allow them to use FinVoice without, having to transfer the calls. Yeah. Absolutely. You know, it's it's one of those things that we're continuing to build, and, you know, it's only gonna get better from here. So, yeah, good shout. We'll continue to work on these, FinVoice to make it better and better each day. Alright. Let's see. Here's this next one. How does audiences work in perspective of incoming calls? So, there's a couple different ways to think about audiences. So if you're deploying via workflow and you have audience data, you can obviously set up that only those that are in this audience with these numbers come into it, or you can just assign one number to Finn and make sure that, the only folks that have that number are the audience that you want Finn, to do. So, yeah, that's how the audience will work. And then let's see. This last one, register your interest in Deploy Voice. Thought I did that when I opened the October 5 email. Okay. Interesting. Daniel, let me let me look up your specific workspace and see what's going on with that. I'll I'll have to get back to you on that one. I'm I'm not sure why you're seeing that. So... Yeah. I think that should be it. But, yeah, we are hanging out. If there are any other questions that you have, we've got about five minutes left. If you are looking for an afternoon snack or whatever, feel free to log off as well. By the way, as as a registrant and attendee, you will get a recording of this if you wanna share it around. We have a couple more recordings from other time zones as well. And, so if you wanna see, like, you know, other things that maybe I forgot to say here or didn't say correctly on the others, those are all the product webinars landing page. But, yeah, play around with the demo page, that I shared in the docs section. So where it says chat, docs, and q and a, go to that demo page, play around with it, share it around. And then, really, that FinVoice at Pioneer, presentation is really interesting if you haven't had a chance to take a look at it. Okay. Lauren, let's do this. Okay. Our team is getting ready to migrate our phones from Freshcaller to intercom, but very nervous to let Finn go live on phones as we're struggling to scale Finn in Messenger as is. Any advice? Yeah. Start small. Right? Like, even with, even with Messenger. Right? Like, as you've been going through some some growing pains with it, like, can't stress enough, like, be particular about the audience that you try it on. It might be worth, it... It's it's shocking to me how many times I'm working with customers and after doing, like, a brief content audit, find, like, several things that could improve. So I talked about how that's important for voice in particular. But even for chat, like, sometimes I found, like, so many duplicate articles or outdated information or or things like that. So sometimes a good old fashioned content and guidance audit is, is what it takes just for Fin in general. But if there's something more specific, Lauren, that you're seeing, I'm happy to, answer a little bit more specific to your use case. Let me throw a couple more on here. Cost for FIN resolutions higher or different than FIN outcomes. So for voice resolutions, they're counted as an outcome. There's a little bit different pricing structure for voice, than chat. You can chat with your account team about, like, specific account level details on that. But, yeah, essentially, like, it's charged the same way a Fin outcome is. If Fin resolved it or if it's hands off successfully within a procedure by design to a team member, those are the only times that you would be charged for that outcome. And then this last one, Jared. Alright. Can you clarify your advice regarding in app navigation? Should they be snippets and articles and text, or is there any other way to ensure the agent has accurate understanding of how to help a user get around our app? I see. So when I was talking about the, article configuration. So just picture, like, so FinVoice can't, like, describe a picture yet. So if you have articles that, are basically, like, a little bit of text and, like, giant screenshots of a UI, that can cause some problems because Finn doesn't have the text to read to sort of understand what's happening on the picture. So making sure, like, you're detailing out, like, what steps you're pointing at in the picture, like, if you're having someone update their settings or, like, relaunch or or what... Reconfigure, whatever the case may be. Make sure it's, like, listed out in text so that Finn can read it, basically, and then be able to speak it. So hopefully that clarifies a little bit, Jared. But, yeah, that's essentially... You just wanna do... Instead of, like, what I don't want people to do is, like, oh, I'm gonna write, like... We've got 200 articles. I'm gonna write 200 voice versions of those articles. Not probably the best use of time. There's probably a way that you can go in and you can audit those existing articles and just change them so that they're, they have, like, more read steps that even a reader will look at and still use the pictures, but Finn can also, like, read them back over voice. So... Awesome. Alright. Well, I am just being told that we're at time. So thank you all. It's great to, spend some time with you as always. Thanks. Come back soon. We'll be talking about evals, here next in about a month or so. So we'll see you then. Thanks, everybody.