Name: Fin Setup & Best Practices
Uploaded: 2025-02-05T20:16:05.041Z
Duration: 1 h 8 s
Description: Fin Setup & Best Practices

Transcript for "Fin Setup & Best Practices": Hello, everyone. I did not promise any free resolutions. Don't don't hold that to me. Welcome to today's webinar for FinCiteUp. My name is Tim. If you haven't seen me before, welcome. Let's, go ahead with our classic. Please drop in where you are signing in from today. As always, I'm calling in from Chicago, and I'm gonna share my screen real quick. And if you don't wanna share where you're signing in from today, let me know, what you've been listening to recently. I feel like I've been in a, music loop, and I'm I'm looking for, I guess, like, new inspiration in terms of music to listen to. Wow. We got representation all over the world. As always, shout out to my fellow Canadians. I think I saw Alberta as always. Guelph. Welcome. Welcome. Montreal. Montreal. As always. Cool. Cool. And let's get started. We're not gonna spend all day, welcoming folks. I love the representation. Keep it coming on. Let's go through oh, new Lady Gaga. I I did hear it stuck in my head. It's stuck in my head. Let's move on. So I am Tim. Everyone, please welcome my colleague, Rigby. He's new to the team, and he'll be commenting every now and then in the comments. So please drop him a warm welcome with a GIF or a comment in the chat, please. We've already been through all this. I guess, I will mention q and a tab on the right hand side if you need to ask any questions. And there's gonna be a poll as well, which I will launch now. And what we're really trying to gauge is basically how far along your company is in your AI journey. We're just curious to see, you know, what stage you're at, how, I guess, in tune you are with our AI products or whether you've tried them in the past and and, just haven't stuck. So cast your vote again on the poll on the right hand side. I've also linked a bunch of documents that are gonna go through a lot of the resources that we went through today, and I'll share the poll results later on. So keep them coming in as we go along. And here's what we'll be going through today. It's gonna zoom by really quick, ladies and gentlemen. We are going through a lot of the basics today, in terms of setting up and launching Foon. And then, I'll be showing you a live demo of what it's gonna look like within your workspace. So if you have your Intercom workspace open already, feel free to follow along. Otherwise, you can let me do all the clicking. It's totally okay. So let's dive into Fin. Whether you're new to Fin, have tried Fin in the past, or are currently using Fin, there are many reasons Fin can help your team become more efficient and effective. You'll be able to dramatically reduce your support volume. On average, we see, at least 50% resolution right with our customers. And then with Fin two, with all these new features, you know, we're boasting up to 82%, in terms of the new features that Fin's gonna be able to help you resolve What's also great is that Fin learns from the information that you provided. So it's gonna allow you to tailor the information that you'll be providing to your audiences. You can choose which audiences get what kind of info, and of course you'll be able to continue improving it along the way. Just like our platform, FIN is easily activated. So all you need to do is add it to your workflows or use a simple setup and sync it with your information all within minutes. And lastly, FIN is multilingual. So it's gonna allow you to provide FIN information in languages for all of your customers, and it's gonna be able to recognize the language of the end user and continue the conversation in that language as well. We'll get more into that later. You'll hear automation being thrown around a lot at Intercom, but really there are three main uses, and we'll go through those right now. So when we talk about automation, we'll be talking about these three main factors. So AI answers by Fin, custom answers by thin, and then simple workflows. At the bottom, you'll see what we used to call them at Intercom. And then, on top of custom answers, ML, that that stands for machine learning, and we'll talk about that a bit later as well. But when we're talking about true, AI using language learning models, those are going to be the AI answers provided by ThinAI agent, which we're going to dive deeply into today. So while we go through adding Fin into your workflows, in this series today we'll be covering the two that fall under Fin AI, which is the AI answers and also custom answers. So you're preparing to launch Fin. Let's go through what that looks like. So, whenever we talk about launch, we have to start from the knowledge, resources, and what information does Fin necessarily pull from. Right? So like I said earlier, Fin only learns from the content that you provided, and there are a bunch of different content sources you can feed it to learn to provide correct answers to your customers. So this can be connecting Fin to your intercom help center. You can sync it by providing a URL for external support content. You can upload PDFs of material and content. For example, like an instruction manual for a device of some sort. Snippets are just like short form text, or short bits of information that Fin can learn from. And then custom answers, which allow you to provide specific answers to certain questions that Fin will try to, detect from the customer. So I always like to say, Fin is only as powerful as the information that you provided. So Fin constructs answers using only the most relevant information from multiple knowledge sources, allowing it to create more comprehensive responses and enhancing its ability to solve complex questions. And I just wanna show you what each of these knowledge sources would look like. So an easy way to update FIN is through your intercom help center articles. In fact, I think that's the one of the strongest, types of knowledge source you can provide to fin. So as you add or update these articles, fin will also automatically update alongside them and be able to provide the most up to date information for your customers. So when you're creating or updating your help center articles, it's always important to keep a fin in mind. So we recommend talking with any customer facing roles to see, things like what are the most common and easily answered questions that you get from your customers, or what are some overarching patterns you noticed in your conversation history. We think that's a great first step, that you can take in order to save a lot of time that your team typically spends on answering simple questions that could be easily answered by Fin within seconds. After that, take a look at any FAQs you already have or macros you've created. If they're already created, it's easy to provide that information for Fin to use right away. So while you're going through these articles in your help center, you wanna be making sure that the content is optimized for Fin to use. The more simple, straightforward, and comprehensive your articles are, the easier it's gonna be for Finn as well as agents, and humans to consume them. If it's confusing for a human to read, it's also gonna be confusing for Finn to read as well. So, focus on the following when creating your help center content for Finn. When we talk about simplifying language, think about how your users typically frame questions and ensure that the answer is worded in a relevant and accessible way. So when we talk about this, avoid basic yes or no answers instead. I know it sounds very straightforward, but use full sentences to, disambiguate the answer for Finn. And when we talk about a scannable structure, we wanna be able to use rich formatting like headings, tables, bullet points. This makes it easier for Finn to scan and pinpoint retrieve the answer your customer or teammate is looking for. And also, just think about who the content is aimed at as well. So if you have a variety of user types and your support content differs for each one, ensure that each piece of content includes a clear reference to the type of user the content is geared towards. You can use audience rules to target content at specific customer segments within AI agents, and I'll show you how to do that, in a bit. As well, you wanna make sure that you are linking up with the product experts to make sure that you are simplifying industry or product specific language. So writing out any acronyms alongside the acronym or providing the full sentence to answers give Finn a better idea how to pull that information. Regarding typos or misspelled words, Finn is capable of understanding and detecting questions even when there are minor grammatical mistakes. So so no need to worry about that. Once you've successfully have fin ingest all of your knowledge content, it's gonna live in what we call the knowledge hub, which is gonna be your central repository for all support content powering your help center and your agents, so both human and AI. As your business continues to change and develop, Fin is always gonna have access to the latest information to best serve your customers. So from the knowledge hub, you can organize content into folders for easy internal navigation and toggle content on or off for use by, Copilot and AI agent independent of each other. Coming soon, you'll also have the space to optimize your content based on previous conversations, including AI powered insights and recommendations in terms of, recommended sources or recommended content that Finn, thinks your customers can make use of. What is a snippet? Like I said earlier, another content source that is easy and quick to add. It's unpublished content, so it can provide quick information to Fin's AI answers. So instead of full article, snippets are helpful as they're quick sources of information that don't need to be formatted to be customer facing. So it's simply little pieces of text that you're feeding into Fin as a knowledge resource. So here are some general use cases for snippets that may apply to, you and your business. Snippets are ideal if you wanna add a short piece of information such as an FAQ or time sensitive notice and make this information available to Fin to serve to your customers right away. So an easy example I always use is, let's say, like, a recent update or patch that's been applied to your software, your program perhaps. Instead of having to write out a full article, perhaps just reference, let's say, the number of, I don't know, potential error code that customers might encounter. FIN is gonna be able to detect that when customers ask about it, and be able to provide a response in terms of what they should expect and what the issue resolution is. We also have the option for, human agents to be prompted to add snippets from answers they provide to customers as they chat. So if this is turned on, you can choose which teammates have the ability to add snippets from their responses, making sure that only your most knowledgeable teammates can inform Fin. So if you look at the screenshot that we have on the slide right now, what I'm basically saying is that, let's say high quality responses, you're gonna be able to feed that content into Fin as well. And Fin's gonna be able to actually learn from conversation history from your more seasoned and veteran human, agents. So it's gonna have the ability to to look back at conversation history and be recommended, let's say, oh, this was a very powerful response. I'm gonna start using that for future responses. What is a custom answer? Like I said earlier, custom answers use machine learning and are providing an answer that you feed it. So, what custom answers do is that they allow you to train Fin to recognize certain questions, and then provide a very bespoke answer. You'll have to train Finn in order to detect this question, which is why we use machine learning for this part. And you'll be able to provide it with a variety of different ways a question can be asked to be able to detect similar questions and then provide it the answer you want when this is asked. Along with providing a bespoke answer, you can also launch an accompanying workflow, which will allow Fin to take additional steps. So you can have Fin ask clarifying questions, collect data, tagging, assigning, and more. So, you're probably wondering when does this come into play, when would we use AI versus custom answers. The real difference between AI and custom answers are that when Fin is using AI answers, it's coming up with a response based on all the content that you provided. However, for custom answers, Fin is recognizing a specific question you trained it on and then providing the exact response you have given it. So both AI answers and custom answers make up Fin. You can really implement both. Fin's gonna be able to pull from both wherever you add Fin in. So here's a sequence of events that happens when you activate both. Finn's always gonna first look to see if there's a custom answer that matches the the question before turning to providing an AI answer if we can. So if Finn can't provide an answer or needs more context, it's gonna let the customer know to ask, a clarifying question to get the information it needs. This can happen multiple times, but Finn is eventually gonna hand the conversation over to the next steps of the workflow. So the user can talk to a human agent, let's say, until their answer gets, and their question gets answered. Sorry. So whether Finn is providing an AI answer or a custom answer, the customer experience remains the same. So what we do is we recommend you reserving custom answers for a small pool of common FAQs or questions that you want to have very specific wording on. So let's talk through when you should use custom answers. These are the general use cases. So mostly what you wanna do is maybe for very, very, nuanced or specific issues that you need, absolute control over. Typically, what I've seen it is is, using custom answers when you wanna override, let's say, the verbiage or or the language that comes from something AI generates. So maybe you wanna keep an answer very technical, very precise for example, or if you want to reserve certain topics for human support. As well, you wanna think about maybe certain instances where you want to, fin be able to detect, specific questions in order to automate specific tasks, which could be tagging or or or, or assignment like we were talking about earlier. So this is kind of the fun part. What's it gonna look like when you build your new AI agent? So here we'll be customizing your bot to your company's branding and image. So this is gonna make sure that the customers retain a smooth and consistent customer experience. So feel free to select a cool name and profile photo for your new AI agents. If you're feeling uninspired, you can just keep it as Fin. So, from here, you'll notice two other settings you can toggle. You can modify Fin's tone of voice to match your company's brand and then change Fin's answer length as well. So the tone of voice is a very new feature that we've added. There are really I think four different settings you can choose based off of your company's brand. You can select options like professional, friendly, humorous, and more. We may release more tones of voice. If you have any recommendations, drop them in the chat. And then you can select how long you want Finn's answers to be as well from here. So if you want Finn to provide more short and concise answers, feel free. Or if you want Finn to go on, you can do that as well. Personally, I've noticed that Finn loves to do, these, like, chain sentences and therefore and so on, which can provide a little too much information to a customer sometimes. So, feel free to select and test it out to your your wishing. Yeah. You'll you'll notice two examples on the screen right here. On the left, you'll see a more humorous setting for Fin, and on the right, a lot more professional. So, like I mentioned earlier, Fin can support up to 45 different languages based on the content that you have it ingest. So if you enable the latest, real time translation feature, what's Fin, will be able to do now is translate your existing knowledge content if it can't find any relevant content in the customer's language. So for example, for my Canadians in the chat, let's say, for example, you have English and French content. If the customer asks a question in French, Fin will first scan to see if you have any French content and then provide a response based off of that. Let's say in the now the customer asks something in Italian and you have no Italian content. All you'll have to do now is toggle on the new translation feature, and that's gonna enable Finn to translate your, let's say, English or French content to, the end user's language. And we support up to 45 different languages. So take a look, and see which ones are supported, but it should be well encompassing. And I've tested this feature out, and it's really, really cool. So, yeah, from here, you'll be able to select a list of options on how you want Fin to reply to customers as well to gauge whether or not they were able to resolve the conversation or how to invite the user to connect with your human agents. You'll also have the option to set expectations for human support when configuring FIN from a workflow, and this is really reinforcing the, AI human hybrid approach that, we love to tout, as a support model in Intercom. So this will be very helpful depending on how you'd like Fin to be set up. So, you'll reduce the amount of pushback from customers when they understand the AI experience they're gonna get when interacting with your AI agent. You'll wanna select your language carefully and set these expectations as well as to encourage engagement with your call to action button when prompted to open a conversation with the AI agent. So as a reminder, these settings can be found in your messenger settings as well as the FINS settings within the workflow builder. So when is FINS officially activated? Well, FINS only becomes live once you add it to a customer facing workflow and once you activate that workflow. So you don't ever need to worry about accidentally setting Fin live and it becoming sentient. You'll have many opportunities to play around with the settings and test out Fin before you launch it. We'll be going through two options for Fin setup. We have the quick setup if you're short on time and wanna keep things simple. But if you're looking for a more customized robust setup, you wanna go with either a new customer facing workflow, or adding Fin to an existing workflow. So let's take a look at the quick setup first. This will basically create fin workflows on the back end without actually having to access the workflow builder based on the settings that you select from this menu here that you see on screen. So, I'll run through the different settings really quickly. Target audience is gonna control who's able to see and interact with the AI agent. You can add additional audience rules like tags or attributes to define different customer cohorts, And you can customize how fin AI agent is going to introduce itself. And then for reporting purposes, you can also prompt fin to request CSAT ratings. So once you're happy with all that, all you have to do is click on set fin live, and then that's it. It's gonna do exactly what I just said, set fin live. So it's gonna set this background workflow live, and then fin's gonna be ready to start interacting with your customers. So if you're following along, you'll notice that there's two options as well. We have set Fin live for chat and then over email. Email is gonna be a bit more limited in its Fin capabilities, but the setup screen is gonna look identical otherwise. So for my workflow, fellows in the chat, if you want a bit more control over the parameters and customize, how Fin's gonna interact with your customers, this will be more of the setup for you. So, Fin is gonna be available as its action, within the workflow. And then if you look on the screen through the screenshot, you'll see two options. That helped, which is gonna mark the conversation as resolved and automatically closed, or you have talked to a person. So from there, you can launch a bunch of different actions, perhaps like asking qualifying answers or or, additional questions, for example, clarifying questions in order to get more information for your human agents, or route to, let's say, different support articles that you may have in your help center. So, just like the previous screen we saw, it's gonna be the same settings that you you saw, from the the simple setup, except you you're gonna be doing it from a workflow. So now you're gonna be able to dictate when Fin's able to appear as well. Do you want Finn to be front facing as in, like, that's the first thing customers see, or would you rather prefer, Finn as a second line of defense, for example, when, after you've you've shared, some support articles to your customer? So, we have templates available if you're feeling uninspired. If you have an existing workflow, for example, you can simply just add Fin as an action block depending on where you want, to to start engaging with your customers. So the best workflow triggers for Fin AI agent are the two on screen right here. It's always when the customer starts a new conversation, whether that be through the messenger or in general through different channels. All you have to do is add the let fin answer action. And these are the parameters that you'll see. The new one you'll see there is ask for more information beforehand over, and those are those clarifying questions that I was mentioning earlier. So Finova email is relatively new since we last did this series. So it can also provide instantaneous support to your customers over email. So it's gonna look like this what you see on screen. Like I said earlier, the steps should be almost identical to what we saw for Fin over chat with the exception of some features. So Fin's not gonna introduce itself over email since it's a bit more reactive. And Fin's gonna be able to to ask for more information before handing off, and that's just the way email's set up. So, a resolution is counted when, following Finn's last answer in a conversation. The customer either provides a question, for example, and then confirms that the answer is provided by Fin is satisfactory. That's what we call a hard resolution. If they disengage or exit the conversation without requesting further assistance, that's what we call a soft resolution. If if a lot of customers are disengaging with Fin, of course, we don't want these conversations to just, like, pile up in the inbox. This is why we have the, auto close, inactive conversations toggle here, where you can set it to, the time, that you desire, basically. And this is just so, to keep your inbox very clean and managed and making sure that we're we're not spending too much time on on conversations where the customer is disengaged. And, of course, FIN also has c set ratings, so you're gonna be able to continue to monitor its performance, from all the conversations you see in the AI inbox, and making sure, for example, that you have the opportunity to optimize your conversations where perhaps, Finn was given a negative rating. So this is where you're gonna configure how the AI agent's gonna hand over to your team. So, of course, when people click on talk to a person, you're gonna be able to control what actions you're gonna automate through Fin there. Now there's a couple of ideas that I've mentioned earlier in terms of what could possibly do once people wanna request talk to a person. You can either assign it to a team or a human agent directly, or you can collect a bit more data so that, you're providing the AI agents, sorry, the human agent with a lot more context before they begin resolving the issue. And, I did mention this earlier, but you can also customize the AI agent experience by audience. So how I explain this concept is more so, without audiences, Finn's gonna look for a needle in an entire haystack. But with the find audiences, let's say I detect that the end user, is from Chicago. I only wanna source, Chicago content, when AI agent is is, conversating with them. We'll search for the needle now in a very small and specific suction of the the haystack based on, audience targeting. And that's where you're gonna be able to create, the audiences. So if you go into your settings and then audiences, you're gonna be able to control which sources, Fin AI agent should be pulling from when it detects these attributes from the end user. So, not only is it gonna make, generating the response a lot quicker, it's it's gonna be a lot more accurate and provide very relevant content to the end user instead of perhaps providing general or inaccurate knowledge. So, when all of this is all set and done, you can test FIN directly from the preview, in the AI agent tab. And we'll go through this in a demo in a bit. And then of course, once you are setting FIN, in a workflow, you can always filter for all your workflows using FIN. And once the first one is set live, you're ready to activate. So with that being said, I'm going to, demo really quickly. But before I do so, I do wanna see what our poll results shown. And it looks like, it looks like everyone here or most people here are currently using an AI agent, which kinda surprises me. And I'll talk about this in a bit, but we will be, doing another part of this webinar in a couple weeks where we're gonna be talking about more optimization for Fin, so you folks might find that useful. Thank you for that additional vote. Otherwise, what we're gonna do today is walk through general setup, but, stay tuned because you might see a lot of the new features that we've just launched, including Fin Guidance, which just launched recently. So I'm gonna reshare my screen, and thanks to all that voted, I'm gonna I'll I'll I'll keep it open for now. And make sure you keep having those questions come in as well. Okay. So let's go. So this is my test workspace. Wanna make sure my screen shared perfect. And And let's go into the AI tab. So for, today's demo, I've had Finn ingest three entire websites. And what it does is it's always gonna ingest the parent URL and all the children URL underneath that. Anytime, any of these web pages are updated, Fin's also going to be continuously adding them to its repertoire. So, it's going to sync every twenty four hours. So it's always gonna be up to date, no matter how many updates are applied to your website. So I have intercom.com, chess Com because I've been playing a lot of chess recently, and then a bunch of random different, other websites. So, this is relatively new in terms of tone of voice. I, actually haven't played around with this yet. So what I'm gonna do is I'm gonna try a humorous tone today and then making sure that, I wanna see how funny it can go. So we're gonna go with thorough. If ever you need additional settings, all you need to do is click here, It's gonna lead you to the settings, and this is where we go to multilingual support. So this is the toggle I was talking about earlier, ladies and gentlemen. Real time translation. So again, right now I only have content in English. But now, for example, we're gonna be able to have, Finn translate, the content from, let's say, our English, language help center content that we have here. And I'm gonna save these changes, and perhaps we can try it in our demo. But back to the the Fin tab, this is the simple setup steps that you're gonna take for, Fin over chat and then Fin over email. Right now, I have Fin activated through this workflow, but I just wanna show you basically, when you're interacting with Fin over email as well, you'll see, again, very similar settings. And then with the simple setup, it's even gonna pick the workflow trigger for you, and all you have to do is set Fin Live. But, again, if you want more control and customization, always prefer going through, the, the workflow route. And this is what it looks like. So this is my general, I guess, like, customer support triage, for example, when people are, first logging into my website. Right? Right now, I have Finn configured so that, if people don't select the reply button, and then wanna skip directly to Finn, it's gonna answer. And we'll go through these settings really quickly. Right now, I have, this disabled. But if I were to enable this, for example, I'm gonna be able to ask, additional questions, before it hands it over, and I'll save that really quickly, so that we can test this out. Before we do that, actually, I'll show you what the custom answers look like as well and how you can create those. So I guess I created one long time ago called how do I reset my password. And this is what it's gonna look like. You're gonna it's gonna ask you to ask the question in a bunch of different ways. And the more questions that you have Fin learn, the more it's gonna be able to detect, different variances of how the customer would ask the question. So this is the machine learning aspect. So anytime it asks something or the customer is gonna ask something remotely similar, Finn's gonna be able to detect that and then create an answer. So right now, I've created a simple test answer here. That's a simple message. But, I'm also gonna automate, for example, tagging and assigning options as well if people select this reply button. From here, you can also launch reusable workflows, or other actions available here as well. Let's say I wanna send an app. Maybe I wanna insert an article about resetting my password, for example. You can automate that as well. And then, I forgot to mention this. If a custom answer is used, it's not necessarily gonna be counted as an AI resolution as part of your billing, but more so a a custom answers just because it didn't use AI. It used machine learning. Now let's test out, this new instance of fin that we've just created. I'm gonna save this first. Let's test it out from the workflow, for example, that I've created. So, I have a hundred and eight workflows. Like I said, I'm just gonna filter for all the ones using Fin and maybe all the live ones using Fin. If ever you need to troubleshoot any issues with your AI agent, apply those two filters. You'll be able to see where and what instances Finn is firing off at the moment and then pinpoint the issue from there. So like I said, I've been playing a lot of chess recently. So I'm gonna ask Finn really quickly, what is the CaroCon defense? This is, an opening move that I'm trying to get better at. So this is really cool. Humorous, super, super long, super detailed, and I think that's why it took a bit longer to buffer this time. This is crazy. And this is what I like most about this, Hulks, is that, it's able to source its references as well so that people, you won't even necessarily need to link help center articles after the fact. People will be able to read, additionally based off of, what references are provided here. So they can do their own research if they really want. If they're really, really, I guess, like, insistent on speaking to a human afterwards, that they're gonna find a way from here, for example. If I talk here, I I just heard a ping go off in my inbox, and voila. It's probably gonna be from exactly. So through my workflow as well, what I did is I enabled an AI summary talking about, what what the customer what question the customer asked, which is gonna provide your human agents with, enough context to start replying to your, your customer immediately. And, again, all of your conversations are gonna live within this AI inbox if you have it enabled here. So you're gonna be able to monitor its performance as it continues answering questions. Cool. That's, I wanna make sure we have enough time for q and a as well. So, let's see what we got. I'm I'm gonna go through in in, in chronological order. Okay? I bet you all have been asking from the beginning, so let's see. Oh, thank you. It looks like, Rigby actually answered a couple. If you want to ask any questions, drop them in the q and a, tab, because that's where I'll be addressing them. Okay. I'll take this one from Mike. Can you give the user options when they open the messenger? Meaning, they can choose to engage with Finn or maybe, like, speak to an agent. Yeah. Yeah. Yeah. Yeah. So sorry. The the questions were getting a little little bonkers. But, yeah, basically, what you're gonna do, Mike, is you're going to create reply buttons, and you can give users the option right off the bat. And then from there, you're gonna be able to build different branches or flows based on the response they selected. So if they wanna speak to an AI agent, that's great. You're setting expectations loud and clear. Or if not, what you'll do is make the other route assigned to a human, agent or or a team, for example. Okay. Mike also asks, back to back questions, Mike. Wow. We're testing Fin out with a limited number of resolutions in our plan. I would like some help determining saved support hours and money. I can't figure out which report is the best to use to help me with that. I don't know if we're gonna have out of the the box reporting for that, Mike. And I can't speak too much about pricing, but, basically, what you wanna think about is, let's say, like, resolution rate and deflection rate. Right? How many conversations like, what's the conversation environment that a typical human agent can, handle within an hour, compared to Fin? CSAT ratings as well. And I think that's really gonna help you present, like, an ROI summary of how the Fin AI agent is performing. That's where I would start in terms of, safe support hours. I'm sorry. The previous question was from Sabrina. If you can choose, let's give the user options. For example, if they open the Messenger. Okay. This one's from Nathaniel. What would be the best way to format a snippet? Is there a creation limit for custom answers? There is a limit on custom answers. It depend it's an addition to your plan. So I can't comment how many custom answers you get. That's gonna be based off of, I don't know, like, the type of billing you select. The best way to format a snippet is, again, rich formatting, full sentences, and perhaps even natural language to instruct Finn on what to do. Can we adjust the time the closed time, the stale time before Finn considers a conversation solved? Yes. I'll I'll, sorry. All these questions are making me prompt. I'm gonna share my screen again just so we can go through some of this together. So from within the workflow, you're gonna be able to determine how long Fin should wait before marking a conversation as resolved. As you'll see here, all you have to do is go to your Fins settings, for example. And, this is the toggle for auto closing abandoned conversations. And then you also have additional toggles here, Nathaniel. So, let's say, you wanna close the conversation, and then you can select from here, for example, and then a closing message as well. So you have been asked for a CSAT rating and maybe just, like, add a message like, alright. We're gonna close out this conversation just because, we haven't heard back from you in three minutes. I didn't really get that very generous with my my customers. While I'm here, I do wanna show off a a new guidance beta that we'll be talking about in the next webinar as well. So, folks, this is really exciting. This beta just launched a couple days ago, and what it is basically is you're gonna be able to give fin guidance using natural language. So, you'll see some examples here, for handover and escalation or context and clarification. So using natural language like this, for example, if a customer asks about the Fin webinar, please hand over and escalate to Tim. It's gonna be able to accept instructions like this right now in plain English or whatever language, and then be able to follow through with, follow-up actions, which I think is really, really cool. So I haven't had the chance to to test this out yet, but I'll make sure to do that, before the next webinar. So, I'll make sure to link that here as well actually for for folks. Okay. Sorry. I just wanted to show that, to answer Nathaniel's question. Or sorry. That was Gonzalo's question. Brian asked, do you have stats around FIN CSAT versus live agent CSAT? Like, not handy right now. It's really gonna depend. The Fin CSAT's really gonna depend on how strong your content is, and then any gaps that you have in there as well. It's gonna vary by, like, industry as well, so it's kind of a hard question to answer. We also have a new feature now where Finn's gonna be able to use AI to generate CSAT if customers disengaged, and that's gonna be based off of, I guess, the conversation that they had and and and what it thinks the CSAT's gonna be. So that's gonna be an additional toggle that you'll be able to use as well. So that's a feature that's coming soon. I I actually think that, thin CSAT tends to be a bit more harsh, because people, on the other end are very, very how should I put this? Like, particular with how an AI should respond versus, like, with a human, they understand empathy a lot more, and they're a lot more generous with their ratings. So I'll say that, like, I've noticed that in the past. Is it possible to use Fin out of the working hours set up on Intercom? That's a question from Jeremy. And, yes, actually, it's a very popular use case for Fin right now. Let's say when all of your agents are offline, they're sleeping, they're on holiday, for example. You can set up a workflow that only fires off, when it is outside of office hours. So that that there could be a a way to support your customers, that are in different time zones or outside of office hours. Rachel. Will Finn be able to reply to properly specific questions and not just general questions? Finn's really good at asking clarifying questions. So, if it's just a general question, it's gonna ask for clarifying questions, especially if it's a question not relevant to the the knowledge in your knowledge base. So it's not like chat g b t. So, if we ask Finn, for example, a question about checkers, I think it would just ask, additional questions just to make sure that it's it's on the right track. Otherwise, it's simply going to, try to hand off to a human agent if it's unable to to address the question. Does the cost of FIN also include the agent Copilot to assist with answering tickets? Great question, Gonzalo. That's gonna be a separate feature. The good news about Copilot, though, it is free for up to, 10 questions per rep at the given moment. So you can check it out, but it's gonna be an additional cost, and that's gonna be on a per seat basis. Yes. I would touch on Copilot. It's not really the subject of today's webinar, but you already have access to test it out. And for for any folks wondering, Fin AI agent is what we call our external facing, AI tool. AI Copilot is more so internal facing. So instead of wheeling a chair over to ask a coworker a question, you can ask Fin directly, for example, a, Copilot. So for example, let's say one of you asked me a a very complicated question about Fin. Instead of panicking live on the webinar, what I'm gonna do is pull up Copilot and ask, how much do custom answers or how much does, Copilot cost? And so we typically have that open, on a webinar. I haven't been monitoring the chat, as I've been re going through the the deck, of course, but I hope you all have been behaving. Yeah. I don't know. Please let me know in the chat if there are any additional features you wanna go through, in the workspace together or or even within intercom. We can definitely take a look together, using using my test workspace. So Amber asks, is Finn using chat g b t modules? We used to. We've migrated, we we've, tested a bunch of LLMs recently, and, actually, we've migrated over to a new model that we've been using. So our our Fin AI agent is actually our our own patented technology, but our models now actually use, anthropics cloud model, in order to generate our answers. We found that there was a lot more, accuracy and nuance, and so we've migrated from GPT over to to Anthropic's cloud. I might take this opportunity to go through some new features that we have coming up for Fin as well. So, I love referring to this page. I wonder if I've linked it on the side. Forgive me if I haven't. But Fin two is basically all of the new features that we're gonna be shipping in 2024 and 2025. People keep asking me if fin two is an additional charge. No. You already have access to fin two. It's simply the the new name that we have for it, basically. And all these features, are not extra. They are going to be included. We're adding them on top of their existing features. And I wanna walk you through some of the, features that I'm really excited for. One of them is called AI category detection. So what it's gonna do is, automatically, assign conversation attributes to a conversation based off of the customer's responses. A good example of this, is something that we've already done within Intercom, where we have three settings, for example, for a conversation topic, which is positive, neutral, or negative. So if a customer is really irate or angry at us or, singing high praise, Finn's gonna be able to detect that and then apply that attribute to a conversation as well. So sometimes I'll get conversations assigned to me, and, Ari, I know that I'm in for a treat with, like, a very difficult customer if I see, like, a a negative attributes been applied to it. Of course, this isn't perfect, but, basically, you're using, again, natural language to describe when, Fin should apply a specific attribute to, a conversation. So we used it more so for sentiment detection. This one could mostly just be for topic detection then applied as well. And, again, this is, to help out, let's say, handover, based on different attributes or categories it's able to detect. So really cool feature there. Also, I already talked about fin guidance, and then content targeting is really cool as well. So we we talked about this briefly with, our our fin audiences, but here's just a better example of when you would use that. So, in terms of different conversation topics or or or different sources, for example, you're gonna be able to control, let's say, detect which audience it is and then which content you should really gear it towards. I'll go through that really quickly as well. Again, these are new features that we have. So we have content suggestions, and then content gaps as well. So this is, these are two things we're gonna talk about, with the next webinar on optimization in terms of how you can continue to refine your, knowledge content. I did wanna show you as well what it looks like from the knowledge hub because this is still, like, relatively new. But this is just a list of all of my knowledge content that I've had Fin ingest at the moment categorized into very loose folders. It's not that organized at the moment, but what I really like is that it's listed in column format like this. And then you can always add, let's say, additional columns here for organization. But if you want to see in a quick simple way like which data is available for AI agent or Copilot or help center, these three columns really help. So let's go to chest.com real quick, and you'll see all 3,000, child web pages I've had it ingest. And so, for example, let's say, I don't want, AI agent to pull from this one. You're gonna be able to do that as well. And these are the help center articles that I have ingested in my workspace at the moment. So I I just put, like, some quick, I don't know, instant pot recipes in here that that Fin AI agent is not pulling from at the moment. But let's say I do want it to start, pulling from here, for example. I guess it's not published yet. That's why. But for example, let's say I want to keep this recipe a little secret. I wanna make sure that Finn doesn't pull from this. What you can do is modify the audience for each individual article to make sure, for example, I don't know, only paying users can see this great braised lamb shank recipe, in a a fin AI generated answer. Otherwise, Fin's gonna be able only to pull from from sources and articles, that it's authenticated to. So you don't need to worry about, inaccurate answers. Again, you have a lot of control over what knowledge sources you have been pulled from. The the the and the mobile app. Yes. You know, there are a lot of complications that come with mobile SDKs, so I I'm, maybe we'll we'll add that as, like, a potential webinar topic in the future, Candice. That's just not gonna be handled by me. It's gonna be perhaps handled by, one of our support agents that are more geared towards, like, mobile SDKs. So, thank you for joining, if you if you're on your laptop. Can we upload a spreadsheet as a resource? Rachel asks. I wouldn't. So there are a couple sources that Fin is is not great at reading yet, which is images. So you're gonna wanna make sure that you you provide, I don't know, accompanying captions for any images as well. And spreadsheets, for example, you're not gonna be able like, customers won't be able to ask Finn to look up, cell d two, for example, and then give a a response. What it can do, however, is use API actions to do complete specific actions. And I'll show you what that looks like real quick. And and, just to answer your question, Rachel, like, I'm not I I wouldn't recommend it, but I'll show you maybe some alternatives that Fin's able to do. So, this is also pretty new. But in terms of some example actions that Fin's gonna be able to take, I'll I'll show you some, quick templates that we have available. It it works really well with, these, little integrations that we have. For example, I wanna show some I wanna show some good examples of, like, custom actions it's able to take. So I'm gonna show you here instead. So it uses API, and depending on what, integrations you have available so this is this is what I was looking for. If you have Stripe or Shopify, for example, using API endpoints, Fin's gonna be able to pull, let's say, order information, subscription info, and be able to either read data and also push or pull data as well. So, instead of a spreadsheet, I would look to viable alternatives. Perhaps I have, like, API endpoints and then have Fin pull from there. So it's gonna be able to, let's say, pull account information when a customer asks for it. Usually, those actions right now are better suited towards, like, human agents, but, again, these are, like, new possibilities that are available through these FIN custom actions that we've unlocked recently. How does FIN know whether they are paying users? Great question, Mike. It depends on what, data attributes you have currently living within Intercom. If you have a powerful, let's say, data integration with the warehouse, like your CRM, for example, or or or through, through a segment or stitch, If you have this information loaded in a CRM of any sort and have it loaded into Intercom, you're gonna be able to use that as attributes in your workflows. So, let's say if I'm setting up a target audience and I only want paid users to interact with that specific instance, you can either use branching logic or or I guess like audience rules in order to look at user attributes to see who's gonna be able to even see that instance of Finn or who's even gonna be able to interact with a a specific subset of, of knowledge content that Finn's ingested. I've been doing a lot of talking and a lot of hydrating, so this is gonna be my time to take a sip real quick. I'll share the poll one last time. It seems like a large majority of you have, used an AI in your past. But for the people that haven't, just know that you do have access to a fourteen day trial. Okay? You can reach out to, your relationship manager afterwards if you're interested in starting that. What I would, I guess, provide as as, final words of wisdom is that hopefully with the knowledge that you've gained from this webinar, making sure that you make the most of those fourteen days. So, like, even doing setup before you launch, your trial just to make sure that you're actually gaining good results, a good ROI like someone else mentioned earlier in order to present to your leadership, with with, like, enough conversation volume to say that, like, okay. Look. The AI agent worked. These are the results. Here's the data to back that up. This is why we think we should implement that. And you're gonna be able to pay, let's say, based on the amount of resolutions that you anticipate having as well. So, if you have a high volume, of course, you can buy a lot more resolutions. If you only want fin to operate, let's say, outside of office hours, you can do that as well. I want to also take this time to plug, our next session on Fin optimization as well. That's when we're gonna be going into a lot more, advanced topics in terms of like, okay. You've already launched Fin. How can you continue to refine and optimize its performance to best, suit your, customers? And that's when I'm really gonna dive into a lot of, like, the new features as well. And as, as well, a lot of, like, the reporting that comes with it, and a lot of, let's say, different, places you can check to make sure that, like, and best practice to make sure that your your knowledge content is up to date, how to identify, like, gaps in your knowledge content, and then how to mediate those gaps. So people always ask, like, what happens when Fin provides a bad answer? Can we tell it, oh, bad AI? You did a bad answer. That's not necessarily how we how how we teach it. It's it's again, it's only as good as the knowledge content you you provide to it. So, you'll have to just, make sure you you modify, let's say, the knowledge content where it pulled the bad answer from, and then fixing the verbiage there to making sure that, it's able to provide a more refined answer to your customers. And with that, I will call our webinar to, close. Thank you so much for for tuning in today. Yeah. Jeremy, I I I didn't wanna touch on it too much, especially for for newer folks joining today. We will go into, a lot more of the fin two features next week. I just really want to give people a good taster of what, came with fin two. But yeah. This fin pull from historical conversations. Okay. One last question from Lisa. Yes. I can go back to as much as, three months, in terms of historical conversations just to make sure it's not pulling, like, outdated conversations. So only only three months as a maximum. But thank you all for taking the time to, come out today. I really appreciate you asking all your questions and being so engaged. Please register for the the next webinar on March 5. We'll be going in a lot more advanced topics, and then making sure that, I don't know, you've made the most of that trial. So, yep. Good luck with any trials that you start. Reach out if you have any questions. Before I let you go, you can always reach out through the messenger to get in touch with, myself and Rigby and our team. So thank you, and everyone have a great day. Bye.