Name: How to adopt AI in Customer Service in 2025
Uploaded: 2025-04-10T02:16:08.296Z
Duration: 1 h 8 s
Description: How to adopt AI in Customer Service in 2025

Transcript for "How to adopt AI in Customer Service in 2025": Hi, everyone. We'll just give everyone a couple of minutes to trickle in. But as Will said, we'd love to know where you're tuning in from, so please say hello in the chat. Thanks for joining us. Jeff, hi. Hi, Paige. Hey, Paige. Awesome. Also, we Dan and I are in the same room, so just let us know in the chat if you can't hear us. That's why Dan is on mute because we're using the same microphone. So just let us know if there's any audio troubles or anything. Cool. We've got people from everywhere. So cool. Dan and I are both in Sydney. Our intercom office in Australia is in Sydney in Surrey Hills. So if anyone else is from Surrey Hills, you should let us know. We'll just give it another minute and then we'll officially kick off. Oh, Matt, you're in Navens. The Bahamas. Oh, sounds amazing. We've been between us. We are we are, Finn. Love that. Okay, cool, wow, there's so many people here. Thank you so much everyone for joining. I think we might kick off. So welcome everybody to our webinar for how to adopt AI in customer service. I'm Beck. I'm a relationship manager here at Intercom, and I primarily work with small to medium sized businesses. I've been with Intercom for four and a half years now, and it's been really cool. Especially over the past two years, I've been able to help many businesses to evaluate, test, and implement AI for customer service in their businesses. So a lot of what we discussed today will be real lived experiences from not just what we know, but what we've seen with other businesses who are getting started with AI and customer service as well. Perfect. I'm Daniel. I'm a senior account executive here at Intercom. So I work with a lot of large, companies, across Asia Pacific, to help design, implement, and manage an AI first customer support function. And I just want to kind of reiterate what Beck said around. We're gonna try and make this as interactive as possible today. So any questions, please put them in the chat or any, anything you want to know, please jump in there because we want to make this kind of as fun and as interactive as possible. As well as, sometimes webinars, can't be that won't be that practical, but we want to give you something to take away to then, help you with your job or business. So, please jump in and interact because that'll make it much more fun for everyone. So, yeah, looking forward to it. So I'll just pass back to Bec now. Awesome. So, yeah, as we mentioned, the theme of today's webinar is how to adopt AI in customer service. And the reason that we chose this topic is because most of you are here probably because you already know there are many benefits to using AI and customer service. And if you did want to know more of these, we actually have a customer service transformation report for 2025, and it's available in the docs section of the chat to download. So if you head over there, you can download that report if you wanna know more. But today, as Dan mentioned, we really wanted to give you some practical action items that you could take away and use in your everyday, work life. So we and as Dan said, we wanted it to be really interactive, and we really wanna enable you today to be sophisticated evaluators and adopters of AI and customer service for your business. So we are going to have our first poll for today. So Will is gonna pop that in the chat because we would love to know where you are at in your AI journey. So are you starting to learn more? Are you looking to trial? Are you already implementing AI, or have you already implemented AI in customer service and are looking at how you can optimize it as well? So you should hopefully see a poll pop up. Oh, awesome. There it is. So we'd love to see, some of your results start flowing in. But we think no matter where you're at in your AI journey, the content that we'll cover today, is important no matter what stage you're at. So we hope that we've structured in it in a way, will it will provide something for everyone. So a lot of you have probably been asked by your business to just implement AI and customer support, which is a big ask, especially if you've never done that before. So that's why we've structured today in three sections, preparing, testing, and selecting. So through Dan and I's combined experience helping businesses implement AI and customer service, as we said before, there's many things we've learned to help make this process a lot easier, clearer, and faster for businesses. So we wanted to equip you with this knowledge as well. Also, just to call out, please keep throwing questions in the chat. We do have some time for q and a at the end, and we'll get to these questions as well. So next, we actually have something a little bit new that we wanna try today. So we don't think this has been done on a webinar before, so we're breaking ground here. But before we dive into the first section today, there's something we wanted to do that we'll get to in a later section of the webinar in the testing section, And that is show Fin live. So Fin is our AI agent, and we wanna show Fin live for one of your businesses. So now is a really great time to test AI for customer service because we have so many ways that you can test your help content within inside and outside of Intercom. And so what we've got on the screen today is something that we call our bring your own help center Fin demo. So you can use this whether you're an Intercom customer or not, just as long as you have a help center that's publicly facing, then you can use this feature. So if you're feeling brave, we would love you to grab your help center URL and pop it in the chat here, because we would love to come back to it later and test it live with Bing. So feel free to put the URL in the chat. We would love to see some of them come through, and we will pick one lucky winner to have their help content tested with Fin live. So the reason we're gathering it now is because it takes, a couple of minutes to generate, so we will come back to it later. But we'd love to see some of your help center URLs. Again, it doesn't have to be an intercom help center. It can be, from anywhere. So we'd love to see some of those coming through in the chat. With that, I will pass over to Dan to lead us into the first section. Yeah. Perfect. Thanks, Bec. And so just, recapping on the poll, it looks like the majority of you are preparing to explore AI or you're in the testing phase, of AI there. So I think this kind of next couple of sections will be very beneficial. And so I just wanted to kind of open up by starting to say, when you're looking to explore an AI first customer support strategy, you need to first decide, like, will this benefit us, and what do we want those benefits to look like? As in all projects and not just AI ones, do these benefits outweigh the costs and efforts of implementation? So today, I'll run through just a simple framework of questions that, Bec and I typically run through on a day to day basis with our customers, just to get an understanding of, where the AI customer support team AI customer support platform will be able to assist. We say this is a great way to kind of understand the key problems that you're looking to resolve, and what you really want the outcomes to be, and how you can kind of show them to management, as a best practice, in market. And so, like, when people come to talk to us, there are kind of two, common scenarios that people usually come to speak to us. The first is that they've been told by management to get AI with no kind of clear guidance on what they wanted to do or resolve. And this really causes quite a few problems in terms of once you kind of had a look around AI, there's not much alignment between the things you've looked at and then what management has kind of had in their mind around what, it should be doing. The second one is when people are looking for a better or more efficient way to do their jobs or make processes better for customers, support agents, and or management. And so over the next few slides, whichever kind of scenario you're in, I have a simple kind of series of questions and, framework you can kind of run through that will really give you a good way to understand, what the AI support platform should look like, how it will help your team, and what it will aim to achieve. So when you are presenting a business case back, to management on why you should implement AI for support, you can always kind of refer back to this kind of framework to begin. So I might just jump in. It's a series of five questions. We've named it the project criteria checklist. Pretty straightforward. But the initial kind of, question, that, you need to decide on when rolling out to AI support, what are my team's current challenges? The purpose of this is to frame up the problem that you want to be solved by AI. Without this diagnosis, it's quite hard to decide if the AI support is actually driving useful change for you and or your business. So for each of these questions, I'm gonna talk through a different kind of lens, I guess. And so what we normally see is people's challenges when they do come to us and talk through this as well. And so those lenses that we kind of think about are the customer, the support agent, and then management as well. So if you were to think about what are my team's current challenges, usually what we hear from, like, the customer perspective is that there's long wait times or varying degree of customer experience based on the channel they're coming through, like is it email, chat, WhatsApp, or location. And so those differing experiences cause kind of different issues. The issues we kind of typically find that people talk through with us on the agent side is that they're unable to access information quickly, or they're spending too much valuable time on repetitive tasks, with difficulty scaling when there's large volumes of customers coming through, for different reasons. And then the final one kind of on that management lens, the problems that we typically see people, having is that there's no kind of centralized way, to have oversight reporting, which kind of makes decision making hard from a management perspective. So once you've kind of answered this question around like, what are my team's challenges, we typically move through into the next question, which is what is the impact of those challenges? So once you know kind of what the problems are, it's good to identify, what are the impacts of these challenges and are they big enough for us to look at a solution that might involve an AI support platform? And is that going to fix or alleviate these problems? And so if I expand on that in a practical manner, or from a kind of three lenses I spoke to earlier, so for long wait times, is the impact of this, high abandonment rates and does that result in customer churn is one we typically see all the time. For support agents, it's when they're struggling with the trying to find information, it increases the volumes coming through, which leads to higher average handling times, and this means that CSATs might suffer, with worse experiences. And then for management, not having that single source of truth slows down decision making, which means that kind of these problems are extrapolated and take longer. And so I guess that kind of leads into that third section. So now you know what the problem is, you know, what the impact of that problem is. The next kind of thinking question is, what is the solution you want to look like? So this can be kind of defined by the capabilities you want it to have, so that you can implement it. And this kind of gives you a nice list of must haves and nice to haves with your AI support platform. And so this list can be kind of dynamic, over time because sometimes when you meet with different vendors and, different solution different solutions, they can kind of frame your thinking in different ways about this. But when we chat to people, typically, it's things such as a generative AI chat, omnichannel integration, outbound messaging, integrations into different CRM platforms or, ticketing solutions, as must haves so that they can then utilize the AI to, resolve those queries. The next kind of question that typically we ask in this framework is that, you've got what your solution is and what you need it to do. It's, what I want to achieve by solving these problems? So what are the positives that are going to come out of implementing the solution we've just chatted to? And so usually we work through with customers, to be, reduction in l one queries that are passed to humans. So they're simple questions. We want the AI to resolve them so our humans can work on, more high value tasks or value driven tasks to, increase revenue from a particular customer base, for example. We, do sorry. There's there's other thing that kind of, we're kind of looking for is, the the need to increase headcount with rising volume. So if we don't need to increase the type size of the team, because the AI is resolving a lot of queries, that's also good. Giving personalizations of experiences, as well as seamless data integration. So there's some of the things we see, when people look to resolve and get the positive business outcomes of these systems there. And then the final question is just then how do you roll this up to a executive decision maker level? So once you've kind of mapped out what's wrong, the impact, what you want to build, and then, what it will achieve, how do you tie this back to a business justification? And usually, this is always tied to kind of three key things. It's either a cost saving, an increase in revenue, or a greater customer experience. And so if you can answer the first four, tie it back to number five, it sets a really, really strong foundation, for going into, deciding on an AI strategy and vendor. So I've just summarized these in a quick table. It's a super simple framework. We use it almost every day with customers. If you want to chat through what you've got with us on this one, we'd love to chat with you about it and we kind of give you examples. I'm just gonna pass now to Bec, who's gonna talk through, once you've got the framework, how do you add metrics to this and, examples so you can kind of understand if it's actually trending in the right direction. And Bec will work through a real world example of, one of her customers that she did this framework with. So thanks, Dan. Cool. Thanks, Dan. So I also did just wanna call out something that I saw pop up in the chat. So the framework that Dan just outlined, even if you've already implemented AI in customer service for your organization, that framework is still really great to call back to, when you're evaluating, you know, the ongoing success of the AI tool. I find that customers who really are clear on, like, the challenges that they're solving with AI and the benefits that it's driving for them, their team, their customers, and their business. Those are the businesses who continue to evolve and continue to see resolution rates increase as well. So it's definitely an ongoing framework that you can go back to any stage in your AI journey as well. But one of the points that I wanted to clarify as well was that you can start to or I wanted to give you some tips on how you can start to identify your answers to the framework that Dan just went through. So when you're looking at, questions one, two, four, and five, you wanna be thinking overall, like, what am I trying to achieve by implementing AI for my customer support team, and how also will I know when I see the benefit? So in the next section, we will cover the trial and testing process, but it's good to start about to start thinking about these things and how you're gonna measure success really early on. So for some of the businesses that I work with, they consider things like ease of implementation. So will the tool take us months and months and months to implement and test and get right and then maintain? Are we talking just a couple of months or even weeks as well? Another thing that I hear a lot is time to value. So once it is implemented and and live, how quickly can I start to see results? Another thing that people use to measure success is the product road map that a tool might have. So is it just one and done? Does the tool, like, exist completely yet? Are there plans for it to continue evolving? And speaking of, if you haven't seen Intercom's, latest product road map and you're interested, definitely comment in the chat road map, and someone will reach out to you to be able to go through that with you because we typically do a half road map. You would know if you're an income customer. But now that we are making such grounds with, AI, we're actually doing a quarterly road map, which is really cool. So if you'd like us to go through that with you, comment road map in the chat and, someone will be able to reach out and run through that with you. But also about measuring success. So another key part of that is, like, the metrics that you're actually gonna use to measure that. So some common ones that we hear are like, what is my resolution rate? How many conversations is the AI agent involved in? Is the first response time of my human agents producing? What volume of conversations have been able to be taken away from my team because I've started using AI? But also a really popular one is by using this AI tool, has my cost per resolution actually reduced compared to a human agent as well? But another thing that Dan covered in the checklist was what does my ideal solution look like, Which can be a pretty big question if this is something that you're just starting to think about. So some great tips is to, like, firstly, look for inspiration. Like, the businesses that you interact with on a day to day basis, are any of them using AI in their customer support? And what does that really great experience look like as well? Another tip is to look at there's a lot of customer stories out there or businesses who have started using AI, so definitely dig into those. We've got some on our website, as well to see how businesses like yourself could be leveraging AI as well. But also, like, I would encourage you to look at how the company of the AI tool that you're evaluating is using their own AI. So Intercom, we are an Intercom customer. So I always tell my customers to message our support and see how we're using our AI bot and test our workflows out to see how we we've shaped the experience because everything we're discussing now, that's a process that our team had to go through in rolling out our own fin and our own AI because we had to decide how we wanted it to work for us as well and how to provide the best experience for our customers too. But as well, like another great tip, you could definitely leverage, sites like g two. They have a lot of recommendations for AI as well and to show you what some of those best, examples look like too. But now I did wanna show you, as Dan mentioned, an example of one of my customers who I helped roll in out with, and, basically, the how they answered the checklist. So I just wanna be clear. They did not have the answers to all of these at the very start. That's something that we're here to help you figure out as well. But by coming up with the answers to all of these questions, we were able to have a really clear idea of how Fin, which is our AI agent, was able to benefit them. So if we start at the top so the first one was, like, what are my team what are my or the team's current challenges? So for my customer, they're in the health tech industry, and they were noticing that their support volumes were rising. And their solution to this at the time was to increase headcount, which as we know can be expensive and also time consuming. And as a result of this, the support volume rising and increasing headcount, it meant that their first response time was increasing as well, which was negatively impacting CSAT and also meant their cost to serve was actually increasing as well. So when we actually started the discussion around what their ideal solution looks like, because this was a challenge that they were having right now, they wanted something that was easy to implement that could leverage generative AI that could be powered by their existing support content, so information that they already had on hand. And also because this was something that they really wanted to track the value of and be able to report back to upper management as well, but also see the direct impact on their team. They wanted something with simple reporting as well. And what did they want to achieve by solving the challenges that they were having? Well, they wanted to see a reduction in human resolved queries because for them, this would mean that they could then reduce the need to hire additional headcount. And with that as well, redirecting some of that traffic that was resolved by humans to AI meant that their teammates would actually have more time to spend on complex queries, which is some of the ones which really drive the most value for their team. And because of that, we would see an increase in patient experience as well. And how would this help the wider business? Well, there were some pretty clear cost savings that we could see in terms of headcount, but also for them an enhanced customer experience For them means repeat business, which will then mean more revenue as well. So I know you might be thinking, where are they now? Well, what basically, what we did was we ran with them a full week proof of concept. And before that, they we worked together to identify that they wanted to see a 30% resolution rate, which meant for them, if we could resolve a thousand conversations a month, we had identified that that's essentially what we would need to be able to see a percentage of their staff retrained in other areas, and to be able to save that headcount as well. And through that trial process, we were able to blow those, goals out of the park. We are now seeing 3,000 monthly conversations resolved with Finn. They have a 51% resolution rate. They didn't need to hire that additional headcount that they thought they might have to, and we had got them live within in under a month as well. So the way that their support team operates was able to completely shift in a way that enhance the experience for their team, but also for their customers as well. So wanted to share this with you as an example of how the checklist really, really helped us partner with this business to make sure that they could have the most success as well, but to see continued success also. Perfect. So thanks, Bec. I think that's kind of a really good example of kind of taking the checklist, put it into practice, and then seeing some real results as well. And so if, anyone wants to kind of run through the checklist with us, we're more than happy to have a call with you to talk through it, to give you examples of what we say with other customers as well, because we do speak to people all across, Asia Pacific in, all industries as well. And we've been doing it for kind of, quite a while now. So even if, you're not looking to buy or anything, we can just give you guidance, on this as well. So now I just want to kind of do another poll just based on that kind of criteria we've just run through. You can select multiple answers in this poll, but we're just reflecting back on, I guess, the project criteria checklist. Out of those kind of five criteria, it'd be great just to hear from you of which ones you think you could probably answer at the moment, based on where you are in the in the journey. So I'll just ask Will to pop up that poll now, and you can select multiple answers. So if you think you can answer one, four, and five or one, three, and four, just pop that in and we'll kinda see how that goes. And so I think that's kinda we'll just pop that poll up now. Yeah. No pressure. Like, you don't have to have the answers as I said before. Like, this is really just to help you understand where you're at in terms of evaluating AI. And if we were to be like or if you were to start exploring it in your company today, like, where would you be? How how ready would you be? Or maybe this will help you identify some areas which you need to, look into a little bit more as well. Perfect. And it looks like a lot of people know the challenges and impacts. But what does the solution look like and how does it help the wider business seem to be kind of a little bit low on this, which is quite interesting. Yeah. But then, I think we just wanted to kind of keep going on the poll because it is super interesting. Yeah, that's cool. Oh, thanks, guys. And so, yeah, I think once we kind of know the preparation is done, we just wanted to go run through the next kind of section in terms of testing, why testing is important and how to do it effectively. So you can kind of see the results, so assign align it back to the checklist we've done we've done there. But, yeah, I think that could be Yeah. Cool. So in terms of testing, so as as we mentioned before, some of you might already be using an AI agent or another AI tool. And you would probably know that quality content and content that is optimized for AI is one of the keys to success of your AI agent. It's not everything, but it definitely plays a really big part. And over the past three months, we've made here at Intercom, we've made a lot of feature improvements to enhance your answers and the quality of your resolutions as well even further, and guidance is one of those, which is basically like instructions that you can give to Finn on how to handle certain queries. So, again, it's great to see that some of you are interested in seeing the road map, and that's one of the big things on there. And it's actually live in the intercom now as well, so really keen to share some more about that with you. But content, even with these features, is still really important, and it's something that some peep sometimes people can undervalue when it comes to AI. So in this section, I wanted to provide some advice on how to ensure that your content is optimized and ready for AI, which will make your testing process that Dan will run through even easier. But in saying this, people sometimes get stuck at the content stage because they either think they don't have enough content or that it's not optimized for AI either. So I would love you to comment on the chat if you have, like, just say yes or no or even to share what type of content you have. But whether you have an FAQ page, a knowledge hub, public help articles, saved replies, or macros, or really any internal docs that your team reference to frequently help respond to customers. So if you have any of those types of content, even just one, then you can start testing AI, which is awesome. So I just wanted to show you, like, even if you have any content, you can get started. So, basically, any information that you have, can you can help get ready for AI. Because I've had customers who felt that their content wasn't optimized, but they switched on fin and saw 30% resolution rates instantly. And when they continue to invest in optimizing their content, they were able to reach over 50%, sixty % resolution rates as well. So what one key takeaway we want you to have as well today is that if you have content ready, you can turn Finon and have really, really good results to start with. But if you want your content and your resolutions to be really great and if you want to be resolving, yeah, as we said, 60% of your queries, that's when we need to continue to invest in content, continue to iterate when we see maybe answers that come through that aren't quite right because we need to really optimize the AI as well. So I just wanted to highlight that. Yeah. You can have great results right out of the box, but if you want it to be really great and continue to evolve, that's when we do need to invest in continually updating content as well and making improvements there. Cool. So now is the time. The first time ever, we're gonna be testing live, someone's help center we've been on a webinar. So so excited to see so many options come through. I have chosen one from the list. So the one that I have chosen is healthy life. So thank you so much for sending that through. Yes. I will now share my screen so we can see this in action. So as a reminder, this is something that you can share or this is something that you can use whether or not you are an intercom customer. Anyone can use this as long as you have a publicly facing help center. So let's do this. I don't know why I'm well, it just it's it's live. So thank you, HolyLife, for sending this through. So this is basically what it will look like, once the demo is generated for you. So it'll send a link to your email, which then allows you to test it here in this capacity. The UI of this has come a long way as well. So you can actually utilize features that are in thin, like, tone of voice. So as you can see down here, we've got neutral, friendly, humorous. That is a funny one if you wanna try that out. But today, I'll just go with friendly. So what it also does is it actually generates a list of questions that you can use based on your help content. So it's basically the AI is guessing questions that might be asked based on the help articles that you have available. So let's maybe say, what payment methods does help you accept? So this is just populating that question in there. You can, if you want to, though, free type the message. So you could really type anything and see see what would work. So there we go. So it has pulled together an answer here, basically saying the different payment methods that are available, offering extra help if needed. And then what it's also doing is it's referencing the help sources that it uses. So, I have got Will as well to share in the doc section, some resources that you can use to optimize your help content. But in terms of, like, what sources you can use, so you can use publicly facing, help content. You can use snippets, which are like text, free text boxes, PDFs, web page URLs. Like, any of those sources can be referenced for Finn as a source. But in terms of, like, calling out to the source, public health articles, we'll be able to do that for you as well. So I hope, anyone from the Healthy Life team, let us know if this answer is is, accurate, but you can test this. And I'll send you the link afterwards so you can have a play with this yourself as well. But anyone who sent their, URL, and if you, if I didn't pick your business today, I'll send in the chat the link where you can try this yourself as well. But basically, what you can do here as well, you could keep asking questions if you wanted or you could say that helped or get more help, which in this example, it won't connect you to the team. But, if it was live, that's what it would do there as well. So, yeah, I hope this was helpful to see that, like, literally any help content you have can be optimized right out or can you can get started right away. And this is a good way for you to start testing it yourself as well just to see how it would go. Let's I wonder what it would look like if it was humorous, actually. We make sure that. It often says well well. I see when it's when it's being numerous. Oh, wait. Maybe I'll this. So, yeah, it's really fun to play around with, but it's also a great way to, like, genuinely test out. Oh, it did say well well well. There we go. So, yeah, you can just have a play around, but as I said, it's a really great way to tell your content will respond, or how how it will power AI. Cool. So I'll stop sharing and we'll head back into the deck. But thank you so much for sharing that. Oh, it's good to see some people love the the human. Awesome. Oh, cool. And great to see, Katrina, that that was the correct answer. That's fantastic. Awesome. So just jumping back into the deck. So the other thing that I wanted to mention, I know some of you shared that you are existing FIND customers or that you are existing Intercom customers as well, which is fantastic. So we have actually just released a new tool inside your Intercom workspace called Fin Testing, where you can basically, connect past conversations that you've had in your workspace. But you could also upload questions manually or via CSV, And then you can inspect these answers to see how Finn pulled them together. So it's basically like the, demo that we just did, on the website, but then this is in the back end. So you can actually dig into the answers and see what sources can use. And it's a really great way where if you wanted to do a bit of extra testing before you set FinLive, you can actually see, okay, maybe I should update this content here because that answer is not quite right, or, oh, we're missing this part of the content. So, yeah, it's just a really great way to see, or notice any content gaps as well. But I did wanna leave you also with just some general content best practices, which are good to remember when you're developing your content or even if you're looking to update it as well. As I mentioned, there are some links in the docs, that Will has shared for us, which go into, like, how to optimize your content for AI. But some really great ones to remember is avoid ambiguity. So, for example, instead of saying, in your content, like, it's easier when you invite teammates, a better way to say this would be collaborating on a project is easier when you invite teammates to join your team. So just you think if you would explain it to your teammate and you're being quite ambiguous, they probably wouldn't get it. So, AI wouldn't either. But formatting actually matters a lot as well in your articles. So leveraging things like headers to identify new sections or organizing, like, lists into bullet lists or if you have a lot of information to display display a table is a really great way for AI to read that information. So that's, some good things to remember there. But then also restating questions if you have, like, an FAQ doc rather than just saying yes or no. So, yeah, the same example I had before, like, can I invite teammates? Yes. Like, you would probably wanna say, like, can I invite teammates? Yes. You can invite teammates to intercom. You can do this by blah blah blah. So that's another good example there as well. But, yeah, definitely check out the content in the docs because that will help you a lot in optimizing as well. Cool. Perfect. Thanks, Beck. So I think the next phase of this is you've got your problem statement, set set out. You know the impacts. You know how you wanna resolve it. You've got your content in a great place. But then there's kind of a four step process when you're actually going out, to do the test. And so how I structure this with customers when I look to do the testing or POC is prioritization, implementation, evaluation, and then iteration. And these four steps are all, as equally as important as each of them. Because when you go through this process, you need to understand the cyclical nature of AI, because it is it is dynamic and it is ever changing. So you do need to kind of go through this process. Even if it's live or you're testing it, this process still applies, to you once you've kind of set it up. And so what I wanna run through in this section is, how to how I kind of do each of these, and then just examples of what it looks like, for customers and what to think about, for each of those points as well. If there's any questions, this phase is kind of, fairly dynamic in terms of how customers set it out as well, because everyone is different how you're setting it out because of what the goals you're trying to achieve are. So if the examples I give aren't relevant to you, let me know what your relevant examples are, and we can kinda talk through them at the end, or, in another session if that would be more helpful to you. So if I just jump into prioritization. So the main question you want to think about is when you're setting your test live, what is the most impactful, thing to test during the trial? We just want to remember that we're not looking to replace the full system you have set up now because that might have been in your business for one, five, ten, twenty years. So we're not gonna be able to fully replicate the full functionality of that system, but we wanna take the most impactful, use cases and test the AI to see if it does make a difference based on the problem we're looking to resolve and the metrics we set out that that kind of mentioned earlier, of defining if this is a success or not. And so typically, when I roll this out, we look at kind of the AI from three different lenses. And this is typically because we have, I guess, three different AI products, at Intercom. And so when I talk about AI for customers, I'm talking about Fin AI Agent, which is our customer facing chatbot, which Beck, ran through just before. When I talk about AI for agents, it's what we call, Intercom Copilot, and that sits within the, agent's inbox, and is an internal facing, AI Copilot, which has access to external facing information, and internal facing help docs as well, as well as past tickets that, agents have resolved. And then when I talk about AI for management, we have, AI analytics, which helps, helps do reporting, and insights based on, all the interactions and data that, comes through. And so typically, what we'll see, from this prioritization component is we'll pick, one use case from each of these categories, or if you're just testing, AI for customers, we will have, a few for the for the AI chatbot. But typically, for the AI, chatbot or finned AI agent, we'll have it located in a segment where there is a high volume of traffic so we can see what the resolution rates are. So in this example, I've said that we would set the, AI agent, live on a contact us page where we get 60% of the volume. For the AI agents, we would we would set it up so that new agents have access to it because the the results and differing in, average handling time and speed to resolution would be more pronounced. So we'll be able to see the results, in a in a better way. And then for the AI for management or the AI analytics, we'll we'll, restrict it to just setting up a couple of reports because typically in modern day corporate, there's hundreds of reports people get every day. So we don't wanna replicate all of them. We just wanna prove out that the AI can do a great job of the important ones that will be able to move the needle. So once we've prioritized the use cases and what we wanna see and how we're gonna measure it, we then move to the implementation stage. And so this is where are we gonna put it, for how long, and and how long will for each of these be deployed. Typically, I do a two to four week period, because AI can be set up and the impacts can be felt very fast. As Bec kind of showed, we we set it up in the twenty minutes we were on the webinar, had it live, and it was answering questions. So if there are common questions coming through, I think in four weeks, we'll be able to see kind of stellar results in, where it comes through from. And then the next thing is mapping out where in the the timeline of the trial you're having these as well. So I would always have the AI Fin AI agent on for the entirety so you can kinda gather as much, resolution information, as possible, as well as the quality of those answers. The next one for the for the Copilot, I'll typically have that turned on for the latter half of the trial, so that we have a a time to optimize the customer facing AI agent, and then be able to optimize it for the sorry, for the customers and then optimize it for the agents. And then the management side of things, I'll typically do that towards the end of the trial just because once we have all the data from AI facing customer, AI for the agents, it's then much easier to pull meaningful reports. And so that's the implementation and the structuring of the trial. The next component is the evaluation. So essentially, it's just drawing back from, what Beck kind of mentioned at the start of what are the metrics we're looking to, evaluate here, how have we hit them, if we haven't hit them, what's going on, and just measuring these over time. I typically do a weekly, recap or review of what's happened, based on each of these. Sometimes I will do two, two reviews a week to make sure that, everything's on track, and that it's caused no issues, during the deployment or implementation. The final and probably most important component is iteration, because AI is not like a I kind of mentioned at the start, it's not a one and done, tool, but it's always evolving and it has as has a maintenance and care. So that's why this iteration phase is so important, especially during testing and once it's live too. So when you're looking at reviewing your AI agents for customers, our results, we typically see around the 30 to 40% mark out of the box. But people wanna go higher up to that sixty, seventy, 80 percent mark. And we have the kind of benchmark of the 56% is the average across, all of our customers. And so what will happen as an example, you will deploy it, you'll get your 30%, but you'd like to see more. So how do you do that? So iterating or using kind of inbuilt tools that we have, one is content readiness, and so that has a report of what did the AI not what was the AI not able to answer, categorizes those questions, and then recommends content to build out and design with your team. It also use categorizations of questions that had low CSATs so that if the AI is giving a poor quality answer to review the content it's pulling for that answer. And so being able to use that information to then generate better or different content allows you to increase those, resolution rates over time. And so the next kind of iteration after that is adding in tasks and, processes and actions so that the agent kind of then starts, doing processes that the agent might have done to get those kind of higher rates there. The next kind of iteration is for the agent. So was AI giving answers in the right tone of voice for the agents to their news? Was it pulling on relevant documentation? And was the uptake of the agents, necessary or applicable for what they kinda needed to see there? And so what training can be provided there, or do they need help implementing, or was it useful? And so just having kind of a qualitative feedback, on that one from the agents and how they feel about using it is a really great way to do that. And then AI for management, all the different kind of reporting and wider oversight for them there. And so with this kind of four step process of implementing your trial of prioritization, implementation, evaluation, and iteration, you're really able to get a sense of, did this tool make the change and fix the problems we outlined at the start? And so that's kind of a really strong methodology that we, really like to use for existing customers who are setting up AI or new customers who wanna test AI, but it really, really works well. So if you wanna use that, for all of your testing across not just us, but other vendors, that's all good. And if you like any, we have a bunch of other kind of decks and frameworks that we run through with customers when setting this up. So if you need access to those or wanna chat through some of those, we are more than happy to help on that as well. But I might just pass to Bec now that we've gone through preparation, testing, now to talk about Silicon. Cool. Thanks, Sam. So as Dan mentioned, now that you've done the preparation by outlining the value you believe AI can bring to your team and the business, you know what your ideal you want your ideal solution to look like and the value you want it to bring. And you've also done the testing and align the test with your criteria for a successful trial as well. So what now? So this is really where we can start to bring together everything that we've learned in the prepare section and the testing section as well. It's so important that throughout the evaluation of a tool, you keep going back to that project criteria checklist and thinking is just as a reminder, what are my slash my team's current challenges? What is the result of these challenges? What does my ideal solution look like? What do I want to achieve by solving this challenge? And how will solving this challenge help the wider business? And during the testing phase as well, you wanna be going back to that list too. Constantly assessing whether it's meeting out measurements for success. And as I said, not just the figures like resolution rate, but thinking about things like, was it easy to implement? What was the partnership like with the vendor that I was doing this trial with? What did our team think of the trial? How did they respond to it? And, importantly as well, what did our customers customers think? So once you've found the solution that is right for you, that meets your criteria, and, yeah, really have you've decided is the one for you, then that's when the purchasing process continues. So for many businesses, this might be your first time purchasing AI. So one thing that Dan and I have come across is that there might be existing legal processes that your company has, that you should be aware of. So I would start letting teams know early on that this is something that you're evaluating, because, yeah, AI is quite new for some businesses. So maybe you do have a legal process involved or maybe you don't, but it's really just good to be aware of that. Same with infosec as well. It's good to be across whether your company has any requirements there too. I know Dan and I have had to fill out some security questionnaires, which it we're always happy to do, but that might actually be a requirement for your business. So it's good to know that early on so that, we or the vendor can do it as soon as possible for you as well. But once you've got an understanding as well of your process for purchasing something like AI, how do you actually roll it out? Well, the good news is, that most vendors should help you do that during the trial phase. So I know when we help customers or, businesses trial in an AI intercom, we generally get it up and running so you can actually test it live with your customers. So by that stage, once you've done the trial, you likely already proved that it can work and it's already set up and so it's ready to go. But now you've decided that you actually want to continue using it long term. And to be honest, most of the trials that Dan and I run with businesses evaluating AI, once it has been live with customers for, like, two to three weeks, you have a really good idea as to whether it's meeting your goals. So now it's just a matter of making it officially something part of your tech stack. So that's why it's really good to keep going back to, yeah, the preparation, the testing, the metrics, and the success, criteria that you have, to make sure that we can move that along for you as well. Perfect. So I think just Vex, an amazing summary there. But just to kind of wrap it all up into a final slide, I think what we wanted to kind of show was how to set up the best, put you in the best spot for, setting up your AI strategy within customer support. And so I think if you have the preparation done of understanding, what it is you're trying to solve, the impacts of solving it, and what you want the solution to look like, having your content ready for testing, and then rolling out a testing strategy and framework that really drop really pulls back to why you were doing it in the first place. It builds a really, really strong business case, for why AI will be successful and then how you're going to deliver that, to the business, from the get go. And then selecting it is all about, like, a partnership because AI, platforms are an ongoing, evolving beast. I think it's all moving so quickly. So you really want to, have a vendor that you enjoy working with, and they've been able to answer your questions and help you through the process as well. And I just want to kinda this is the first one of these webinars we've done, about kind of how to set up your strategy. And so we just also love to know, is there other, webinar topics you'd like us to do into the future? Maybe that's, once you've set up AI, how to manage it, over the over the next period or how to increase AI adoption internally. We'd just love to hear your thoughts on what some of those, might be as well. But I think that's, everything from us. Any questions, please reach out to us. I'm more than happy to have a call and chat through any of these components as well. Even if you're a customer, not a customer, just interested, please reach out. We're happy to chat through it all as well with you. Yeah. Anything else from you, Becky? Well, we actually have some questions in the q and a section. So I did try to answer some of them live, but I did wanna call out some that I think could be really useful for everyone to hear. So there was a question, basically asking whether you can teach Finn what to answer and what not. Like, basically, giving the example that there's a customer with a special setup and you've been instructed not to present certain features. So can you teach Finn what to say and what not to say? So the answer is yes. So one of the functions that I was mentioning before guidance is actually perfect for this. So the way that I like describing it to businesses is imagine you're you just hired a new teammate. This is basically giving them the instructions that they would use to interact with your customers. So we have you can help them contact, ask for context or clarification with customers. You can also teach it a specific communication style. So I know one of the other questions was asked, can Finn adjust its tone based on perception of a customer's sentiment? So definitely, yes. In guidance with the communication style, you could basically say, if customer is frustrated, use a really assuring, like, gentle tone, acknowledge that you hear their request, things like that. So definitely, yes. But, yeah, you could call out in guidance, like, don't mention x y zed. But, also, one thing I wanted to flag is that you have control over what content Finn uses. So if you have a help article about a feature that you don't want Finn to be discussing, you can just choose not to use that for Finn. So that's the option. You have we have a knowledge section in Intercom, and you can really easily select, like, not to be used by AI. So that's another way as well. We've designed it in a way where we want you to have, like, complete control over what, Bin can provide. So, yeah, having the content selected is the first way, and then leveraging something like guidance is another great way as well. Just looking at some of the other questions. So, oh, so this is a good one about, real time translation. So if you have the question is if we have our FAQ pages in multiple languages, will the chatbot draw answers from the language the customer is using in the chat process? So, yes. The answer is if it basically, real time translation, which for those who don't know is a feature with Bing where even if your help content, let's say, is all in English, if a customer asks a question in, say, Spanish, Bing can then recognize that and translate your English content to respond to the customer in Spanish. So it can work in that way. But if you have already your FAQ pages in multiple languages, it works the same way as well. The main difference there is that it can actually reference the article in the different language as well. So Jeff asked a good question about, do you have situations where fin AI trials have failed? And if so, what are the learnings from those cases? And so it's, like, a super interesting question because it all depends, I guess, on what do you want understanding what you wanted to achieve in the first place, and then having, I guess, the right, knowledge and preparation set up. So where I've seen them fail is there's a misalignment between what people think Fin will do, and then what its capabilities are. And so just the reason we kind of run through this today is to get everyone kind of on the same page, before we start the trial. And then the second component is making sure we have the knowledge, base set up that can resolve the amount of queries that need to happen. And so where I've kind of seen them, I guess, not work so well is if there's misalignment on what we're trying to achieve, and then maybe the knowledge isn't, isn't the right knowledge for the answers, the questions that need to be answered. And so there's kind of two components there and that's really kind of why we, did a deep dive on those two pieces today. Yeah. In terms of making sure that people actually see the the quality and it aligns to what you want. Yeah. The the good news is if you have done, like, the preparation that Dan and I outlined in the first section, I can't think of any examples where I've helped one of my customers do that, and then they haven't had a successful trial within. Because we as Dan said, we wanted to enable you all to be best set up for success, and we believe that that framework really will help there, as well as, yeah, things like content best practices as well. So, I hope that that's encouraging because it really like, this is what we've learned from working with many businesses as well. And for your second question, Jeff, like, what is the technical road map in AI? So, I mean, as we said, we can share this, with the team. But in terms of what it will like, in one year or two years, as we said, we are evolving Fin. We have, like, quarterly road maps. So this is something that we have a really big team. We've made, like, a multimillion dollar investment into Fin, and, evolving our AI. So this is something that we're gonna continue working on, forever, basically, which is exciting. But some of the really cool things that are coming, like, we are at the moment doing a testing of fin tasks. So we can already do fin actions, which is if you need to see information that's in one of your third party tools to make an answer really personalized. You can also if you need to update something as well, you can do that already with Fin. But we're sort of just, like, sophisticating that even more and taking it to the next level. But things like guidance as well are a really great example of how we're iterating it even further. And on that road map piece, the Fin actions is, it's like if, it triggers off a bunch of different processes and essentially is taking Fin into that agentic space where it can, act, as a agent may do, to, trigger off a bunch of different actions, not just a singular action to update an address, but it will go, trigger the full, credit card cancellation and refund policy for fraud. It's getting kind of more advanced in that one. We also had a really good presentation about a week and a half ago called Built for You, and the executive product team announced some really amazing new features, one of which was, Fin AI Vision. So you can now upload images to Fin. It can read them and then decipher what the question or error is, and then give a response based on the image that the customer has uploaded. The one that they used was a crushed Uber Eats delivery. And so off the back of seeing that it was a crushed delivery, it then took action, to to refund, the food order. So we can share the link to that video because there was a lot of really cool features and discussion about what's coming as well, but we can send you out the road map as well to chat through that as well. Definitely. Cool. Any final questions? I mean, I can see one about, like, packaging of Finn. So essentially we fin is priced on resolutions. So we really designed it in a way where we only want you to be charged if the customer has received a successful resolution. So something that you might have noticed about Intercom, we are really transparent with our pricing now. It's all on our website. So you can definitely go and have a look at that. But also, if you're an existing Intercom customer, reach out to your account manager or reach out in the support chat and we can definitely chat about it with you more. Let me see inside. So, like, some of the things that this question is asking about is you don't have many questions with straightforward answers. So that's actually a really good point. So and I feel like that's also a reason why some people are hesitant to adopt AI in customer service. But I would encourage you to definitely still test it out because as we were mentioning, like, AI is becoming more and more sophisticated and so is Fin. And so Fin actually can now answer complex queries. If you have the content for it, Fin will be able to draw it together and help answer the question. But as we were just discussing, if it's something that maybe requires checking something in another tool or taking an action, that is things that we can do now as well. So definitely reach out if you, want to discuss, like, how your queries are structured because, we've been able to help a lot of businesses who felt like they had really complex queries actually to resolve those within as well. And the resolutions are priced the same as well. The other one on the complicated queries question is because Finn has access to previously resolved, conversations or tickets, it can then use, conversate oh, so previously resolved conversation or tickets. It can see if those complicated questions have been answered before. It doesn't just have to be in your help center, content. So you can use a variety of different, sources to source that information. And so I've found working with, I guess, larger professional services companies that might have, more complicated queries, having access to those, previously answered customer conversations and tickets, will provide kind of, better and better resolution rates than expected on those phones. We've just only got twenty seconds left of the webinar. I don't know if it cuts us off, but thank you so much, everyone, for joining. Any questions we didn't get to, I will definitely get back to you afterwards. So thank you so much for joining. Continue to reach out. We'd love to chat with you. And, yeah, thank you so much. This was fun. Yeah. Thank you so much. We'll catch you soon. Bye bye. Bye. Bye.