# 144 How to Build a Website AI Agent That Books Demos While You Sleep (Free n8n Blueprint)

# 144 How to Build a Website AI Agent That Books Demos While You Sleep (Free n8n Blueprint) Speaker 2: What if you had an AI employee that could answer customer questions, book sales demos, and update your CRM all while you're asleep? And when you wake up, you've got three new qualified leads already scheduled on your calendar with a full conversation transcript waiting for you. Now, that's exactly what my guest Ashley Gross has built with an AI chat bot called Bella, and she's going to show us how to build one ourselves in this episode. Now, Ashley's been using AI since 2020, Way before things like ChatGPT were mainstream, and she's tied her AI implementation efforts to over $25 million in converted pipeline at the enterprise level. So in this video, we're going to cover how to build a 24-7 AI sales agent for your website, how to train it on your SOP so it sounds like your business, and the exact n8n workflow that you can copy and paste to get started today. So we're going to go ahead and dive right into this week's episode. All right, Ashley, why don't you tell us what we're going to learn today? Speaker 1: Hey, Corey. Today we are going to learn about how to build an AI agent that can hang out on your website so that while you're sleeping, anybody that comes to your site, asks questions, wants to book a demo, can still have their needs met. And when you wake up, that lead's going to be in your CRM, all of their questions answered and ready to go for that next day demo. Speaker 2: Awesome. Well, that's going to be really cool because that's something that can apply to pretty much every small business out there. So if you're a small business owner and you're watching this, you're pretty much going to get the blueprint for how to implement this in your business. So before Ashley shares her screen and starts diving in, Ashley, why don't you just tell us about your background and why it is that people should listen to you? Speaker 1: Sure. I started using AI back in 2020. So before tools like ChatGPT were mainstream and it was truly the Wild West, I was using AI tools to reclaim my time. I was spending almost 80 hours per week working. And I was able to use AI to automate a lot of that busy work down to 15 hours a week, which means I was still doing the equivalent of 80 hours, but in 15, because a lot of it was frequent work that was really dragging out and not strategic and not a good use of my time. So I started automating every process that I had to do more than twice a day. I've been able to tie AI implementation at an enterprise level to revenue. Specifically was able to prove that $25 million converted in pipeline was applicable and possible because we added AI implementation into product customer success, marketing and sales. Speaker 2: So you've been doing this before. It was cool, which I think is really cool because a lot of people, I mean, like myself, I only started getting into this world once ChatGPT really came on the scene. So I'm excited to learn from you today. Cause I think you've got more knowledge than most people. So with that said, do you want to cut you, you gave us kind of the brief intro of what we're going to be building today, but why don't you give us more of a, like a fleshed out version of like, what are we building and what does it exactly do? Speaker 1: Sure. So what we're building is a chat bot that can sit on your website. So think Freshdesk, think intercom. Anytime you have a little thinking bubble that you have a question about and you click on it and you're met with a very clinical tone saying, what's your question? Here are categories I can help you with. And it just basically reroutes your questions over and over again. I didn't want that experience for my customers, but I also needed to sleep. And I found that a lot of the time I was getting questions was the middle of the night. So I needed to figure something out so that I wasn't using my sales reps human time in the middle of the night to potentially stay up and ask questions or potentially not have any questions at all. So I build an agent that responds every time the button is clicked and someone says they want to talk to an agent. It will actually come on and answer questions for them. And her name is Bella. Do you want to see how she works? Speaker 2: Yes, absolutely. I want to see how you built her too, kind of step by step. Speaker 1: Sure. So this is step by step how I built her. So I host my website on Framer. So there is a webhook. And every time somebody clicks that button, it sends a webhook, which is the trigger for this workflow. And then it takes all of the conversational pieces, Basically, every part of the transcript, the name, email, phone number, company name, any questions and answers, so that I can have a full transparent look at exactly what happened. I'm big on trust and authority, so I want to see everything that went in that people were asking Bella, but also everything that she was giving them back, because that's equally as important. And then once she had an exchange with them, her goal was simple. It was always to answer their questions and then get them booked on our demo calendar, which she had our availability for. And once she did that, it was really just a simple note of checking if we already had the email. So like, is this somebody that has form filled on a lead gen magnet for spoken to us? If we didn't have their email, then she actually went into active campaign, and she created it, or created an updated the contact depending on what information we had. Send us an email so that when we woke up, we had the demo information, all the questions and answers, and then the little blurb about her. And this is what she looks like. Speaker 2: Nice. Speaker 1: Anim AI. There's a bunch of different avatars you can use, but I chose to use Anim and I won't click start because she can talk and we can't necessarily hear her, but she can hear me if I'm talking. So she just gets really confused basically. Speaker 2: Yeah. Speaker 1: But she was trained off of a secure Google Drive. So every single SOP that exists in the company that is safe and has a certain level of visibility and tiered risk, she has in her knowledge. And so the idea is every time she answers a question and she doesn't know the answer and I have to help her with it, it gets logged into her memory so she'll never ask me again. So she's truly agentic in that sense. Speaker 2: So it's so cool how you have it set up that way. And so just to kind of summarize what she does. So essentially it's a chat bot that's trained on your business's SOPs and the way that you guys do things in your business. And then it sits on your website. So if anybody has questions at any time of day, they can chat with this, essentially ChatGPT for your business. And then from there, the goal for the bot is to get that person on a demo call, because I assume you guys probably know, oh, for X number of people we get on a demo call, we close Y. So it just becomes a numbers game at that point. So like right then and there, by being able to book those people at two in the morning, if that's when they're chatting with your model, then, you know, that's a lead that probably wouldn't have booked a demo elsewhere. So I can see this directly translating to revenue. Speaker 1: Exactly. And I like using Anem because you can choose between the different large language models. So it's really frustrating whenever you become married to one and then there's a new feature update on another one and it blows it out of the park. So I like the ability to switch them out depending on what their latest update is and how well they're performing. But also I really enjoy having one avatar because I have two businesses and they're different and she's trained on both of them. So I don't just have her on one landing page. I have her across three so that if somebody is asking a question and they're not a good fit for one business, but they are for another, She can handle that back and forth transition. Whereas I feel like if you had actual humans working that middle of the night shift, you would have to segment them into different areas of the business and in separate businesses. And she doesn't work like that. So she's truly like a big picture thinker of where are they at in our life cycle, where they belong. And also. Simultaneously, she gives us a repository of all the questions so that when we create new content, we can create content off questions that people are actually asking us. Speaker 2: That's super smart. So yeah, so that's how you're getting your content ideas too, is just, hey, what are people asking our AI agent that sits on our website? So I absolutely love that. Like I said, I think that's directly applicable to just about any business out there that gets inbound leads, which I'd say is pretty much any business. Speaker 1: If they're doing it right. Yes, absolutely. Speaker 2: Yeah. So do you want to kind of dive into the n8n flow itself and kind of talk through each module and what job it plays and potentially how someone could swipe this and really set it up for themselves? Speaker 1: Yes. I added the setup. So it's step-by-step what you have to switch out. But it starts just with that webhook. So all the webhook does is actually use the trigger of the button being clicked so that everybody when they attend that specific section of the website and they click that button, they understand that they're about to voluntarily get on a call with an avatar and ask questions. They could chat, they're deciding to call. So that call, that click on the button is actually what signals the trigger to start this workflow. And the conversation could be 30 seconds, it could be five minutes, it could be 10 minutes. So we didn't want to do a time-based interactive node because we didn't know how long those segments were going to go. And we actually found out that when you tell people that they're about to talk to an avatar and it's the middle of the night, they're pretty understanding that you're being transparent about it because everyone's got to sleep. So the conversations were actually looking more of an average of 10 to 11 minutes, like pretty long conversations. Speaker 2: Wow, that long? I'm actually shocked by that. I figured it would be like, oh, most people figure out they're talking to an AI agent and then they hang up after like 40 seconds. But like 10 minutes is a long time to be talking to an agent. Speaker 1: Exactly. So that's why we only have one node for the processing conversation. And you can see that in this conversation, these are all the bits and pieces of the lines of code that she's able to get. So visitor name, the email, phone, the company, the actual session ID that took place. We're going to talk about what page URL the user was on when they asked to talk to Bella. If this person was referred, what the full transcript is, how many characters were in the message count, duration and minutes, start time, end time, summary. I'm the lead source. So we can see truly like where all these people are coming from. Yeah, we have their digital footprint across the whole entire site. And the best part of this is they're willingly giving this information to us because it makes their experience a lot better. As a bootstrap founder twice, I can't tell you enough how expensive tools can get that promise you that. The fact that you can find a way to keep sales human to human, but also find mindful places for automation when it makes sense and also get this level of data. I think that that's like the dream for any small business owner, in my opinion. Speaker 2: Yeah. I mean, it's truly a win-win. Like you said, the customer's getting a better experience and you as a business owner are delivering And we're here to help you build a better product with a more efficient cost structure. So yeah, absolutely. A win-win. Speaker 1: And then the next node is just checking whether or not that person gave an email. And if we have that email in ActiveCampaign. So ActiveCampaign is our CRM. So whenever your audience gets this little blueprint, all they would need to do is switch these two nodes out with their CRM of choice. It could be HubSpot, Salesforce, Marketing Cloud, whatever their nodes are. And then from there, it merges the paths. And then after that, it just sends a message and it just says, hey, there, you have a new lead from last night's discussion. And it actually grabs the summary of that processing conversation data node. And gives us the summary and the lead source and the message. So when we wake up, we know exactly what happens, but also they're already booked on our calendar. And the best part about this is let's just say, hypothetically, they book a demo the night before and it's the middle of the night. And then two days later, they want to reschedule it. They can actually reschedule with Bella. They don't need to email us or call us. She's got our schedule. So we don't have to give out our Calendly links, which I find really helpful because sometimes it just gets abused naturally when people are super busy and we don't like to give them out for that reason. So we haven't had to, but Bella has them. So she knows they're next available and she's able to have that kind of an interaction with them and still maintain our data privacy. Speaker 2: And what I love about this flow is like you said earlier, you can substitute out the different And today we're going to talk a little bit about how you can configure these pieces of this automation for whatever tools you're using. So like if you're using GoHighLevel, for example, then you just switch that out where you have your active campaign nodes, you switch that out with GoHighLevel and then it's kind of plug and play with how you do things. And then if you don't want to receive an email with your daily summary of the people that it talks to, then maybe you want to receive a WhatsApp message or a Slack message or even a text message. It's really adaptive to you as far as how you do things, which is what I love about these kind of like node based. And we're here to talk about automation and AI agent tools. Speaker 1: Yeah, you hit on a really important part. So it's interesting. I came from the enterprise space. So it was a little bit strange for me leaving the enterprise space and having two of my own startups because I was conditioned already to look at having a lot of tools. And when it comes down to it, if you have really good communication, really clean data, and that's all in one place, you don't need Slack. At least I found you don't need a lot of those tools. So my company operates off of Notion. That's where we put all of our information so that it's visible in one place. So even though we're using ActiveCampaign as our CRM and we use Clay and a couple other tools for our processes, there's always one viewpoint and like one master dashboard in Notion for everyone to see. So it's been a nice little pivot and switch because Having all these tools and getting constantly pinged with all these updates was not productive for me. It was actually more distracting. Speaker 2: Do you guys use a messaging tool of any kind, like a Slack or a WhatsApp or you're saying it's all in Notion? Speaker 1: It's all in Notion. Everything has a database. It's got back pages. There's SOPs, SLAs, contracts, everything from even, we have our database for who we are sending emails out to and then who we're interacting with on social media. That's all in one database. That's kind of the beauty of having a small business and being able to automate this piece by piece and then scale it out as operations and processes start to make sense. Because if you do it this way, even though it's a crawl, walk, run method, you're actually kind of interrogating every process to see if it even makes sense to begin with and if it's effective. And then once you realize and establish clear benchmarks and ROI, you can then automate it and it stays and it doesn't break later on because you did that work up front. Speaker 2: Yeah, you're basically forcing yourself to optimize that process, which goes, I mean, hand in hand with the framework that I will talk about from time to time, our AOA framework, audit, then optimize, then automate, right? You can't go out of order. That's something that we've learned the hard way several times is a client will come to us and say, well, like, hey, can you automate X, Y, Z task? It's currently taking me five hours per week. Made the mistake of just saying, sure, we can do that. And then we go into actually automate the process and we realize the process itself is broken. So then you've got to go back and do that process optimization work in the first place and then go back and automate. So it ends up taking twice as long and being twice as expensive. So I like how you kind of framed it that way. Speaker 1: Thank you. You have to be married to your tech stack. That's the rule I like to say. Speaker 2: Yeah, absolutely. Now, so and I understand you've got something that you're giving away as part of this kind of breakdown here. You want to go into that briefly and then I've got a couple other questions for you on how people can kind of go and set this up for themselves, like how to plug in their data. Speaker 1: Sure. So I'm giving my blueprint away. So you'll actually be able to just copy and paste this and bring it into your own n8n environment. And like I said, just switch out the node for webhook and then the two CRMs. If you don't use Gmail, you can switch out for anything that you'd like. But I'm giving this away. I think that this is a really good use case. And I'm really, really intentional about sharing and not gatekeeping the small business owner success stories because it takes a village. And so if you can find a way to authentically and intentionally use technology, I'm all for it. So everyone that is tuning into this, We'll get this blueprint. It's got this nice little setup with five steps. Super simple. And you'll be able to take this and go wherever you want with this. And if you like Anim, which I highly recommend using, you'll be able to choose between Bella or about a dozen other avatars. And you can choose which large language model you use, which avatar, obviously, voice, And then how you actually train the avatar. So you can configure all of this. It really is personalized to the user's liking. But I just gave her my knowledge base and my Calendly and I gave her very specific prompts with roles and speaking styles and useful context. And she was amazing. So it's truly a plug and play solution. Speaker 2: Well, that's incredibly generous too to give away for free. So for those of you listening, if you're listening on the podcast, we'll have a link to get that for free in the show notes. And if you're listening on YouTube, we'll have that in the description. And like she said, that'll pretty much be a, you know, download and then upload straight into your n8n version of this automation. So that'll be super useful. Now, Ashley, so if somebody is watching this and they're like, okay, well, I want to kind of try to tinker with it and Hook up some of my own data sources to it and really personalize this and make it my own. Is it as simple as just arming the agent with your SOPs or kind of like you said that you did, hooking it up to your Google Drive so it has access to all your docs and sheets and whatnot? I mean, is it really that simple or is there something more that people need to do? Speaker 1: No, it's really that simple. I would warn and caution rather that the Google Drive originally that you give should only have certain pieces of documentation in it just so you can see What that breadth of knowledge looks like once you give the avatar a little bit of it and then as trust in between the humans and AI develops and as your comfort level goes up, then give it more information. This is also just a really good skill set for how to pressure test agents in general. If you give it access to everything, then you don't really know what to expect or how to mitigate issues. But if you give it layer data and add slowly, it's a lot easier to fix something and correct something proactively. Speaker 2: Well, I think that's a great point too, is a lot of people just, they hear that they need to build a knowledge base or they go to build an agent and they just kind of data dump everything that they have on them. They give them all their SOPs and access to all their tools and their whole Google Drive and the whole nine yards. And then they realize the output they're getting is not that good because they've got context bloat. They've basically given it information overload. So I think that's a good point is like, You know, really figure out what is the purpose of this agent or the purpose of this tool and then giving it only the data, only the tools that it needs to do that job and do that job really well. Speaker 1: Exactly. And then making sure, obviously, that you're reading the documentation you give it beforehand so that everything is actually accurate and not outdated. That can get you if you're not careful. Speaker 2: Absolutely. No, that's a great point. Well, Ashley, is there anything else you want to go through as far as this particular workflow is concerned? And if not, I've kind of got one question for you to transition us out. Speaker 1: I'm happy to answer that question. It really is a simple plug and play solution. So I think I hit on Bella pretty well and I can't really take any credit for her because I didn't create the company and she's flawless with her execution. So I don't know what else I could say. Speaker 2: I mean, if anything, half of what I'm building too is I just see, I'm like, oh, this guy built something cool. Like I'm just going to go Basically, practice building that for myself. Speaker 1: Yes. Speaker 2: I feel like that's the best way to do it. Speaker 1: Absolutely. You're not going to waste your time that way. Someone already started. Speaker 2: Right. Absolutely. My last question for you. Ashley, let's say I'm a small business owner. I'm not technical. I'm pretty heads down in the weeds of my business every day, but I want to start leveraging AI. I want to start automating some stuff in my business. What's the one thing that you would tell me to do to start? Speaker 1: Focus on revenue. There's a million different use cases for AI agents and automation in a business, but you have to be focused on revenue, bringing more of it in, preserving it. So I would say think about the use cases that generate and you know, generate more revenue and then go on YouTube. It is a knowledge source. I still think it's like the best place to go anytime you want to learn anything about these agents. And start looking at videos and tutorials of people that are building agents using your tools and your tech stack to bring in more revenue for the business. I would say don't try to change too many things at once. Focus on money and then focus on agents that connect to the tools you already have. Don't try to switch them out. Speaker 2: That's very good advice. And I think a good example of that is like a speed to lead agent, which is kind of a variation on what you showed us, right? It's just. I guess for when people do get those inbound leads, it's having an agent that, like yours, trained on the SOPs, trained on sales, and kind of goes back and forth with the goal of booking them on a call or a demo, just like yours. So I think that's the best place to start. We built those for clients. They're probably the one thing that I can think of that most directly translates to revenue is how quickly do you respond to leads. Speaker 1: Yeah, you can start looking at it through funnels. You know, if you start by having something that can interact and do better than a chat bot in the middle of the night to help kind of guide the way for your prospects, then the next layer on top of that can be like enrichment and you can start to play around with tools like clay and actually like enriching those, those folks so that the bonus part of that is you wake up, the demo's already booked, but also you know everything about them because of your enrichment tools and you don't have to block your calendar. You know, you can kind of keep it. Loose in terms of like if a demo pops up, you're not going to be sweating because it's a high priority and you have to clear everything to go like deep dive and research about them. So it's like this nice little layering component where you feel comfortable, you get something right, you build structure and a foundation and then you go and build on top of that. Speaker 2: 100%. Well, Ashley, very good feedback, very useful build that you showed us today too. And so again, for everybody watching or listening, you guys can get a downloadable copy of that exact automation that you can just upload to n8n yourself and use right away. We're going to put that in the description on YouTube or in the show notes on Apple. So Ashley, before we go, where can people follow you or find out more about you? Speaker 1: LinkedIn is my favorite social media channel and I am constantly on there. Speaker 2: Awesome. So just Ashley Gross on LinkedIn. She is a great follow there. So Ashley, thank you so much for your time and for everybody watching. We'll be back next week.