How Smart Agencies Use AI to Save 40+ Hours a Week (Real Systems & GPT Examples)

Hello and welcome to today's session. We're going to be talking all about how to leverage AI in your digital marketing agency. And as you guys know, here at 7 Figure Agency, we're all about helping digital marketing agencies grow, scale, so they can make more money, have more freedom, have a bigger impact. And I really believe that the agency game is changing, right? And if you're not evolving with where things are headed, you're going to wind up being left behind. And we've had a lot of experts come on and talk about AI. We had Jeff Woods who came on and spoke about AI-driven leadership and how to think at a higher level within your agency. Recently we had Andrew come on from the AI Marketeer and talk about how to really leverage AI in a very tactical sense. And today I'm really excited to have Kushboo with us from E2M. E2M is one of the, one of the top white label web development, SEO, white label agency service providers in the marketplace. And they've sponsored our last couple of intensives and I've got to know Kushboo and the team there pretty well. And they're doing some very advanced things with AI. So I asked Kushboo if she could come on and do a workshop with us. And previously she's only done this like in private environments and share with you guys some of those latest ideas and insights. And so I'm going to spotlight Kushboo here. Kushboo, thanks for being here. And I'm going to hand you the floor and I'll be interjecting as we have questions and keep things on track.

Sounds like a plan. So super excited to share all the learnings that I have so far. We are going to start with the use case discovery because before we start hands-on building things, I really want to set things right in terms of foundations, as in what are the different use cases where we could really use AI in terms of the operations? And then once we have these use case discoveries done, we'll dive deeper into how do we build that, like on ChatGPT, on cloud, and in it in automation. So that's all we are going to cover in this workshop. Now, in terms of use case discovery, from starting up until now, I feel there have been a lot many areas where we feel that, okay, I don't think so AI could be used in these areas or not. But then these are all the areas that we have figured out works really well for us. It's all functional. It might not be accurate, but really saves a lot of time and builds a lot of efficiency. So we could have agents in Sales Insights to just give like a sales summary, have an agent, and it gives you sales insight based on the summary or the numbers that you share. We will be seeing that in working. Customer intelligence agent, like feed all your customer information to an agent and then ask any email or code or anything about that client, and it just gives you an answer from a generative AI perspective and not just like a generic answer from a knowledge base. So that's one area that we could build an agent for, like a dynamic SOP AI agent. A total game changer. If you have processes, you have SOPs, this is the time we need to create agents around it, which is where you don't give SOP or a process documentation to any of your employees or to any of your clients. Instead, you're giving an agent so they could ask questions rather than just going through all these lengthy pages that we create processes around. A financial analytics agent, it gives you insights based on what client is profitable, where do you need to spend time into analyzing the customer experience, where the pattern is breaking. So you can just give in all the details, like the whole, I would say, ledger of the month's financial, and then it gives you the whole segmentation of where the profit and loss analysis are. AI-powered lead generation. So we are going to build that today in the workshop, how to generate leads using AI. Automated outreach AI agent, lead qualification agent. So there are a lot of areas that we could already use AI for in terms of our agency operations. These are just a few, and then we will be seeing all these use cases today. But just a quick exercise for all of you before we dive deeper into how to build things. Before we even go into how to build things, we need to know what to build for our business, what to build which would help us free up time, or what to build which is going to help us get more revenue. Write 3 tasks where you feel you're repeatedly spending time. You have process around it, you have an SOP around it, but you don't have an agent or an automation around it. So those are the ones that we should be working on first. And then once we go through how we build those things, right after the session, you can build those agents and automations for you. I'll give you a list, like an example as well, like a proposal creation. We all have a template, but the requirements are very custom. So we go on creating custom proposals every time we have a new requirement. So how about we have a template wherein it's fed to an agent, so now all we have to do is just give the information and it just auto-populates the proposal based on the information that we give. So that would be something that we spend a lot of time every week, but that could be automated. A sales outreach. There is a sales team who does an outreach every week, but what if we could automate like a certain aspect of the outreach email? It's gonna save a lot of time. So these are just a few examples, but now every agency would have their own unique ways of dealing with things and doing things. One last exercise before we dive into how we build those things is now we already have a list of things that's going to help us free up our time, but now we want to have a list of things that help— that's going to help us boost our revenue. So anything that you feel you're working on right now, which could be automated, it would directly or indirectly help you increasing the revenue of your client. It could be client satisfaction, which is going to help you with the renewal and then LTV of that client, or it could be new value propositions that you could be giving to the client. So just giving you an example, like sales outreach optimization. If you outreach to more people using automation, it's gonna help you boost your revenue indirectly in certain months. Upselling and cross-selling, like an agent identifying a cross-selling and upselling opportunity, and then you use AI to build that momentum within your existing clients. Customer retention strategies, building an agent who defines and predicts the churn and then gives you some strategies on which you can help and make that client stay more. Now we'll be moving towards what could you use to build those agents. Today, what we will be seeing is like ChatGPT agents, Claude projects, and automations, but there are different tools on top of that could be used to build automations around. One is Make.com. The other is Landflow. All right. So we go on the GPTs here. We just go and click in there, just click on create. You see a window like this, wherein you just write the name of the GPT that you want to create, the description of what this agent is going to work on, like what exactly this agent means. Then we write the instructions, as in what this agent has to do. Now, I would say a really good recommendation on how to write your instructions, just use AI for writing an agent instruction. So just tell an AI that, hey, I want to build a world-class agent for possibly like a dynamic SOP creation, think like you are the top-level SOP creator. Write the instructions in a way where the agent will be behaving like, and then just give the pointers of how you want that agent to behave. So just gonna write you the instructions. So you're using AI to help you build AI agents. So here we are not doing anything from scratch. All we have is the raw idea as to what we wanna create for, and then we'll be using AI all through this journey to create the agent or the automation idea that we had. So we just write it in the instructions here, like, what do you want this agent to do? Conversation starters is what kind of questions you want the user to ask. So it just gives them an idea that what kind of questions the agent could answer. So that goes into conversation starters. I'll just pull up something that we have built, which I'll also show you exactly how that works. SOP. Okay, this is like an SOP agent.

I'll get my SOP agent working here. Give me a moment. Edit GPT.

Okay, so you see it's like an E2M account manager guide, which is like an SOP agent. We give this agent to every new client account manager that joins the team. Now, before we had this agent, we had 50 pages of documentation in terms of SOP for every client account manager that they had to go through in like the first couple of days, where it had all the data about the renewals and how to handle escalations, what kind of clients, what kind of requests, the different services. Like it had everything that a client account manager should know. Now, practically, it's next to impossible for any new person to go in and check everything that's there in that documentation in the first 3 days. So basically what we did is, and this is gonna give you an idea also how that looked like. This is how the whole SOP agent, SOP documentation looked like. It had like onboarding details, which is like quite lengthy, and then new client onboarding, the whole processes, renewal process, escalation guide, emails, initiatives, like everything. So it was all in Coda before we had this agent. Now what we did is we just downloaded all these Coda documents that we had in terms of SOPs. Provided that in the knowledge base here. So we just attached all those SOP documents we had in the knowledge base. And what knowledge base is basically, this agent is going to use the files that we have attached in the knowledge base to give any answer. So whenever there is a question that is asked to this agent, the agent is going to answer it from the knowledge base, like from the SOP files that we have attached to it. It's not going to answer it from the internet. It's not going to answer it on its own. It's going to answer it from the knowledge base that we have attached. And now it could get further as well, wherein you can just have an API of Coda connect to the ChatGPT agent. And now you also don't have to, I would say, upload all these knowledge base documents. It just directly connects with Coda. And then any change that you make on any of the SOPs, it just directly reflects in the agent. But here, for better understanding, I've just downloaded those files and appended in the knowledge base. Now, if you see the name, the description is like it's guiding account managers. Now, if we'll go through the instructions, like you are an assistant dedicated to providing actionable guidance specifically for account management team. What it has to do. So there are 4 documents that it has to go through and then provide guidance, assisting with the client onboarding, guiding account manager. You are an assistant centralizing, streamlining account management processes. So just like I got these instructions from ChatGPT itself, like I explained how I want this agent to work and it just gave me the instructions that I could just go and put in here. So this is what the instruction structure would look like. The conversation starter is if anybody's using this, what could they ask? What are my daily tasks? What should I handle if I'm a new client account manager that has onboarded recently? What does the renewal process look like? So these are like the conversation starters. You can upload more files from here. Now this is something we just, I think ChatGPT launched last week. Earlier we didn't get to choose which model we want to use for an agent, but now you can choose which model you want to use for an agent. So if you say no model, no recommended model, it just takes the best guess and use the one that it thinks should be appropriate for the kind of agent you want to create. But if it is something I would say more logic-based, or if it's more, there's coding in it or some math calculation in it, I would recommend going for ChatGPT 4.5 or O4 mini or O4 mini high because they have logical reasoning inbuilt on those models. So in that case, the accuracy is going to be higher than any other models that you choose. But this is more content-driven model. So we are gonna take, we are gonna allow ChatGPT to take the best guess. And just not choose the model. Web search, you could tick, you could have it if you want the agent to go outside of that knowledge base and find answers to the question. But if I wanted limited to the knowledge base, I'll just take that off. So I don't want the agent to go and do a web search and provide answers. I want very specific answers based on the SOP. So I'll just check that off. Yeah, and like the agent is ready. We'll just go ahead and update it and then we'll test it.

All right. Let's just write, hey, it's my first day today. What do I need to do?

Okay. Let's just ask like a pretty dumb question. So it just gives me like a step-by-step guide on what do I need to do, like login credentials, communication tools, email signature, Calendly. Like this is all from the SOP. I didn't have to go through any documentation to get these answers and get me started with what do I need to do. Okay, I'm good with day 1. I have an escalation. Can you help me? Let's see. Okay. Anything I need to know about escalations? Any Tips. Okay, so we do have escalation tips in the SOP that we have. So probably it's just giving us tips from that. And if we are asking straight out a question, it says that, okay, you can reach out to Khushboo, DM her or Slack her. It also gives my email address, which is also in the part of the SOP. So this just makes super simplified even for your client. Or for your team members to go through things. Now, this same use case we have replicated in a way where we have a chat model. So if there is a client asking for a certain service, let's say GoHighLevel, we have a chatbot for GoHighLevel. So once we are done with that call, we'll give that agent's link, the GoHighLevel agent link to the client. So in case they have a question like, do you do lead gen in GoHighLevel? They could just go and ask to the agent and the agent replies. So So they don't have to wait for email responses. They don't have to wait on another call or maybe ask for a time to chat again. They could just go to an agent and ask those questions like, what do we do? What do we don't do? So everything could be handled by agent. Now, if there is anything that the client is asking which is outside of written on the web page, there's an email that is triggered to me saying that the client asked this question, which is out of my knowledge base. Can you connect with the client? So in that case, even I know what are the questions that is being asked to the agents. and if I feel it's a really potential question or this is really something that I want to emphasize on, I can just email back to the client. So this is just one use case. If you have any processes, any SOPs, any documentation that you want to have an agent for, there will be a lot of questions and answers that it could be answering by itself from the knowledge base that you provide. And then after that, if you want to route it to your email or Slack, if any other questions were asked, you could do that with N8N Workflow. So this is just giving you an overview of how these agents are built, what are the different use cases on how would you use these agents for. So this was one use case, giving you more tactical use case, like a report generation use case, right? So we do have monthly utilization chart insert. So I'm not sure if everybody follows the OKR approach, but we are quite big on OKR. So what we did is we just created a PDF. BF version of Measure What Matters, provided that book to our knowledge base, to an agent. Every division can go in and say that, hey, I have a sales division. This is what my functional goals look like for this quarter. And then based on that book, based on the ideology of that whole book, it just helps us create more structured OKRs. So now we are not spending time on how to structure our OKRs. Based on the book, based on the philosophy written in the book, it just helps us structure those OKRs in a way it should be. The standardized or the industry accepted way it has. We'll go to the client monthly reports. This agent itself is saving more than 600 hours of our team every month right now. So basically, we work with more than 350 clients. We send them a monthly report at the end of the month based on the hours that they have used with us. Now we have built this agent which creates at least the foundation of the report that we send. I would say on an average, the client account management team spent about 3 to 4 hours analyzing all the data, going through the sheets, going through all the projects, and then what are the different tasks we did. And then based on that, we used to create reports. But now we just provide a spreadsheet to the agent, and it just creates the base foundation of the report that the client account managers could build on. So what we are going to do is I'm going to show you how this is built. The description is you are an agent that creates detailed monthly reports, instructions. Like the intent is you have to assist account managers to create a comprehensive monthly report, analyze the data based on tab selected. So we have tabs for every different months. There's a whole spreadsheet, but then it has tab for every different months. So the agent is going to prompt us which tab we want to select in terms of creating the reports. It's going to ask us the tab. Then it's like hours per— hours usage per project. We have given the formula for it. Hours summary. How do you define the summary of the hours? The average hours used, trend analysis, hours usage trend per week. And then the final step is, okay, what do you need to do in terms of providing the graphs, the bar chart, the line chart, and all those things? And then prompt status. So prompt status are here as well. What do I need to do? So I just need to provide like a CSV sheet that could be uploaded to this agent, and then it just creates the report for us. So we'll try in here. Hey, can you help me create a report for fourth line?

And I'll just attach it here. I'll go to desktop, demo files, fourth line, and we'll just add it here.

Alright. It's asking for the tab.

Let's go for Zoom.

And I'm also going to show you how that file looks like so you know what it's summarizing.

We found some problem. Yes. So if we go for— if we go for no.

So if we are going for June, this is how like a spreadsheet looks like. So it's just has the project names, the hours tracked and everything. Yeah. So this is how the data looks like. It's quite unstructured, just has a daily log of hours and where we spend those, the projects and all those things. Now, if there's a person doing it, it takes a lot of time for them to summarize all this data, but now we have this agent doing it. So it looks all geeky when you see all this coding going on and you see all these things happening on itself, but we haven't done anything. All we do is we just write prompts and all these things happens automatically. It seems task name doesn't exist.

Thank you for your patience. Column name is now corrected.

And I guess one thing that we need to know is the agents aren't going to be perfect. So if we feel that it's not doing the job correctly, it's not meant to do it correctly, or I would say with 100% accuracy as of now. But even if it's doing 60% or 70% of the job that we do, it's helping us saving a lot of time. And then there's always fine-tuning involved. So we can keep on fine-tuning these agents to perform better and better. But having something is good than rather than having like nothing.

Okay.

There's a question here.

How about the log data? How do you feed it? That's what Steph was asking.

So there is a context window challenge that is there in the agents right now, wherein we cannot feed in a lot of data, but I guess we can do a larger context window on the N8N platforms. Like we use N8N a lot, so I know a lot more about N8N. So we can create agents on N8N, which takes more, I would say, data as compared to what Claude or OpenAI could take. And if we still hit, I would say, the context window limit, we can use an AI model like an OpenAI node on N8N to summarize all that data. And then we can provide that summarized data as an input to any of the agents that we create.

Got it.

Okay, so we have it.

I will download it as a doc. Okay.

So it just creates like a downloadable file and then we just give this file. So it pushes out to our Slack group as well. So now every account manager can take this base report format from their Slack group and then build on those things. So it's gonna regenerate. While it regenerates, I'm gonna just give you an idea of what it looked like. I just did one.

And then cheat sheet. I don't know where it went. Just give me a second. It's gonna help us do it. Okay, it created the charts.

I recently did it for one of the clients.

I should have it. Give me a moment. And so I just did it for NNM.

Yeah, I just did this one. So this is how the report looks like. I'm not sure if you are seeing my screen.

Yeah, we're seeing the Word doc.

Yeah. Yes. So this is, I just replicated it and saw if it's working fine. So I did it for one of the clients for like June 2024, but this is the kind of reports it generates at the end of the whole process it does. So now this is not a perfect report that we could send it to the client, but it's at least a good starting point for account managers where they have all the information. Now all they do is they just go and fill in the information on top of this Word document that is generated by the agents. So it saves a lot of time for account managers to go through all this 100, 200 fields of different projects and hours and all those things. It just, it's a game changer when it comes to creating reports. We have a similar one for like SEO reports, audit reports. All those things as well, where we have connected the third-party APIs with the agents. But this is a base agent wherein you can plug in your processes and then build the reports with these agents.

All the concepts.

Yeah, so this is about ChatGPT, how we create agents, what are the different use cases, and then how we build agents on ChatGPT before we move forward with Claude and Enniton automations. Do we have any questions on ChatGPT agents or anything in particular about ChatGPT?

There's a question here from Linda, and Linda's asking, so I, so if I have raw data month by month on Google Spreadsheet, I can upload it and the agent will analyze it and create this style report with proper instructions.

Yes. Yes, that's right. And you can also connect your Google Doc here. So it does connect with Google Drive. So you can have your Google Doc in the drive, like the sheet. And then you can connect that sheet directly to the agent. So you don't even have to upload anything. You can just directly connect to Drive.

Super good, Linda. Hopefully that answered the question. Jeremy said, would it be better to use custom GPTs so that you can do deep research with it too? What's the case for the agent versus a custom GPT?

So this is a custom GPT. So custom GPT is an agent. So there are two things that you could build on ChatGPT. One is agent, which is the custom GPT. And the other is the project. So you can build ChatGPT projects. So now we have projects for almost all the clients. So our projects, different from custom GPT, is custom GPT is derived to do a very specific task. That's what custom GPT does, which is agents as well. Now project is something, let's say you have a client, you have all the branding guidelines and the quotes they have approved, the sales summary, the onboarding call. Like all the details about the client, you just feed to the project. Now this acts as a database for that client. And then any, anytime you have a new request from that client, or you want to check the email, if you want to send to the client, you just paste in here. So now what it's going to do is it's going to take context from everything that you have gathered about that client and going to provide you more accurate and precise information on how you should be reaching out or sending that email or getting that approval about that client. So this is more, I would say, generic in terms of accumulating a database of the information about a specific product or a client or a project, and then giving answers, whereas custom GPT is more on doing specific tasks where the output is very well defined in the prompt itself.

Super good. There was two questions here I think would be relevant for most of the people on here. Kyle and Valentin were asking, if you've got all of your project management inside ClickUp and you've got like, they're tracking hours and everything there, it'd be better to pull that data directly from ClickUp and/or maybe use ClickUp's AI function that they've recently rolled out. Any thoughts on that?

Yeah, so if you already have it on ClickUp, ClickUp does provide APIs. So in that case, we will not be using agents in the OpenAI interface. We will be using agents on platforms like N8N or Make, where you can integrate agents with your existing ClickUp and then define what kind of output you want. So in that case, it's not just going to be agent, it's going to be your third-party application integrated with an agent, with a generative AI, and then output is showing in Slack or email or probably your ClickUp itself. So in that case, it's like a whole automation and not just an agent. So we will be building that outside of ChatGPT or Claude, but yeah, that's something that's possible.

Perfect. Let's see here. They're saying when you summarize the data, did you notice an impact of losing important data so the analyst— so the analysis won't be impacted significantly?

Yeah. So in that case, we do run into an issue when we are summarizing the data. It might lose on the most important aspect of that data. So in that case, what we do is when we write the prompting for the summarization, we also write it in the end that, hey, this is the data that you don't summarize. You just keep that data as is so we don't lose the essence of that data in the summarization. So in that case, we just mention like special instructions at the end of the prompt that we give, and then we write what it needs to exclude from the summarization. So that's how we think that it doesn't hallucinate a lot when it comes to when you're mentioning what is something that it has to exclude from the whole summarization of the data.

Super good. Valentin's asking, can we connect an entire Google Drive folder to a GPT project? Because we've got Google Drive folders for specific clients that we use as a database. Could you use the folder instead of the individual file?

Yes.

Yeah, you can just connect directly from here. Perfect. So I guess we are good on—

Great, great questions, guys. I appreciate you guys engaging. I just had like one or two more. Kushboo, I know we got to cover, but you, you We get to like, all right, so John Tate has a question here. He goes, how do you distribute the projects and GPT links to your team members?

That's a great question. We have a platform Veeam, which we have been developing. So we create agents and projects on Veeam for internal use. We have a ChatGPT team version. So in that case, you can directly share the link of the project. But even if you are on a Plus plan and you want to share the links of the agents, you can always go here. Let's say I want to go here. Let's go to this, Edit GPT, Share. So you can always decide how you want this GPT to be shared. So if you say only me, it's going to be private. If you say anyone with the link, you can save it and then I can share it with you.

Let me just share it with you here.

So anybody here would be able to—

I'll just paste it here.

So anybody here would be able to access this. Now they'll not have access to the whole description or how I have prompted this or how we have built it. They'll just be able to use it like as the front end, just like how we see the website on the front end part. We don't see how it's built. The same way they could be able to use the agent, but they'll not see how this is built. So you can always give the link, make it public and give that link to your teammates and then they can use this agents. But project is something that you cannot share. You can only share project if you are on a ChatGPT team and then, you know, they are a part of your team account and then that's how you can share the projects.

Awesome. Yeah, I think it's an important question, John. Great answer. Let's keep going, Kusum. This is great, guys. How— I love the fact that this isn't like a theoretical session. Like she's really showing you here's how it's set up. Here's how you build it. Just drop a hashtag if you're getting value so far. Let's keep going. Kushboo, this is great.

Awesome. So we'll just see the workflows because this is something that I really want to show. The cloud is something that we could go on later. Let's go to the workflows. All right, we'll do the regen. This has been like a game changer. This automation itself is doing a job which like 5 team members would do on a weekly basis. Now, all those 5 team members, we haven't let them go off. We haven't laid them off. They are helping us fine-tuning this. So don't worry on that part. But yeah, what exactly this does is N8N is something where if you have agents in place, so let's say you have a lead scrapping agent, you have an outreach agent, you have ideal client profile agents. So when you have all those agents in place, and if you want to connect these agents together so they could talk to each other, or they could take the information from one agent and that could be fed to another agent without any human intervention. In that case, these kind of automations are game changer. So here we have used N8N because it's super easy. Even if you're a non-technical person or you don't know coding at all, you would definitely love N8N because it doesn't require any technical background. So now what we do here is we have just grabbed the data saying that, hey, These are the top 10 ideal client profiles that I want to fetch the information for and then create automated outreach emails for these 10 clients. Now, ideally what we would do is there would be a team member who would go on and find information about all these 10 clients, get all the information that they could get from the website, see how it aligns with our business model, create like a whole personalized outreach email. But now that whole thing is being replicated here with this Flow. Let's start this flow and then I'll explain you how this is built. So it just takes the leads that you have mentioned in the spreadsheet, which could be extracted from an agent. Now it's going to extract the websites from the emails that you have provided. It's going to upload it in the Google spreadsheet that, hey, these are the websites. Now we are doing HTTP request. So we are scrapping all the information from their websites that available on the internet. Now that goes to an LLM. We are using OpenAI chat model here. What we are going to do is once we have scrapped out all the information, we are using LLM to summarize that information because it's gonna be a lot of information to process and write how it aligns with our model. So just one question we had earlier is how do we summarize or how do we provide large datasets? So in that case, we just summarize it. So now all the content that we have from the website. We are using OpenAI model to summarize all of that. Once that is all summarized, we are again scrapping the data in a way where we have specified exactly what we want. Like we want their latest LinkedIn activity, we want their latest other social media activity, we want what recently they did, the blog post. So we have mentioned like those parameters, like other than the summary that you have extracted from the website, these are the data we want to fetch on, get from the internet. So that's what it's going and fetching from the internet. Again, feeds the LLM models. It's gonna add it into the summary that it had already created. Now once that is done, we have specified the parameters and okay, find information on employee count, industry followers, the kind of work they do, and then go and update the spreadsheet. So once that is all scraped out, it's going ahead and updating the spreadsheets. Now we have to merge all of this data that we have gathered in all these steps. So we are here, we are merging all of that data and putting it into a spreadsheet. And then once we have all of this data structured, we are comparing it with our service offering, seeing that how does it align with our service offering? What should we be writing about in the personalized email we send to them? And then once that matches our ICP, it goes ahead and provides, okay, What are the emails that you should be sending to this client? So even those emails are generated by LLMs right now, and then it will be updated in the spreadsheet, which we will see. So right now you could also send these emails automatically using ActiveCampaign or Mailchimp if it's like a constant database that you're using for sending these emails. But I would say in terms of accuracy, there is still a lot that needs a fine tuning. So you might not find these emails to be 100% accurate in terms of the way you want it to be. So it requires a lot of fine-tuning. It still has those smileys and then space shuttles and all those things which really looks like AI. So once you fine-tune it and make it look humanized, you can directly connect this node to Mailchimp and it could directly trigger and automatically send those emails to your clients as well. And then once they book a call or if they reply back, that is when you get it on your inbox. Other than that, everything is automated. So we'll just see the spreadsheet here. How it looks.

All right.

So we just saw it was like this. These were just the 3 details that we provided to the automation. And then if we see the spreadsheet 3, it has given like everything, like first name, last name, job title, email ID, LinkedIn, industry, and also like an email. I came across the Media Captain while exploring innovative agencies. I'm reaching out from Eterne Solutions, many digital marketing fails. So this is like a more personalized email from the summary that we have extracted. Rather than a very generic email that you would send to 100 or 200 different people. So going with this approach, yeah, going with this approach, the team spent anywhere around, I would say, 8 to 10 hours to do customized email for 50 to 60 clients. But now with these agents, we're able to send 500 customized emails within an hour. So that's the kind of the impact that we are seeing with this automation. So we are not sending the emails as is right now because super obsessed about what goes to the client, but even getting all this information and like a base point on how this email should be structured is a great starting point, like saving a lot of team's time right now. So yeah, this is one of—

Awesome.

Love it.

So this is like one of the use case and how you build this. But if you're trying to build something like this at first, I would highly recommend that you know, break down this whole automation into different agents. So build your ICP agent, build your lead scrapper agent, build your email outreach agent, build your follow-up outreach agent. And then once you have those agents in place, then you build a workflow like this, which is talking to different agents and then getting some few things from applications you are using, connecting it with the CRM. So don't try to build this at first. Try to build the agents and then these automations will quite be easy for you to just connect things on the agents that you have already created. Yeah, this is one of the workflows like this. We have about 56 automations working right now internally, which is doing one or the other job. This was, this is also an interesting one. Everybody has so many sales calls and then you always need to take 10-15 minutes to prep up before going on that sales call. So what we have to do here is if you go in here, test works now. So what happened here is if you are just providing the website link and the LinkedIn link, it just goes ahead, scraps all the information, creates the whole summary that you need to know. And now you also don't have to go through like the whole documentation of everything that it has created. We have an audio AI node. So you will be sent an email with a 2-minute audio, which you just have to hear before your sales call, and then you are well aware about who you're talking to, what are the questions you should be talking about, what are the things they have done recently. So it just helps you prep up for that sales call, wherein just exactly does the process which a human would do, because that's what we had provided in the prompt. And now instead of just like an email, we are taking it one notch higher where it just gives like an audio note. You can just listen to it and then go on the sales call. So yeah, here, if we see, it's provided for the website. You have provided the link, so it extracts all the information, spit out all the details, fill in the parameters that we have provided. Then when you have provided the LinkedIn link, it scraps out all the LinkedIn information, removes the duplicate that we have already found it on the website, and goes ahead and provide it to OpenAI model, so it could summarize all the information from the website as well as LinkedIn. And once all of those things are done, then we process it, give it to an agent, the agent give it to the Whisper AI, which is like the voice part of it. And then it just creates like an email wherein it's sending you an email as well as the AI, I would say, note as well. So we'll just let this whole flow run. And then at the end, we'll just get like the whole email with everything, which is noted down and summarized. So yeah, we will go back to it. It just takes a couple of minutes to run this because it's like a voice note in it. But yeah, it's super cool, saving a lot of time when you get on, I would say, 10, 15 calls a day. Even if it's saving you 15 minutes, you are saving an hour or 2 hours on your time when you just want to summarize. So this is something which we felt that, okay, we just need 2 minutes of audio note from an AI. But if you want to go in depth, if you want to scrap all the social media information, all the campaigns, all the website information, everything about that client, you could have a 5-minute AI node or a 10-minute AI node where it could scrap all the information and just send it to you in a voice note and an email.

Yeah, that's super cool. A lot of people in your— Yeah, that's really nice. I love the idea of that. Yeah, I think most of us have preset pre-meeting rituals, like pull up the website, run the ranking report, do the conversion analysis. And if you can imagine what it is and it's happening the same way every time, you could probably build an agent to run that for you. That's really cool use case right there.

Right.

So if you just see, it's like Brent Weaver, e2m Solutions, key insights meeting.

And if you just go in here and I'll just play the—

I'm not sure if you guys hear the audio. Do you hear it?

We don't hear it. If you could stop the share and then share again with, like, there's a little button. Oh, play it.

It's just playing, but I guess it's my AirPods, so it just plays in here and just doesn't play on my—

If you hit share and then you hit, there's a little toggle, share with sound, it, we should be able to hear it, I think. So if you stop the screen share and then share again and click that toggle. Okay.

Let me share it again.

All right.

Okay. Do you see it now? Do you hear it or no?

I don't think so. Yeah, I don't, I don't think so.

Let me just download it and provide it here so we could see it.

All right. I'll just go back to the snow. I will have it from my downloads. All right.

Audio file.

Let's move it here. I don't know, I'm not able to share it. Just a minute. Snowy chat. Do I know it's something else? Let me just stop sharing it. All right.

Guys, while she's getting that together, if you have questions while we're still live that you, man, I'd love to get her perspective on this, that, or the other, drop that into the chat for us.

All right. I just shared it. You all should be able to see it now in a while. It's uploading. All right, there you go.

Got it. Yeah, just hit the chat. Thank you.

Yeah.

So yeah, these are all the different use cases that you could build on. So it's all about thinking, rewinding the mindset in a way that, okay, let's just start building the little blocks of AI in terms of the way we operate, the processes we have. And then eventually, once you have those building blocks, it just gives you an idea and it just opens your mindset in a way that, okay, how about I go one step higher? Now build automations around it and then build like a whole agentic AI flow. So when it comes to building like a whole AI agentic flow, basically what that is, if you have every area of the business, like you have agents or automations for in a way, and then if you're linking it all, it just creates like a whole AI agentic flow for you, for your business. But in for every business area, you have like an AI or an automation that runs on it itself, and you just need a control in terms of You just need to do a trigger and then everything happens by itself. So that's kind of fun. Josh, how are we on time? I still have—

I'm fine on time. Do you have a hard stop that I should be aware of or?

I have another call in 10 minutes, so I really wanna show like how to build things on Claude.

That'd be great. Let's get into it. We'll move quick. There are some questions here, but I'll send them to you after. Maybe we can shoot a response.

Sure. So Claude is something that we are a huge fan of when it comes to anything which has logic in it. So if you want to create anything which has like coding or logic in it, I would hands down say go for Claude. So how do you create agents? They have a different terminology in Claude. They call it projects. So how do you create projects in Claude is, let's say, it's a chat here. You go on projects. You create a new project and it just gives you like a brief on what you're working on, the description of it, and then you create a project. I'll just show you one that I've already created. So this is like an N8N workflow templates automation. We have built about, I would say, more than 500 N8N automations so far. So what we did is we have just downloaded all the JSON files of all the N8N workflows, provided that as a knowledge base to Claude, so it knows exactly how to work on those JSONs. So now what we exactly do is we just go here and write, hey, build me a lead gen agent using Google Maps API. So what it's going to do is it's going to create me like a whole JSON, which I can just copy, paste it on initn. And it just creates like a whole functional flow for me automatically. All I have to do is just go and provide API keys and auth keys, and that's all. It just creates like everything. But yeah, how do you create these things? That's a question. So once we have that knowledge base of, okay, you have done enough workflow automations or you might have done enough agents, just have that documented in like a Google Doc or have it documented it somewhere. And then attach that knowledge base to a Claude project so it knows exactly what kind of references or the past work that you have done around it. And now when you're sending the next request, it's gonna go and refer to all the references that you have provided and then give you the automation or maybe an agent or a prompt based on all the references. So that's a big plus for Claude is I would say it's smarter when it comes to you wanted to think logically from the references that you have given. So I would say ChatGPT, if it's like a grade 5 student, Claude is like a grade 8 student when it comes to logic and math and reasoning and knowing into the knowledge base. So we are just taking out the JSON here. Once this is done, so you could build your own end-to-end workflow automation here. And then I'm going to share like a resources link where I I have given all the nodes, all the JSONs of all the Anitin automations that we have created so far. So you can just go download that, provide that as a knowledge base to Claude Automation, and then build your own Anitin workflow automation template. Once this is done, I'm just going to show you how this works and how you could be able to use it. So yes, Claude has already created a JSON for you. So all you have to do is just copy this. Go into a platform like initn. Let's take a completely new node. So it's completely like a canvas right now. So what I'm going to do is I'm just going to go and paste the JSON here.

All right.

It's working. Let's go back again to Claude. Copy it and then paste it here. I don't know, it's not working right now. It's not working. Hey, can you give a smaller version? This It doesn't work.

So it might not be like a complete JSON file. So it's not working. So it's not printing it on initn. Yeah, so you don't need to do all these initn automations from scratch yourself. If you have resources, just directly plug into the knowledge base here and then use a project to create like a whole automation template. And just go and create automations on it. This is a simplified version.

Let's see if this one works. Okay, we have copied it. Let's go on here. Okay, this works.

Wow. Nice.

So now what exactly you do is you just go ahead and connect your applications in it, wherein you just say that, okay, AI agent, let's bring it here. So all I do is just go ahead here, see, I'll connect it with my OpenAI account, just set it up here. The scaling temperature, I guess we'll make it 5 because we want to make it more precise. Just go ahead there. The Google Maps API, we already have provided like a basic schema in the notes. You don't have to go and write it everything in the note here. You just have to connect with your account, take your, like the API key and just provide it here. And then you'll have this up and running. The whole point is we don't get to do all these things like from scratch. If we really want to build these things, there are materials available out in the internet that you could use to build these workflows and automations for you. And I'm going to share the resources link as well. So I'll stop sharing it and I'll share the resources link, which has all these cheat codes and cheat sheets and all the workflows and automations that we have built so far.

Everybody's eager to jump on that link and get access to that. Super generous. Thanks for sharing that.

Absolutely. I really want everybody to use this, start building agents, start creating automations. And then in case if you feel you need any help in terms of the logical functioning or the segmentation of it, feel free to message me. We are very big on having AI kind of penetrate through the agency ecosystem. So we'll be more than happy to help.

Kushboo, this was awesome. I love the fact that you broke this down for us and actually showed visual examples. I love the use cases. I love like that you shared a valuable resource. Thanks so much for sharing.