Transforming Agency Operations with AI Agents (not just ChatGPT) with Khushbu Doshi | Ep #784

Transforming Agency Operations with AI Agents (not just ChatGPT) with Khushbu Doshi | Ep #784 Speaker 1: I want you to stop working for free. If you're tired of getting ghosted on proposals, the endless follow-ups and We'll think about it as a response. There's a better way. The top agencies aren't sending proposals. They're getting paid to pitch and close bigger deals way faster. I want you to grab this free resource I just put together at agencymastery360.com slash strategy. That's agencymastery360.com slash strategy and you'll learn how to flip the script if you want to get paid for your expertise instead of giving it away for free. Speaker 2: Welcome back to the show, Khushbu. How are you doing? Speaker 3: I'm doing good. Thank you for having me again. Excited for our talk today. Speaker 2: Yeah, we're going to talk about AI and how it's really changing the landscape and some of the cool stuff that that you're seeing out there. But for the people that have not heard of you or know you yet, who are you, what you do? Speaker 3: So I'm the COO at E2M Solutions. I have been with the E2M since I would say eight years now. We work with agencies. We are white table service providers. With my experience of working with more than, I would say, 500 agencies, I stepped my toe into AI eight months back, did a lot of research, implemented a lot of automation, saw the productivity boost by 20% every quarter on quarter, and now we kind of educate agencies on how to leverage AI and then help them with the AI journey, like taking the first steps. Speaker 2: You know, I learned a lot. You know, you spoke at our event at Elevate in Phoenix not too long ago, and you were talking about AI agents. And I think I'm like most people running agencies of going, yeah, AI. You go to ChatGPT, you go to Gemini, you ask it questions, and it gives you really cool stuff. But I wasn't really familiar with what an AI agent was. And then when I started diving into it, I was blown away. So tell everybody what an AI agent is or what it is, how it's different from just going to ChatGPT. Speaker 3: All right, so AI agent is something where you can define that what exactly it needs to do. So usually when we start using ChatGPT, so let's say I have a client email that I want GPT to connect for me. So maybe I'll just go on a chat and paste that thing. Now again, if I have something for that same client, I might go and search that chat and see where it has all the references and all the memories and the context about that client. So I don't have to feed the same information again. So now you go on and search that chat. Now, when you're kind of pacing on that chat, you feel that I'm getting more, I would say, contextual responses, which is more specific to that client because you have kind of inputted all the client's memory in that chat. And that's the reason the response that you get in that chat is very different than a very generic chat. Now, to solve this thing, ChatGPT came up with agents where you can create a custom GPT where you can train the GPT to behave in a specific way. So now you don't have to go and search for chat. And even in chat, when it gets lengthy, it can just go back and retread to 300 last messages and nothing above that. So now to solve that limitation, you can always go ahead and create custom GPTs where you can feed specific information about any of your business domain or a client or maybe a content outline or anything that you think of. So, when you know that you have a very specific set of instructions and this is how you want your ChatGPT to behave, this is what an input would be, and this is what my expectations are. This is what the ideal output would look like. So, when you have everything laid out, an agent would work better in terms of any other normal conversation that you have with ChatGPT. Speaker 2: Yeah, I mean, after your talk, I started really diving into it. For everybody, like just as an experiment, you can go to ChatGPT and then you can go to like explore and then you can hit create. And when you hit create, you can give it a title and you can tell it its role. I want you guys to think about this agent as a role. So one of the ones I created was this marketing strategist that will help you out with landing page, email copy, all this kind of stuff I wanted to do. But in order to make it smarter, What I started doing is taking all of our success stories over the past 11 years. So all of these success stories, I almost said like sex stories, that would be really bad. That's a whole other topic. But all these success stories was in Trello. So I was able to export that data out. I was able to import it. Into this agent, I was enabled to also upload our branding guidelines, our brand voice. I was able to upload our past clients' biggest challenges, like thousands and thousands of data points. So now I can share out this URL. So a lot of times we hire out a lot of agencies or we bring on a lot of people. So, I just send them this market strategist link and said, ask it any question that you want and it will help you. So, if I go, what's a good email I should send out on April Fool's? It could say, oh, there is this story of Zach who went from 10,000 to 800,000 by sending out this one April Fool's email. Use this. And it would write it for you based on all the data points you give to it. It's pretty unbelievable. You also use it for onboarding your team members. Talk more about that as well as QA testing because that was fascinating to me. Speaker 3: Yes. So one of the major, major, I would say breakthrough that we got when we kind of started building this agent was people always, always, I would say, feel very lethargic when it comes to reading SOPs of 50 plus pages and reading all these procedures, which they don't know when they're going to implement it, but they still have to go through it because it's kind of in the onboarding. So we're like, how about we make this dynamic where we can get all these SOPs in one document, feed that document as a knowledge base to the agent, and then train the agent in a way where we can say that, hey, you are an expert dynamic SOP agent, where you just give the information, whatever it is asked to you, don't give anything, you know, I would say additional to what is being asked to you and do not search internet. So I guess that is the biggest thing that we always, always forget is we always stick on search for web. But when you know that you have a very defined knowledge base and you don't want a ChatGPT to go out of it and add its own two cents, just, I would say, get rid of that check mark, which says go on web. So it will not go on web and it's going to give you results from your knowledge base itself. It's not going to give any additional results. So that way we kind of gave our SOP as a knowledge base. And now any new client account manager we onboard, we don't give the SOPs where they have to read everything in the first week itself. They can just go and write in, hey, it's my first day today. What do I need to do? And it can give you a step-by-step approach. Once you're done with that, you can say, hey, I'm done. What do I need to do next? So then it's going to give you incremental information just as when you need it instead of overwhelming you with all the information. So that was one of the biggest, I would say, breakthroughs. And it's not just that we made it and we didn't use it. We did use it. So we onboarded about eight new account managers after that, and we gave that to them. So we gave an SOP guide as well. And then we gave this agent as well just to see the human behavior, what do they choose, what they feel more comfortable about. I would say they were very fascinated just by the idea of they have somebody, they don't have to go to basically somebody and ask these questions and feel, can I ask these or can I ask that or how about I'm asking too many questions. They know that they have an agent that's not going to judge. They can just ask as many questions as they want. If they want to see what's in there for them, you know, what kind of services and procedures and everything. So it's always based on curiosity that you could ask the way you want it. And I feel now we are at a stage where after evaluating everything, we don't even give those physical SOP guides anymore in the onboarding. We just give the link to this agent and it works kind of really well. And the same happened for quality testing as well is when you know that, okay, you have a business domain where you have very specific processes laid out in terms of you already have a checklist, you already know what needs to be done, how that needs to be done, everything is laid out super, super precisely. In that case, these agents work really, really well because you already have the data that is needed to train the agent. Now all you have to do is just create the agent and give the information that you already have laid out in terms of processes. So we have this 500 points checklist, which we make sure about when it comes to delivering a new website to the clients. So now what we did is we created an agent and said that, hey, you are a quality testing expert. Make sure you pass through all these checkpoints and just give us the report on all these checkpoints. Now there might be some limitations based on the model that you are using or the properties or the parameters that you have in the checklist. So you also have to mention about the fallback strategy, that if there is something that you are not able to check right now, just don't have a tick mark on it. Just give me a note on these other things I cannot check, and this is why I cannot check. So now it's also like after 500, I guess it's checking about 350 to 370 points. And whereas it just mentions out, I would not be able to check because it was a back-end access needed, or it was a third-party integration, or anything that it is. So now we have like a whole report. So now instead for a QA to go and check all these 500 points, they just have to go and check this remaining 150 points. So that was also amazing. Yeah. Speaker 2: Yeah. And I just want everybody because just was blowing my mind. An agent works as a role or a team member and one specific role and what you can do, what really makes things really neat is you can make agents like your marketing agent communicate to your creator agent and they can talk. So like think of if you were building out a social media team, what do you have? You have the CMO, you have the strategist, you have the creator, you have the tester, you have all these different roles. You can create a workflow for all these agents and it's always working for you. And then you can have human beings add elements like one of the things that we're building out right now for agency mastery is. There's a lot of people that are just spitting out content from AI all day long. I want this team that we're assembling, these AI agents, to give us ideas. And then when I say, yeah, I like that idea, create that. But I want to review it and still put my personal touch on it outside of that. But it does 90% of the work, which is incredible. Speaker 3: And with the ChatGPT task now, it has become so, so easy for you to kind of even map things out what the competitor is doing. So let's say I have ChatGPT task, which tells me whenever their website copy update on any of the competitors or whenever they are posting any blog posts, it gives me a summary of it. What is everybody talking about in the digital space? How is SEO evolving? So you can specify these kind of topics. And set a ChatGPT task so it could just give you all the information that you need just once a week. You can say that, okay, give me all this information on my Monday morning or like a Friday evening. And then you don't have to go through all these different sources to kind of scrap this information out and summarize it for you. The ChatGPT could do it for you. And today we're going to talk about, you know, leveraging the existing features and kind of seeing how we could use and leverage that model for the real time applications of it. Speaker 2: For all of you listening to this, your guys head probably hurts just like mine when I was First, really kind of really understanding it and especially diving in. We're like, you know, Khushbu did, you know, what's working for the mastermind members and really dove into one of her workflows and AI agents and all this. And I was getting chats from all the members are like, my head hurts. I get it, but it's already here. The people that leverage AI, I started my agency really when websites started taking off and the internet was really this resource, this amazing technology advancement. AI is that times a thousand. And if you don't really start looking at it outside of just going to ChatGPT or Gemini and just building prompts or building little things and not looking at how can you streamline your agency to a part where it makes you more efficient, to make you deliver bigger impact. Every one of your clients is looking at you to solve the biggest challenges. And one of their challenges is AI and how it's going to disrupt their business. So you, if you can start understanding it now, getting ahead of all of that and really start utilizing it for your agency and for your clients, you're going to succeed greatly. Like this is a huge advancement that I barely understand it. And I'm so excited about it. Do you have the same feeling Khushbu? Speaker 3: Yeah, and I guess with all the new models that kind of are enhancing itself, you just feel like every day there's a new breakthrough, just like recent ChatGPT image generation. Now they had DOLL-E, which used diffused model. Now they have this image generation, which is an altogether different algorithm and the logic. Now they're kind of working on a more advanced version with Gemini's new work in the AI Vortex Studio. So I guess the way that we have been working is definitely going to change a lot in 2025. Right now, if I'm mapping out how it has changed the digital space, like when I talk to people who have already adapted AI, what kind of building products on AI or maybe integrating it heavily You can definitely see that change. So now instead of writing long codes, people are using Cursor AI to write the codes for them and just preparing the user flow, which they could feed to Cursor AI or maybe how to write the prompt in Cursor AI. That's the thing that they are learning. They're not learning how to code in Python or how to code in kind of, you know, PHP. So that's kind of evolving. That landscape is evolving. And now in business analysts are no more creating the documentation of how that I know user flow is going to work. They are kind of preparing this flow of automation on what the data points are going to be, how the automation is going to work. So I would say every business area in the digital space is evolving, and we might not be doing the same thing that we were doing back then. Now we are evolving it in a way that how we could incorporate AI, leverage it for what we really think that we would have more time for, and really save that time and invest it somewhere else where you kind of lost funding. Speaker 2: Yeah, it's incredible. Well, this has been amazing and we've probably heard a lot of people's minds. Is there anything I didn't ask you about AI that you think the audience needs to hear about? Speaker 3: Yeah, I guess, you know, one common notion is we always consider ChatGPT or OpenAI or Gemini as AI. And then we kind of always feel FOMO when we see all these fancy tools was doing all this fancy stuff. But I guess the models itself, if you are leveraging those models, just know that how to use that. So for ChatGPT, we already know that we can create custom agents and projects and tasks But do also check out Gemini. Don't forget to check out the Google AI Vertex Studio where you can stream online. Literally, you can share your screen. So where you are stuck, it can give you audio on what to do as a next step. So think like it's an expert 24-7 there to guide you if you're stuck somewhere. So not everybody's aware about these tools which are there in the market. We know more importantly just Gemini and LLAMA and then ChatGPT. But there is much more than that. So don't forget to check out like Google's Notebook, Ellen, Gemini and Google AI Vertex Studio. Speaker 2: Yeah, I love it. And everyone, don't worry if if you don't want to put Hundreds of hours in understanding this. We do have a solution for you. You can book time with Khushbu and her team. If you go to their site, you go to e2msolutions.com and look for fractional AI consultant. And look for that and you can book time with them where you can have them build all this for you. You just tell them the logic. You tell them the process and the things, how you have it now. And if you could use AI and leverage it, they'll build it all for you. So they're building a lot of cool stuff for us right now. And I highly recommend you guys go check out E2M solutions and reach out, have a conversation. It can't hurt. I don't know many people doing this and they're crushing it. So go check it out. Well, and until next time guys, Have a Swenk day!