DeepSpecialization_EP 96_Khushbu Doshi_Video_Edited_V1
===
[00:00:00] but I feel this is like a drug, once you start seeing the, efficiency and the productivity boost, now you wanna just put in more and more time and see that, okay. What else it could do, how it could affect my revenue. honestly, Corey, I feel. Anybody you feel who is super excited about it and who is not scared about it, is the person that should be held responsible for doing it.
[00:00:22] You're not gonna lose your job to ai you, you're gonna lose your job to people who are leveraging AI better, who is using ai.
[00:00:29] You never know that when that tiny bit is going to be a huge automation because that tiny bit is going to get, you know, I would say people are going to get addictive to it. Welcome to the Deep Specialization Podcast, the show where we blend focus, strategy, and client intimacy in order to scale and simplify our businesses and our lives. I'm your host, Cory Quinn. Let's jump into the show. Welcome back to the Deep Specialization Podcast everybody. Today I am thrilled to have Kush Brew Doshi with us.
[00:00:59] Kush Brew is the Chief operating Officer at E two M. They're a, an amazing company. They're a sponsor of this podcast, and what they do is they specialize in helping agencies grow their business with. White label services, they're one of the best in the market, and I'm particularly interested in talking with Kushman today because we're gonna dive into the role of AI in agency operations.
[00:01:28] If you're an agency founder, and I know there's a lot of you listening to this show, and you want to scale your revenue by leveraging AI in the business, you will not wanna miss this amazing episode. Welcome Kush to the show. Thank you, Cory. Thank you for that wonderful, wonderful introduction and I'm kind of excited to share all the AI magic and knowledge that we have gained so far.
[00:01:54] So to kinda kick things off, just to give listeners more context of you and the business, E two M, could you share anything about the business, what you guys do, type of agencies you work with, the services you provide? Sure. So we have been in the white label space since last 12 years now. So from the day one, we have been just working with digital agencies, just like our listeners.
[00:02:17] We work with them day in and day out. So we exactly know what their pain points are, how to help them scale, and then kind of help them with all the different executions. So I guess, you know, the way that we kind of. I would say pivoted into AI and kind of have introduced that as a services, because we know that there are a lot of agency owners who really want to include ai, who really wanna know more about ai, but they don't know where to start.
[00:02:42] So I guess that's, that's where, you know, we are kind of trying to do that. Awesome. And about how, so you've been in business for 12 years. How many agencies do you guys serve? So right now we work with active 357 agencies, and so far, I guess we might have worked with more than a thousand or 2000 agencies, so that that's what we have been doing since last 12 years.
[00:03:03] So, yeah. Mm-hmm. Yeah. So I imagine based on that number, a lot of agencies, a lot of clients that those represent, right? Each of those agencies have a lot of clients that you're helping 'em out with. So you probably have a lot of. Surface area, a lot of opportunity to figure out how to leverage AI to help not only your, the internal workings of E two M, but also for your agency clients.
[00:03:26] Is that, is that fair to say? Right. That's fair to say. And Corey, I guess the way that I started, I never thought that if you wanna give that, give this as a service to our agency clients, because it's like when you know, you kind of start seeing AI and then kind of start reading everything about ai, you always feel that, how about that?
[00:03:45] I use like a tiny bit of an. Bit of it and see exactly what it could do to the business. So I, I kind of started it as, I wanna integrate AI in E two M and see that okay, how it could boost efficiency and then profitability and how it could impact revenue. And just by like I would say two months, three months into it, there's no results that I could see And, and then I'm like, I don't know if ai, AI is really a thing, is it really going to help business tat PT is really good for content, for emails and for all the other different things, but I don't know if it's really gonna make some difference how, how to use agents, how to build agents.
[00:04:20] So that's where that journey started to kind of see that. Is it really worth spending that much of time? Like I've spent almost like what thousand plus hours in AI research by now just to see that. What could be the applications of AI in the real business? What are the solutions we could build around it and then go, I.
[00:04:38] A level deeper into it and see that, okay, it's not just about AI or LLM models, it's much more than that. It's about how you are training those models, how you are kind of connecting those different agents to talk to each other and create an output that you really want. So I guess that's where the journey started, where we wanted to do something for.
[00:04:58] And then when we felt that, okay, now we have really good use case of. Everything that we could do on each of the business areas that we have. Yeah. And we started seeing that impact in efficiency, in productivity, in the revenue. So we were like, okay, how about we leverage this knowledge that we have gained and then we could use this and give it to the agencies where now they could increase their efficiency, their profitability, and give that as a service to their end clients as well.
[00:05:25] Awesome. Well, I can't wait to dive into the details and the specifics here. So. Let's start with, you know, where should an agency maybe who is leveraging Claude or some of these AI tools for more copywriting, but really haven't been leaning heavily into ai, where's a good place for agencies to prioritize, leveraging and really integrating AI into their agencies?
[00:05:50] I think the first step is kind of identifying that what really AI mean to you. So do you really wanna use AI for the content part? Then in those cases, these models that are kind of prevalent in the market, which is chat, TPT, Claude are really good where you wanna just limit the excess of AI to. You know, just the content part of it.
[00:06:11] But do you really wanna do things with ai? When I say do things with AI, is either do lead generation with AI or create some meta titles or some tags with ai or create like a blog post outline with ai. So do you really wanna do things with ai? If yes, then okay, let's, let's kind of lay out on what are the different things you do on a day-to-day basis in your businesses?
[00:06:33] Where is somewhere where you spend most of your time? I. Either has a process around it or it's a repeatable thing. It's a no brainer thing. It doesn't involve any creativity. So just creating a list of those things and then the next step is okay, breaking down that, uh, whole thing into different objective task.
[00:06:50] So I guess biggest, biggest lesson that I learned, Corey, when I started this is, yeah, I was, okay, I wanna build a sales agent that could exactly do what I'm doing right now. And that's what the problem is. It's like we always wanna build something which is like an expert. It should work like us, it should behave like us.
[00:07:06] The output has to be like us, but AI is kind of ever evolving. So you're not always going to get the result the first time, which is going to be perfect the way that you want. So you end up with either frustration or you just feel that AI isn't capable of doing what I really want it to do, and I don't think so.
[00:07:22] AI is gonna be helpful in what I'm doing because what I'm doing is kind of completely unique. Sure. But then I took a step back and I was like, okay, if people are using it, if it's making sense for different businesses, why isn't it making for me, there has to be something which I might have not figured out, which is something, there is something which is not right.
[00:07:40] So I'm like, okay, how about that? I break down that whole expert task into different objective task. So either of creating like a completely expert sales agent, which could do everything. How about I break that down into different chunks of what a sales agent is supposed to do? So let's say a sales agent is supposed to scrap the data.
[00:08:01] A sales agent is supposed to generate the leads for me, qualify the leads for me, summarize the leads for me a draft, customize emails for me, send those emails, get the responses, create follow-up sequences. So these are the different objective tasks, which is going to make an expert agent. So now, instead of creating an expert agent altogether, who's going to do everything for me?
[00:08:25] I started creating these objective task agents. So one thing at a time. First thing I created is the lead gen agent, the data scrapping agent, the follow up agent, the email agent, and that's where I started seeing the success. So instead of going for something which does like 10 jobs at a time. Just, you know, just identify, okay, which are the different areas that it could be divided into, segment it out, and then take one thing at a time.
[00:08:52] So now when I have all these 10 different agents, it becomes very easy for me to now create an automation around these 10 agents where they can talk to each other and give me the output that I always wanted to. Got it. So you, you take a larger task, maybe more global task or, or, or role within the organization sales.
[00:09:10] Yes. And you break it down into these objective tasks, and then you build a, an AI prompt or a tool to somehow execute that specific discreet task. And then you can do the same for all of the objective tasks within that. Function. That's right. And then you effectively automate the process where one goes to the next.
[00:09:31] Could you give a, an example, maybe of one of these objective tasks? Can you give us some detail? You mentioned a lead, lead gen agent. Like what does, what does that look like? Uh, so lead gen agent is basically using open ai. You can create custom agents. So now I created a custom agent where you can just say that, Hey, find me a hundred digital marketing agencies in New York.
[00:09:53] So it could go in, we have a Google map, API that is integrated into the custom agent. So just like how a lead gen expert would do it. Go into the Google Maps, give the county or specifics of, okay, this is where I wanna find the agencies from in New York. So just like that we have kind of, uh, I would say prompted and trained the agent, where if I'm writing a prompt that find me, these many agencies from this state, particularly with these many employees, and the companies should be at least like 12 years into the business or 15 years into the business.
[00:10:25] So based on all these parameters, now what it is doing is it is going in and fetching the data for me, and then it's creating an Excel sheet for me. So that's one part of it where I don't have to kind of put in any manpower to get in all this information. I can just mention the, uh, parameters of how the information or how the data should look like, and it's kind of scraping that data for me from the internet.
[00:10:52] Sure. And the assumption here though is that it's only able to source data that's available on the internet. If there is bad data on the internet, which there is, or if there's outdated data that doesn't, it doesn't qualify actually, the the leads. It just, it just goes out. You tell it. I'm looking for agencies with the minimum of 20 employees in New York City.
[00:11:18] It'll go across the web, do searches, try to find these people, put together that list, and then give it back to you. But it's. The assumption isn't that it's a perfect list. It is a, it's a probably a pretty good list that you need to take, you know, further, further steps. I. Yeah, that's, that's where this whole AI game starts, is now scrapping the data and preparing it in Excel sheet is very simple.
[00:11:43] You could get it from a LinkedIn sales navigator as well, but now training the model where it could identify, does it match with your id? Client profile is somewhere the AI plays the role. So now the way that models are trained is we have given the model, like what does an ideal client profile look like for each one.
[00:12:02] Like the company has to be in the business for like five to 10 years. They should at least have five to 10 people. They should be giving WordPress, SEO content, all of those things. So we have given like a complete detailed, I would say, documentation on what an ideal client looks like for EM. So now what it does is this LLM processes all of the data that the agent has extracted and treats it like, okay, does this match the ideal client profile of E one?
[00:12:31] Yes, then it qualifies the lead. If not, then it just gets it off the list. So now you are not just kind of scrapping the data, but you're also qualifying the data with the LLM model that you have trained to kind of identify which are the ones that you should be talking to, and which are the ones that you know is, is not badly your ideal clients as of now.
[00:12:52] I love it. I love it. So we're talking about more of a sales and marketing application of applying ai. I know you, you focus a lot on, on sort of the operational side of the agency business, but what would be, let's say, for an agency who wants a quick win? Like what would be one. Maybe objective task that's universal across agencies that they could apply in their business to get some, some nice, you know, value from ai.
[00:13:18] Anything come to mind? I'm curious. Yeah, yeah. I guess one thing that has been proven, and it works really, really well for all the agencies that we have been working with, is creating client profiles in chat, TPT. So usually the common practice is you just use a chat or a thread on OpenAI or a chat like a chat GPT, where you just put in all the information about the client.
[00:13:40] Yeah. Yeah. And then next time, whenever you wanna talk about that same client, you just go and kind of search that thread that that, okay. Which thread did I talk about? For this client, so it just gives me relevant information about, so I don't have to give any backstory about that client. But I guess the way that it could be done better is using chat TPT project.
[00:14:00] So now if you see the chat TPT project folder, just create the folder with your client name. Give the global set of instructions about the client. So any person about the client, the business that they are into, what are the services that they're taking from you, any specific information or guidelines that they might have given, any successful projects that you might have delivered for them, any emails on which you might have received, like really good response from them.
[00:14:25] Uh, the tonality, the brand wise and everything. So now you could use chat TPT projects for it. So now, next time you don't have to go and search for the thread. About where did you talk about this client in chat, GPT, which thread or how many days before? You could just click on that project about for the client and just put in anything that you may have received.
[00:14:45] It could be an email, it could be a proposal, it could be a content requirement, anything. So now the answer that that you are going to get is going to be more personalized to information that you have fed for the client. As compared to any other information, because what happens in the chat, it's gonna take the context of the last 300 messages.
[00:15:06] And if you're going to go beyond that, it's gonna forget what happened before it. So now you're not going to get a complete context of everything that you might have shared about the client. So, which has been the case with the project. So I guess if you are, if you start creating projects for the clients, that's gonna be a game changer.
[00:15:23] I love that. That is so powerful, and I agree. It's something that. Any of us can do immediately, right? I mean, if, as long as you're familiar with what projects are and it's, uh, it's fairly straightforward. So I, I love that idea Kush. Thank you. Who within an agency, organization should be dedicating time to AI research and exploration?
[00:15:44] Like who, who owns that in the agency? Yeah, so honestly, Corey, I feel. Anybody you feel who is super excited about it and who is not scared about it, is the person that should be held responsible for doing it. But I would say that initially when you talk about ai, I would say I. 80% of the team would think either that it scares them out or it's just, they just feel that, okay, none of what I do could be done with ai.
[00:16:12] And I don't think AI is gonna be like, you know, really useful. But I guess at the executive level, if you just choose one person out of each team, so let's say you have a content team and then you have an SCO team, or you have a web dev team, just choose one person who is an Evid learner, or you feel that who is not kind of, I would say.
[00:16:33] Not feared enough where if there is something which is challenging to what they do, they'll kind of adapt, learn more and kind of go deeper into it. Yeah. So just identifying that one person who could be that person who could bring in different initiatives is the one that you should be relying on.
[00:16:50] Because usually when we speak with agencies and their teams, we feel that, okay, the owners or the CTOs or the CEOs, they're kind of super excited that, okay, we wanna have that EI Edge and we wanna do. This with AI and that with ai. But then now when we are talking with people on the ground level who is going to implement these AI solutions that we built for them, they're like, ah, but I didn't think it's gonna make my life easier.
[00:17:13] Or, how about that we do these many tweaks, or, I'm not getting what I really wanted, but. The thing is, at the first time, anybody is not going to get the perfect solution, but it's all about adapting it and giving us feedbacks to make it the way that you really want it to perform. Yeah. Yeah. So to answer your question, somebody who is kind of really, really very curious and who is kind of excited to integrate AI would be the one who is going to fine tune the model and make it the way it should be performing.
[00:17:42] What's a reasonable sort of timeline or expectation around when you should start to see tangible results from. Let's say allocating internal resources away from clients or other things and really investing in ai, like how, how quickly should an agency expect results? I think the first couple of months would be a learning curve because there's a lot that's happening in the AI space.
[00:18:03] Every other day there's some news where you have to kind of put in more time and kind of see how that is impacting the business solutions that you're building. So I, I feel ideally, a couple of months of time is enough. Uh, for an agency to kind of go do a deeper dive into it, build solutions, and kind of start seeing some results in terms of efficiency and productivity.
[00:18:25] But I feel this is like a drug, you know, where once you start seeing the, I would say efficiency and the productivity boost, now you wanna just put in more and more time and see that, okay. What else it could do, how it could affect my revenue. Yeah. So a couple of months of time is, I would say, enough to get.
[00:18:42] Used to how AI could implement and kind of refine and redefine the whole flow. And after that, you are never going back to not using ai.
[00:18:51] You mentioned a moment ago about some people in the organization maybe intimidated by ai. I may thrill feel like their job is threatened, whether that's. Realistic expectation or fear, that's to be determined.
[00:19:04] But how would you counsel agency founders and executives who really want to lean into AI but are also sensitive to the, the fact that some of the employees may really not be as excited as they are, like, how would you advise them and, and coach them on how to navigate that situation? So I guess the way we did is we did these, uh, quarterly AI demos where we would want to incentivize people who are coming up with new AI initiative, showing us how it could boost their productivity.
[00:19:37] They could do that on a quarterly basis. So if I'm seeing somebody doing a demo on AI and seeing that, okay, what else AI could do, I'm kind of inspired if it's my teammate who's doing that, or if it's somebody that. I feel that, okay, how about I did not get that idea? How about I did not do that? So I guess inculcating that feeling of a competition within where we feel that, okay, if I don't do this, maybe I might have, I.
[00:20:04] An outdated knowledge, or maybe I might not be as competitive as the other person is. The other is AI news and updates. So basically just having a team, uh, who is kind of just, you know, regenerating the ai, uh, news that is happening with the business that you are into. So, just like how the model context protocol came, MCP.
[00:20:26] So it's, it's about like a new terminology, a new thing that's happening in the AI space. So now it's, it's like we don't need to know how CPS going to kind of act worldwide. How, what are the problems it's gonna solve? What we need to know is how it's gonna impact the digital business or the digital agencies, or what are the business applications of it.
[00:20:46] So there has to be a team who is kind of refining this whole AI concepts and everything that's happening in the AI space for the digital business. The other thing is just making them understand that this is not a, it's like a counseling. That this is not here to replace the jobs. It's just helping you do something where it's going to save you the time, where either you could do something or you could spend more time on something which you really love.
[00:21:13] So let's say if you really love writing content, so now all you have to do is you already have the baseline, you already have the outline of what the block should look like. You already have somebody. an AI who has done the research for you. So now you're just spending time into something which you're loving the most.
[00:21:29] So now, if you were able to just write one or two blog posts with all of these AI tools, everything that's helping you, you could either write three or four. So it's just helping you amplify your own output where you're doing more. Yeah. Then what is expected from you. So doing that one-on-one counseling really helps where we just make them understand that this is not here to kind of take your jobs, but it's here to just make your jobs easier.
[00:21:52] Yeah, I think of it like, you know, change management. I worked at a, at a company, we started off with a hundred employees. How many, how many employees do you have there at E two M? Right now we are at 323. Yeah, so we went from a hundred to a thousand employees. When We were a hundred, when we were a hundred employees.
[00:22:10] We were like a small little ship on the ocean where we could turn right and turn left and do a 360 and it wasn't a big deal, right? Everyone was on board, and when we got to 300, 500, 700, a thousand people changing a couple degrees, you know, in direction as we're going along, created a lot of opportunities.
[00:22:31] I'll say the positive way, opportunities to practice change management, which is. Finding ways to communicate to the organization, Hey, this change is coming, this is what it's about and this is why it's good for you. And getting ahead of it with communication. So I think of it as, you know, being, being a good leader is part, partly being able to help make sure that the company is.
[00:22:55] Right. You're able to communicate changes in a way that the company feels confident and they're leaning in versus left behind. I think that's a big skill that, yeah, it's a soft skill and AI is certainly one of those, right? It is a big topic. It's still, as you said, rapidly changing, rapidly growing, and as a result of that, there's a lot of uncertainty.
[00:23:14] It's hard to define what this actually means for people, and I love the saying, I'm not sure, I'm sure you've heard it, but. It's that you're not gonna lose your job to ai you, you're gonna lose your job to people who are leveraging AI better, who is using ai. Yeah, that's true. Exactly. Yeah. And so, you know, this is a new technology and, and I think the, the, the baseline is that you need to, you need to embrace it and find ways to leverage it and, and use it to help you to create more value for your employer and for your clients.
[00:23:44] So. Yeah. Yeah. But I feel Cory, as an employer, like as an agency owner, if you are obsessed about AI and if you say that, Hey, I really wanna use AI in everything that we are doing, even if it's a tiny bit of it, or if it's a huge automation that we are building, I feel that reflects into your team as well where they are now not threatened to use AI in what they're doing.
[00:24:07] Yeah. It's just like, okay, this is the way that we are gonna work moving forward, where we are going to use at least. A smaller aspect of AI into whatever we are doing, even if it's a email generation or maybe just, you know, content generation or SEO or anything that we do. Just use a tiny bit of it now.
[00:24:26] You never know that when that tiny bit is going to be a huge automation because that tiny bit is going to get, you know, I would say people are going to get addictive to it. So when you start using it, a tiny bit of it, you're kind of going to expand that horizon of AI that you're using on a day-to-day basis.
[00:24:43] And eventually in 4, 5, 6 months time, then you'll see that majority of what they are doing right now is now they're leveraging AI for it. So just like, you know, helping them to do that Ignite and then the start. Yes, I love that. How do you think about measuring success or ROI from AI related initiatives in the agency?
[00:25:03] I think the way that we do is time to value. So if, if we are spending, let's say we talked about lead generation. So we had like three people team where they spent about four to five hours of their daily time on finding these leads, and then a couple of hours qualifying these leads. So if we are talking about five hours, three people, it's about 15 hours job on a daily basis.
[00:25:28] Now if we are creating an agent or an autonomous floor around it, let's say it's gonna take us 60 to 80 hours to build this thing, but now the ROI that we are going to get out of it is uncom comparable to what the team would have spent if they would do it manually. So that's how we map it out. Time to value where, okay, what is the time we are going to spend to create something, how much value it's gonna bring.
[00:25:54] How much time they are spending right now manually to get the same results that we are getting with these AI agents. So one is the time to value and the other is the KPIs. So basically how much of the productivity boost it was. So eventually every quarter the target is the productivity should at least increase by 10%.
[00:26:15] With the ai, so you could either use it in development or quality testing or any other aspect of business that you're doing, but the goal of every team has to be like, you have to use AI to boost your productivity by 10% every quarter. So now what happens is we are posting the productivity 40% over the year, just breaking it down into 10% every quarter, and then we map it out that, okay, what went really well, and then what are the things that we should be working on?
[00:26:45] That's a great way to measure. I love both of those. Time to value really clear, right? Manual versus leveraging these tools. How quickly are the, or what's the impact of the tools and the AI in creating value from a time perspective as well as the, the 10% productivity gain per quarter. That's a great metric.
[00:27:02] Uh, especially in the world we live in. That's really, really strong. I like that. Who, who owns that metric in the organization? Is that the CEO or is that the, the head of content? Like who, who would own that 10%, uh, productivity bump per quarter. So every division has one leader who's kind of working on this productivity, who measures that.
[00:27:21] If a developer, let's say we have a team of 200 developers, and if they could produce like 10,000 manpower hours with the same team, they should be able to produce at least, you know, 11,000 manpower hours. So every team has a head who is responsible for this, and the CEO kind of gets the report at the end of every quarter.
[00:27:41] Like, what was that 10% boost look like? Like in terms of numbers? Yeah, so there are a lot of sort of off the shelf AI related tools and platforms like Chat, GPT and Claude Perplexity. Grok. I'm sure I'm leaving out a few. Do you use any more custom built solutions? Like are there other, other tools that are not so off the shelf that are worth looking at?
[00:28:06] I would say 70 to 30% ratio, where 70%, we use the tools that are more prominent and which is kind of very easy to train and fine tune and get the results that we are getting because we always quantify and kind of, uh, see that, okay, if there is a problem, which is being faced by so many people or so many agencies, in that case we'll just use the tool because the tool is definitely solving that problem.
[00:28:32] Yeah. So in that case, we don't go for a custom solutions. Now if there is a use case scenario, which is very much, I would say, personal to the situation inside the organization. So for an example, we deal with 357 agencies. So all of them have different processes, different SOPs. The way that they interact is kind of very different.
[00:28:52] So now creating. Autonomous agent for something which is kind of in line with the operations that we have internally. Yeah, we won't have any ready mid tool for it. So in those cases we would go for a custom tool where we are kind of creating an autonomous agent around the whole tool stack that the development team or the team is using.
[00:29:10] But other than that, if it's a common problem or if I would say if it's a generic problem, don't go for custom tools or you know, spend time into building those tools around your processes. Just go for off the shelf tool. Yeah. What is your perspective on sort of ethical or maybe privacy concerns when dealing with clients?
[00:29:31] You mentioned like uploading emails and maybe conversations, transcripts, things like that. What, how do you think about that from an ethics perspective? I. I think that's the biggest concern, and I guess that is, that is something that we don't have clarity on when it comes to ai, which is about like the data security and the privacy and on those aspects of it.
[00:29:52] So usually what we do is when we are talking about sensitive data or when we are talking about data, which is like historical data, which could mean like a lot to the company, but it has a lot about their clients and their, you know, revenue and all those. Things. Then we always, always make sure that we use agents which are kind of hosted locally on their servers, where we are not taking that data out of the system.
[00:30:15] So in those cases, we use LAMA or Crock, the basic version of Crock, which could be self-hosted on their servers. So in that case, everything is within our, I would say, reach. And then we could have all the firewalls and all the security, whatever we want. So in that case, like localized servers are used for those data.
[00:30:33] But apart from that where you're not really, uh, afraid of, you know, having the data out on the internet or maybe models are using that data to self-train themself. So for anything related to content or anything related to, I would say research or summarization or anything of that sort, we are using models, uh, that is kind of the open AI or the claw or any other models.
[00:30:54] But other than that, if we are doing some very critical work around AI right now, because it's not kind of very much known as in. What security measures are being taken care of, or is the data is being private or not. So in those cases, we just make sure that it's on the self-hosted models. That makes sense.
[00:31:12] It's, uh, it is kind of the wild west, so it's better to be more cautious than Yes. Yes. And so kinda looking ahead into the future, next three to five years, what are some AI trends that are exciting to you that, that you think are gonna have a big impact on the agency space? Recently I came across this model context protocol.
[00:31:32] So basically what it does is when when this whole open AI era came in, there was just elements which we feel that, okay, now we could ask anything, we could get the responses. Then came agents where now the LM models could do some task, like, you know, sending those reports to you or maybe messaging you on Slack or open air operator could do something on the internet and go and do things for you.
[00:31:57] So I guess step-by-step LMS and then agents and now model context protocol. So basically it's, it's something like a generalized server where it is going to have all these applications in build. So now you don't have to stack it up. Right now the whole complexity is when you wanna have agents talk to each other, or when you wanna have tools talk to an agent.
[00:32:19] You have to know exactly how to build that user flow, either using NN or Zapier or make.com, where you can at least build like a four to five connections easily. But when it goes beyond that, it just becomes very much complicated to make sure that everything is going in place or not. So now I feel. To solve that purpose.
[00:32:40] MCP is gonna play a major role where you don't have to stack these tools up or integrations up. It's all going to have it inbuilt within itself. So now let's say Zpr has a hundred different services that it gives. Now you don't have to connect to those a hundred different services. That MCP is going to have all those a hundred services in itself.
[00:32:58] All you have to do is just connect your LLM and an agent to the MCP and the MCP decides based on the prompt that is being sent to the model, which is the service that it should be using. So now you don't have to stack it up, you don't have to do those automations. It's gonna do it for itself. Awesome. So I've got two final questions for you.
[00:33:17] This has been fantastic. I love all the specificity that you've been sharing. I think that's really helpful for me, and I'm sure the listeners. One question is, what would be your sort of parting advice for agency founders who are maybe dabbling in AI but really wanna find a way to help it to increase productivity?
[00:33:36] Let's say 10% per quarter. Any any parting advice that you would have for them? Yeah, assessment is something that is definitely needed by agency owners, so do that assessment with your team. Uh, find out tasks which are redundant, which does not involve any creativity or which has very tight need processes or SOPs around it.
[00:33:55] Those are the ones that could easily, easily be automated, so. You know, do that assessment with the team. Find out what could be automated. Start automating that first, and then go forward and kind of automate task where the team is spending most of their time. If you don't have SOPs, create SOPs around it.
[00:34:12] And I feel data, data is going to play a huge, huge role. Be it you're using LLM or you're using any AI tools or automations. It's gonna be, the output is gonna be as perfect as the data that you might have fed to that agent or the model. Yeah. So now starting to structurize the data is, is really, really important as an agency owner is, have the team obsessed over the data, the structuration of the data.
[00:34:38] Making sure everything has to be consistent. So now as and when you wanna kind of have an agent for you or an automation ready for you, you already have the baseline, uh, data ready, which is like the structured data ready. All you have to go is just build the automation and connect the data point, and then you can start seeing the results.
[00:34:54] So yeah, assessment and streamlining of the data. Beautiful. That's awesome. That's really clear. Thank you. My last question for you is, what's your motivation? What's my motivation? That's, that's a great question. The motivation is if, if I get to see even one agency saying that, okay, I have seen that productivity boost within my team, or I've seen like revenue spike within my team using ai.
[00:35:21] That's kind of, I feel that, okay, now we are helping humans to do what they really want to do. Where they could leverage AI to do all the stuff that they felt that, okay, I really don't like doing it, or maybe that's too lengthy for me, or it's too boring for me. So now you could leverage AI for that and use your, and I would say channelize your creative energy into doing something which is more meaningful or which is more impactful.
[00:35:45] I feel that's, that's the, I would say, motivation factor for me where, you know, helping people understand that, okay, this is how they could build. Automations and then channelize their creative energy into doing something which is far more impactful than what they're doing right now. Beautiful. Well, I think that this has been a really great conversation.
[00:36:05] Again, I love the detail, the specificity around agencies and how they can begin to implement this or improve their implementation. I love things like time to value measuring this, right? If, if you have to measure, if you want it to get better, you have to measure it. So time to value that 10% improvement in, in efficiency and productivity.
[00:36:24] I love all that. Yeah. I know that you guys have a new product for agencies to help them to really embrace ai. Could you share a little bit about what that is and how agencies can get, uh, you know, benefit from that? I. Sure. So we recently launched the fractional AI services where we could get into your business, we could understand the areas that we could automate for you.
[00:36:46] So it's kind of doing that assessment for you where you feel that, okay, if I don't have the time to do with my team right now, how about I have a dedicated AI expert who could come in, do that assessment with my team, find out the areas that could be automated and start doing it for me. So we just recently, we, uh, recently launched that where we could do that assessment, build the automations for you, help you understand which areas could be automated.
[00:37:09] So I guess that's the area, uh, that I feel that agency owners could leverage when they feel that, okay, I'm running out of time, but I really want to integrate AI and I want someone to just come in and do all that AI magic for me. That's awesome. Where, where's a good place for people to learn more and sign up?
[00:37:26] We have, uh, as a page on our E two M solutions. It's like higher fractional AI consultant, uh, that's on the page, uh, on E two M solutions. So you can definitely go and check it out more on that page. Awesome. We'll make sure to put that into the show notes. Kush Brew, thank you so much for coming on the show and for sharing all this wonderful knowledge.
[00:37:45] Thank you, Corey. Thank you for having me. Thanks for tuning in to the Deep Specialization Podcast. If you haven't checked out my bestselling book, anyone, not everyone, you can download the audiobook for free right now by going to anyone, not everyone.com. That's anyone, not everyone.com. And finally, a special thank you to our sponsor, E two M.
[00:38:08] We'll see you in the next episode.
