I'm going to show you how I start every AI consulting project. And knowing this right now, even though you don't have any clients, will give you massive confidence to get those clients because they sense that you are confident in the first steps you're going to take with them. We do it in two parts. One is gathering enough information about the client when they sign in in terms of scraping information about the about the company or about them from the internet.
And point number two is we do progressive building, giving them quick wins from assistants to co-pilots to autopilots to agents. I'm going to show you the quadrant and that's how they'll guarantee massive value and it will keep on working with you forever. That being said, let's jump into the video and get started. So, how I start every $10,000 plus AI consulting project without hiring a dev and without using a big budget that kills profit margins.
Of course, this is more on the client delivery side and not client acquisition, but it helps massively to understand this in client acquisition. When you go through that sales call, when you do the content, when you do your marketing or outreach, people sense or or you radiate that confidence and conviction that I know what I'm going to do as the first step and they're more likely to convert. So, the promise of this video is as follows. We're going to get started on the AI use case as soon as you sign up with a client and deliver actual value immediately through a two-step approach.
One is feeding an LLM or a special project that you have for an AI project, the client context. I'm going to show you exactly how I do that. And two, we're going to do progressive building where we go from low integration, low automation to high automation, high integration, aka AI agents. If you don't know who I am, I my name is De Mahaja. pursued my PhD in aerospace engineering, worked on satellite systems, in particular spacecraft system engineering, sent a couple satellites into outer space and now I integrate AI systems within businesses.
So how I look at AI use case and every single AI use cases through four quadrants. Quadrant number one over here are assistants. Quadrant number two is co-pilots. Quadrant number three is autopilots.
Quadrant number four is agents. And on this side you see everything that's low integration. On the bottom is low automation. Here's high integration and here's high automation.
What does that mean? Assistant, just imagine your chat GBT text that you send it to it a copy paste text and you say write this email for me. You copy that email, the text that it generates for you, you put it in Gmail and then you send it. That's an assistance.
It's very low integration, low automation. Agent is high inte highly integrated within your business operations and systems and it does things highly automated like you don't need to know and tell it what to do. It just figures things out. and think about customer support, AI agency, N8N and all those tools. So this is how I see everything.
Let's look at the business AI use case which is very very popular and I see this selling massively which is a customer support example and this is essentially the first level support in any business so you can make it more efficient. What could we do in terms of an AI use case or how can we use AI to make it more efficient in terms of supporting our customers. So I look at it this way. Same quadrants.
Low integration, high integration, low automation, high automation. We start over here which is assistant is essentially a support rep. That person in responsible that human is using chat GBT to create better answers and faster give answers to the customers. Then we have go one level higher here.
You do a co-pilot where the supports reps get smart AI suggestions based on the tools they are using such as let's say outlook and outlook has context in terms of what that person is getting in terms of emails and suggest certain responses and so on. So that's highly integrated but it's not highly automated because at the end of the day you need to just accept or choose certain things. Autopilots essentially are low integration high automation which is an AI chatbot that answers common questions 24/7. And it's highly automated, yes, it's on your website, yes, but is not that much integrated within the system and the business itself.
And then you have agent, which is an AI chatbot that not only answers these questions, but resolves certain things on the fly, such as refund requests and such as customer issues, such as things that it not has not seen in the past, and with minimal human involvement. So this is the same exact use case I and this is the main point I'm trying to make here which is AI in customer support but different delivery levels by AI and if you think about this how did we do it in terms of offers and delivery methods we have done for you do it yourself done with you this essentially is the AI version of all of this so if you're starting a project right now and you understand this matrix where do you start in particular my suggestion and what I do every single time is I start in low integration, low automation no matter what the AI use case is because the first objective is to prove that this thing actually works for this use case because even though on theory and paper it looks good and it will work, it may not work because of the probabilistic nature of how AI is. So I always suggest we start with human involvement, strong human involvement and strong supervision where we can see under the hood in terms of what's happening. I start with a custom GPT let's say over here and then we start integrating it a bit more then start automating it and using low integration and then jump to agents.
So the heavy or the use case that I've seen or the way people fail at this is they start agents and they go all the way back to assistant then they work their sums up here. So always think think about when you're starting this consulting project the client wants a fast win. They want to see ROI quickly even though it's not the full solution. They want to see something and you just give them an assistant.
You prove that it works. Get some metrics and then you start to move your way up. And don't think of cat chat GPT as you are used to right now in terms of it's just chat GPT. It's just a custom GPT.
It's just a GPT wrapper or whatever you're going to say. But if you package it within a system with the proper evaluation metrics and with the proper documentation, it actually becomes a very powerful tool for testing and simulating in engineering in in the engineering world of whether this use case is going to work with AI or has a high probability of working with AI or not. And once you do that, this is the approach I take. Do not go from agents all the way to assistants to come back to agents because this is not a good place to start.
Even though the most of the value that you're going to get from adding AI within any business operation is over here with minimal human intervention but time and time it has proved by different entities who have started here which went straight for the promise AI such as Clara over here and such as McDonald's with their voice agents such as Air Canada with their chatbot that was agentic that was handling different queries about tickets and so on. They got sued by the way you can check these use cases out. They started here, then they moved all the way over here, and now they're working their way back up. So, always I'd suggest start here, move all the way to agents, and don't do the opposite way because we've seen that does not work.
So, pick the tools that allow the transition from assistant to co-pilot to autopilot and to agents. Meaning that if you choose tools, let's say an agentic tool or solution, some startup, some vendor comes to you and tells you this is an amazing tool. This is going to do exactly the job the AI use case you want. The problem here is that you get locked into this contract.
If you're a big entity, you get locked into this contract for let's say 5 to 10 years and you don't have the flexibility to go back here. Big massive loss and AI does not have the capacity to handle such mistakes in terms of time because it's moving so so fast right now. And if you get locked into a certain thing, you're going to have a big cost lost in terms of money and the ROI and maybe it leaves a bad taste about AI in your mouth. So that's this part.
So if you come to the tools and go b a bit more into this tools essentially to me are very boring. They're and everything I design in terms of AI solutions and when I start an AI consulting project or trying to implement an AI use case, I make it as tool agnostic as we can. Meaning that we don't care about the tools. any tool that comes in the future, we can just create and use the same exact framework with a bit of like let's say 20% 10% needing of tweaks. So the second question you're going to get asked and as you start this project is okay do we make this in-house now if especially from a big entity who has the capacity and the budget to do this should do we make this in-house or should we just pay for it and it depends on multiple factors and this is something you're going to go through depending on the project that you're starting but do it it depends on the control how much control do you want how much how fast do you want to go in terms of building this AI use case do you have the talent inhouse to actually build it out do you have the flexibility to shift and move around or it's a very stringent environment you're going through or how how much of the ownership do you want in terms of security the data and so on the key and the very important key and I repeat it again flexibility is very important in this probabilistic approach we're taking with AI use cases because most of the time AI use cases does you can't bring it to life or see the ROI with a theoretical thinking or recommendation that you have in your mind right now we need to test it out in different ways And I'll give you an example on this just so it sticks.
If you put that the same prompt, the exact same prompt to the same exact model two different times within the date, you're going to get different responses. And that's how AI and LLM work. Essentially, Gen AI works. They give you the first term, then they they suggest the next term and next term and next term.
So if you have a different starting point, everything after that is completely different. And this is why every single AI use case, you have a certain accuracy to it. And you don't lock yourself into the tools because some tools do not make sense even though they make sense. And this is the process that I go through with an AI consulting project saying that okay should we build this in-house or should we go and buy the solution for it?
Should we do a hybrid approach? What exactly do we do? So the first question is I ask is and and this is done within the call. when you're going through the call with them maybe in the first initial calls you can say okay the AI use case you want to now go and start implementing within your entity is it prototype has you proven value of this AI use case in the past or you're just starting out and you just want to see it's your first time doing this if it's their first time you go this prototype route if it has proven value they've tested it multiple times okay let's look at the production route and see how we can take this so let's go into this branch over here assuming it's their first time they have never done this solution within their entity. They're you're going to ask okay what's the strategic value you're going to get or the ROI you're going to get out of this solution is it very high is it very low how do we quantify it and so on so if it's very low maybe buying the solution is fine but if it's very high maybe we look for vendors with the solutions before going and buying it.
So we look at the vendors and saying okay let's see who maybe exists within the marketplace that could sell this solution and if it's acceptable the performance of the solutions you go and buy it. If it's unacceptable then you ask yourself do you have the internal resources to actually build it out with employees infrastructure the hardware and so on. If you don't have the infrastructure and the vendors do not provide the solution let's say those startups and so on for this exact strategic value you want to get from this AI use case that you have never tested out the obvious answer is let's re think about this do you really want to do this or not because we're going the hardest route over here but if it's tested and you have the internal resources and all that maybe you could take some kind of hybrid approach then if we look at this part over here if it's proven value in the past you can have a strate strategic value over here where it's low let's say in terms of strategic value you do a hybrid or buy approach and if it's high you go and make it yourself because it's proven and it has high strategic value to you so you go and build it out so this is a nice framework you can see to just decide whether you want to build it what do you want to do exactly do you want to buy it or whatever it is and let me just go a bit and talk about the tools I'm going to give you my recommendations of what I use when I start with these projects but and by the way I just want to make this clear I have an Excel sheet that has all these tools and I see okay this is for autopilots this is for assistants this is for agents maybe you can create one for your own use cases and so on but essentially I'm just going to give you the tools and I'm going to give you the top three tools I use in every single project so for assistance you have cloud cloud projects you have uh the c which is essentially custom GPs of cloud and then you have cloud code and all of these kind of tools then you have your custom GPTs and you have chat GPT that's an assistant and then you have PO which essentially is all the LLMs that you can choose and you can switch between and chat with as well. So for you that's in terms of low automation, low integration.
So if you go to low automation and high integration, you have the co-pilot. So you can see in Microsoft Outlook over here you have this top thing that comes by summary by copilot essentially telling you what the email is about and how you want to respond. So it's highly integrated not automated that much. In let's say Gemini in the Microsoft tools you can see this over here in Excel. that's in Google Sheets.
You can see here the co-pilot over here that's telling you, okay, do you want to do this? Do you want to do that? And so on. And here's a co-pilot as well of Gemini that I'm showing you, which is above here.
And there are essentially highly integrated systems. And I can tell you based on experience that these systems are not highly valued, let's say, by employees. They'd rather use assistance on their own subscriptions rather than go with co-pilots because there's a trust issue over there. This is practical experience but this is how it is currently.
So then you have the autopilots over here. Autopilots are tools like N8N like chatbase and chat GBT task. Essentially you tell it every single morning schedule this task for me and it's going to do it every single morning for you. So that's high automation and low integration.
Not very much integrated but it's highly automated. And then you have again NA10 with these AI agents and nodes. So NA10 depends on the nodes use Lindy AI for example. If you're on the development side, you have crew AI to build your multi- aent platform and also then you have level over here which is for websites and so on.
So these are a lot of tools and you can use more tools and if I keep on talking about the tools today we're not not going to finish this video but there are a lot of tools and try to be as tool agnostic as you can because every single day a new tool is coming and try to deliver to your clients when you're starting this project in a tool agnostic way because maybe the tool gets out of business tomorrow and so on. So if I had to choose one tool for every single quadrant that I showed you over here, I choose for assistance chat GPT especially with thinking chat GPT5 thinking mode is is really good for planning and doing low automation things and the custom GPT also extracts more information than plot. That's my experience. And then you have if you have to go the co-pilots, I'd use N8N for autopilots.
No, I mean this is this should be shifted by the way. And u yeah so for co autopilots will be NA10 over here co-pilots will be NA10 for autopilots will be N8N and agents will be N8N and this is these are my tool stack essentially that I'll be using for different quadrants and the plan is to use one tool for any use case. Don't just bombard yourself with a bunch of tools. So build it across uh four solution types as well for your AI use case.
Once you choose a tool, take your AI use case to assistance. Just first test it out with custom GPTs, then a co-pilot solution over here with Gemini, let's say autopilot solution with NA10 over here, then agent solution with NA10 as well. And you can just have the mix over here. So one assistant tool chat GBT, one copilot tool, let's say a sheet plugin over here, Gemini, and one autopilot agent tool, which is NA10.
And these alone can take you a very, very, very long way. And this is exactly what I meant at the first part of the video saying that this is progressive building. You start with assistants, then you move to copilots, then autopilots, then agents and you repeat and rinse and repeat over and over again. Now for client context.
So of course this applies only to non-confidential stuff and non-confidential information where I create and when I so just a tip here how I manage everything is I do it in cloud code and I have an cursor and the cursor ID in particular. So that's my documentation assistant and in that ID I drop certain MD files, markdown files which essentially are doc files giving context around the client that I've been talking to. The context includes the proposals, the emails, the transcripts, anything that's not confidential and statement of work and also I use the fire crawl MCP and perplexity API and so on to scrape the information about this certain company. If it's a bigger company, they have more marketing material that you can get info about and just build upon.
And then also I build a hierarchy in terms of the organizational tree structure. Of course, I take as much as I can from the client, but I build a hierarchy in terms of from LinkedIn scraping using a emails and roles and also I look into the Apollo. You can check on the actors on Appify as well. And you get all these information and put it in certain docs in terms of the context about the client.
And this will give you a clear sign. And this just blows their mind when you go on the meetings and on the calls and you talk as if you are part of the team very very quickly. It shows influence you they have. It shows their preferred working style, their risk they willing to take, their motivation, their interest and what they've been publishing and just keeps you up to date.
And I think this is the best way to go with AI projects because if you do an AI project not doing it an AI way I mean it it raises question marks why are not not using AI within your own business and in your own client delivery. So with those those two together what you'll have completed if you're just going through the AI project in terms of system design and or system engineering approach is you've explored the concept you know what the mission is for this project or use case that you have you've know the stakeholder expectations you know the information about the stakeholders over here extensively and then now you can only start with the system requirements in terms of saying okay what is the requirements of the system what is the preliminary design what is the detailed design let's implement it And now let's test the values out and see if this worked. And this is equivalent to the approach that I've shown you in the past which is you have the discovery phase understanding what's the idea is what the concept is and you get to the concept stage and let's build this MVP right now and pilot it and then ship it. That's discovery with the objective being that the rest keeps on coming down through the exact process that you take them through.
Some people think that okay I'm going to think about this when I get my first client and so on. But the problem is that your marketing, the way you do sales calls, the way you present yourself in front of the prospects that potentially can be clients has that smell to it in terms of you're not confident about your first step even. And trust me, this is very very important that you may think it's woo woo or or nobody feels that and so on. I can fake it till I make it.
But it usually does not work because I've seen it time over time with people in different fields that the market senses how confident you are. So understanding your first steps will 100% give you an edge in terms of actually getting the client. So it's funny how everything is connected to each other. So, if you enjoyed this video, like and subscribe and see you in the next