Automation Module Teaser with Laurens Nys

Just going to get going here in a minute. Sounds like so far most of the folks on the call are, are fairly new to, to N8N which is what we're going to be talking about in the, in the kind of the second half of the call today.

And.

So what I'm gonna, I'm actually gonna get going with my quick intro here. So I'm, I'm Jared Waxman, one of the co founders of Go to Market Engineer School and people asking why did, why did you create this? And you know, it's basically I created something that I, I wanted for myself. I wanted to learn all these tools and I wanted to learn from really smart folks that are doing this every day and it didn't exist. I thought, okay, well let's, you know, I've got enough connections and so on so forth. I think I can go put something together here. And once I did, I started hearing from people all over saying the same thing. People in growth and sales and a rev op saying that they know that they could use AI and automation to be more efficient, but they don't. They're just at the beginning of it and that they feel like they're getting left behind. And not a surprise. These tools are moving so quickly. The things you could do now you couldn't do three years ago. So, so anyway, we have this course. It's, it's really geared to get you to go from zero to, you know, sort of 0.5, like be comfortable enough to continue your learning journey. And it's like a taste test of different workflows, different tools and you'll learn from, you know, different experts. And so you know, whether we're talking about CRM hygiene or scaling personalized outreach or see staying responsive to buyer signals, refreshing content, moving from SEO to aeo, these are all things that we'll touch upon in the course. I'm going to go through the rest of these slides very quickly so that Lorenz can get going, doing my little overview. Then we're going to get into the main session, then there'll be a little Q and A. The school itself, you know, like I said, we cross tool expertise is the focus here. Hands on, it's very practical. Students in the course get free credits for a couple months so they don't have to worry about their credit card getting blown up with tokens being used. And then we've got the slack community. Our cohort last time really enjoyed it. Nine out of ten recommended it to friends and about 30% of them actually get new roles within two months after the job, sorry, the course ended a little weird build here. But the course schedule basically have five main modules. Research, enrichment, scoring with Yash from Clay and then messaging, personalization and setting up a brain for your PMM with Zach from Octave. And then if you are looking at a lot of internal data sources to build your workflows and your agents, Tariq and Max are going to show you some ways to do that with Cargo. And then we're going to get into content and moving from SEO to enter engine optimization. Other things you can do with Air Ops with Steven and Andrea. And then Lawrence here is going to show us about sort of no Limits with Nadan and all the things you can do. Each week we've got a lecture which is two hours usually on Wednesdays from 8 to 10 Pacific. We'll record that and then we've got over the next week a bunch of different live help sessions, debugging sessions. You could pop into them, bring the async exercises, homework exercises that we've given you in the, in the, in the Wednesday lecture and help get on. You know, they're going to help you get unstuck, help you get through it so you can share your screen. Small groups and, and that's, that's how we want to get you over the learning curve. Also, just to know, we've got a substack with lots of newsletters. We got a newsletter coming out every week with sort of a brief of what's going on in the industry. And we've got podcast episodes that we're dropping every week. And Lorenz here was kind enough to, to record an awesome podcast with us and then we also have a go to market engineer jobs board that you can access.

And.

That'S, that's my intro. So this is a little about the course here. The course is, is sold out for this next, this next cohort starting Monday. But we hope to see you in a future cohort if you're not already in this, in this coming one. And that's it. Now I'm going to stop my screen share and hand it over.

We'll just go ahead. Yeah, I mean, quick intro for the, you know, in case you don't know me yet. So, Lawrence, I'm a GTM engineer. I've built my own GTM engineering studio, GTM Sigma. I've been doing some sort of GTM engineering for the past two years. Initially started as software sales and then always the sort of tech aspect always spoke to me. And yeah, now I'm just a GTM engineer helping software companies build automations use AI across their go to market. And let me see to the next slide. So yeah, just quickly today as well as like Tarik in the last webinar, I'm going to give you my opinionated view on what I think a GTM engineer is, how I think GTM engineers should think. I'm going to move through these as quickly as possible as well. And then I'll end off with why I think NAN is a decent or even a great tool to use as a GTM engineer and then show you some real NN flows, both flows. We'll dive into into the course and just maybe give you some ideas to try yourself. So first things first, what is a GTM engineer? So I like kind of dissecting words when looking at what does something mean? GTM engineering to me is like obviously it's two words, GTM and engineering. I'm assuming most of you know what go to market is, but to me the very first principle definition of it is anything you do to get, keep and make your customers worth more, that's go to market. So a real side note here is if you're just doing outbound, you might be an algorithm engineer, but in my opinion that's not what a go to market engineer is. You should be looking at GoToMarket as a system across all of those. So that's marketing, that's sales, that's customer support. Again, like literally anything to get, keep and make them worth more. And engineering means building systems. So that's not list building. It's like building systems that can live on their own. And where there's an input, there's an output, there's a process. This is kind of my definition of a go to market engineer. Anyone? Engineering systems across the full go to market, then the next thing kind of okay, some of these were already open. Those are kind of like to me, sort of the first principle and the core ideas I think you should kind of know and internalize as a go to market engineer. I think the first thing, which might be an obvious one is go to market is a system and we're just going to dive into it a little because I'm assuming most of you just want to see Nadan. But to me, go to market as a system basically just means just like any system, there's an input, there's a process, and each process has a certain time it takes to fulfill that process. And there's a certain percentage of conversion or in system things, it's often called waste There's a potential to increase both time and increase conversion. And there's certain output now in sales or in go to market. That's pretty obvious that your input is like your prospects, your web visits, et cetera. And the output is just revenue, it's ARR or retention, et cetera, increase. And then ideally there's a feedback group. That's not always the case in go to market it often is. Like for example, ICP is a decent feedback group to where if you look at your go to market you start closing some leads as soon as you have a lot of closed companies. That in turn teaches you to understand your ICP better, to define your ICP better, making the inputs or the I run process, for example, more efficient and more effective. So that's like core idea number one. The second sort of idea is okay, now we have go to market as a system. If you look at it as a system, how do we actually go in and make a system better? And I think theory of constraints is to me one of those core ideas or lenses you should be using when looking at go to market. So the core idea or the core thesis is like each system, there's steps in a system and each circle here represents a step and then how far it's filled represents the capacity for that step. For example, there's inputs that's inbound leads coming in. And the first kind of step in your sales funnel is a meeting with an SDR. And the SDR can take 5, 10, a certain amount of meetings and and then goes off to the next step. If this is at half of capacity, this will never be like the limiting factor. Like if there's 10 leads coming in, but your SDR can take 20 meetings. The amount or figuring out automation or AI in that first step of making the SDR more efficient better isn't really going to move any metric because that's not a constraint. But there might be a constraint in like for example, you need to make proposals and you can only make two proposals a week. If there's more than two companies getting to that stage each week, then that's the constraint and that's the thing to solve for. I'm not sure if it's fully commonalized, but it's one of those things that it's very logical if you think about it, but it's not often practiced. But I think this is again one of the core ideas. So that's the second one and then the third one is a more simple one or a more straightforward one. It's like more better new. It's like when you found the constraints, you kind of run through, okay, how do we solve for this? And a problem or something a lot of people tend to do very fast or too fast is they just get excited with AI and all the possibilities and they just immediately go to what's the most advanced thing we can do here? What's some new process we can build or some new thing we can do with AI that we couldn't do before, which isn't always the most useful or best way to solve for a constraint. So the framework I use is more better new more basically just means, okay, so there's 10 right now. How can we double it? What's the process right now? And how can we do without changing anything? How can we do more of what we're already doing? Then the second one is better. Which basically means, for example, the rep is doing research every meeting which caps the amount of meetings that can be held every day. How can we make the prep or the conversion better? Maybe only has 30 minutes to do research. Is there certain things like pieces of information that we can add to the research that might be useful but just aren't happening because he only has 30 minutes, for example, reading annual reports is a great example here. No rep has time to read an annual report, but there's usually very useful information in those. AI can read it decently fast and fast enough to increase conversion. And then the last thing obviously is new to where I think you should touch new. Only if you know for sure you can't do more and you can't make it better. Only then should you consider like, oh, this is still a constraint, or like customer cost still doesn't make sense, we need to change this. Only then I would go out and actually do something completely different and it's already been done. The next one, this is more like the meat of it. NADN as a GTM engineering tool. I think in essence, if you look at because there's so many tools you can use, I mean, GTM engineering also just encompasses a lot. Like building systems means like every AI tool, every automation tool is all up for grasps. I like NNN personally because I think you can kind of see every tool as sort of a layer of abstraction of code, right? Like in essence, if you would go out and build fully custom things, you would just be writing code. But that's very hard. There's a big learning curve there. Maybe cloud code catches up one day, but right now there's still a decent learning curve there. Nadan to me sits just high enough above code to allow for a decent learning curve and make it still pretty straightforward to build out automations. But it's also close enough to it to where it allows for a lot of capabilities because the higher you go in abstractions you kind of lose a lot of capabilities. With code you can do anything you can imagine and then the higher up in abstraction you go, you can do less of what code can basically. Another thing I think N then really nailed is their AI capabilities. They just realized there was something N then would be great for and it just shipped like crazy basically. And especially like the agentic part of it because a lot of AI isn't really agentic. And what I mean with agentic is an AI kind of, and I'm going to show you like in a workflow later on as well is where the AI can like reasoning itself, have multiple steps, can call a tool, can look for data and can kind of loop through steps and really think through steps. Like for example like ChatGPT, it's a chat interface. It's like you send a prompt and it responds. When you build out agentic stuff you can send out a prompt and then maybe it goes off, takes the website based on what it finds on the website, it goes and scrapes LinkedIn based on what it finds on LinkedIn it goes back to the website but with a different query, et cetera till it reaches its goal, which I think NIL did really well. And then context engineering is also maybe dive deeper into that later. But that's like a big part in building production ready AI systems. But any then also allows to connect it to a database, to a rack, to a knowledge graph, et cetera and it's just extremely cheap as well. It's like 20 bucks a month which some automation tools are just as cheap but that's decently rare, especially in the tools targeted towards GTM engineering. Most of them are priced a little higher then this one is kind of context as well. So when you build out like production ready stuff you'll Rarely just use N8N usually you'll also use some sort of a data storage system like a database, a data warehouse, et cetera. And this usually is something that especially if you talk to GCM engineers that are starting out, it seems more complicated than it is. In essence Google Sheets can act as a data warehouse and then postgres which is like the legit database or whatever you want to call it. If you look at it it's just the Google Sheets and you just use it Using a language that's SQL. But if you have AI, you rarely write SQL yourself. But it's just as a sort of a side note, if you build out real systems, you'll usually build out your data storage, which for most companies is CRM. And you'll build out the database with the data and eat it on top. And then that's connected to some sort of an interaction layer. That interaction layer basically being where the user actually lives. Like for example, the sales rep lives in the CRM. So the CRM sends notification if something happens in their target market, sends, adds a note for company research, adds the context automatically, etc. The core idea really is that the sales rep, the marketing, customer support, never get into any. Then they never go to the database, they just stay where they are. And the tools that they're currently using just get way better and just add capabilities. Basically, this is the theory part. I've basically just opened a few workflows I've built in the past to just kind of run through N8N. Especially like the first time you open it, I can imagine it's like you just open a canvas, it's fully blank, you're like, what the heck do I do? And we'll obviously run through the courses to where do you start? What does this all mean? I think the really basic basics of it is N8N is kind of just you map out workflows and then you have all these individual, they call it nodes and they just all perform one action and then you just give it to the next one. And this is like. So I'm just going to show you what this one is. This is a deep research one. I tend to work with companies with usually some sort of vertical aspect in the solution they sell. And usually an ACV or a deal value that's decently high to where it makes sense to go really deep into research and have a lot of information pre call. So this is like one of those agents. And then each of these is actually a sub workflow. So just give you the TLDR on what this does. This gets triggered by some sort of an action in their CRM. They just press a button, runs through, or there's a meeting already planned. It looks for the company, it checks if the company is there. If not, it creates the company and runs it through again. And then these all just kind of enrich new information from the company. So for example, this is to get product information and I'll open like the full work focus. This is kind of like a matrix, like A workflow in a workflow in a workflow. So this one for example just goes out and then these are like DI nodes and these are the tools the AI has access to. So for example, this is just an AI model, which in this case is just GPT 401. This is like a Think tool, which is like an NAN native one. But it just allows the AI to, when it goes through tasks to kind of have. It's very similar to if you're doing research, you just write stuff down while you're doing research. That's basically what AI is doing with Think Tool. And then these are just tools like Firecrawl is Scrape for Web. Firecall Map is just to get the sitemap to know what are all the sort of links on this website. And this one just goes through, maps out all the products a company has. Then for each, then the first thing it does is, okay, these are all the products, which is like their flagship product. So we know we're like targeting or getting information on the product they care the most about as well. How mature is their product? Can't really show the prompt here because it's some sort of, it's not secret, but it's sort of IP or whatever. But it kind of maps out the product they currently have and then compares it to like what's the market average? Like what are the features here? How advanced are these features? And you can create for your full total addressable market. Let's say you're selling into software and you're enabling some sort of a new capability that makes them better compared to competitors, which is the case for the company this was built for. What this does is then gives you an overview of, hey, this product or this company is targeting this type of other companies and is targeting in this geographical area. And, and this is the features they have, and this is how mature that makes them. And this is how it compares to similar companies targeting the same customer. Because then like when you do outbound or whatever, you have like full context on how to target them, right? Because for example, if they're targeting aviation or whatever, and you notice they lack two specific features that all other aviation companies have, you can bet they're probably losing deals based on not having those features. So that's something you can use in your outbound, something you can use pre call to sound smart when you have the call, et cetera. And then the gap analysis is basically just kind of synthesizing everything. Usually you. And this is like some sort of a AI thing, but the more context you give into your AI. And maybe if you use ChatGPT or cloth or whatever, you notice that yourself as well. But the longer a conversation goes on, the more tokens are basically in the context window, the worse the output starts to get. So something that's very easy and smart to do is you make sub agents, either in sequence or beneath each other, that separate certain tasks and then carry over context. So you start with mapping the products and then this one just sends out a message to this one. It's like, hey, these are the products I found which one is the most relevant? And then this one takes the relevant one, gives it to the next, and it's like, hey, this is the product. We need research it instead of building it all into one agent to where context will be way too high and then just not effective, basically. So that's one workflow and there's like, with any 10 possibilities are sort of endless. Which makes it more exciting and scary in the sense of like, you have just like a canvas, you can do everything. So what do you do when you can do everything? That's why there's like some sort of theory I started with. This one, in essence is also a deep research one. It works in a similar way. But the main change here, and so like the TLDR here is the same. You saw the trigger. It gets a company and it builds out a full picture of their supply chain. Because this, for a company that sells two companies with supply chains, some sort of supply chain software, and it uses like sub agents. So the cool thing with Nadan, with sub agents, you can basically have this agent talk to this agent, so you can compare it to like when you work together with a colleague. Like, this one just has the thought of like, hey, build out a map of their supply chain and then just sends a question to the one below it. And I can actually like show you the question as well, because we've run it previously. But it just sends the question of like, hey, how many employees does this company have? And then it gets an answer from the sub agent and it kind of iterates through these sub agents, these conversations, until it has all the information it needs and then it gives it to the next agent that then goes off and do his thing. But the cool thing really is when you start building out these systems and enabling these main agents to call to sub agents. For example, when you talk about research agents, there's really anything that's findable online you can get in there. There's really no constraint on what you can Build what the agent can do. Because right now it's like research is passive in that the agent just gets information, but you can add other stuff to this as well. You can get the agent to send an email in here. You can also get it to call with some voice models, but you can really get the agent to do anything you want. The only real requirement is that the tool that you want the agent to interact with needs to have some sort of an API, which 90% of tools have. Like most tools have an API. Like Gmail has an API. You can get the agent to read all your emails, reply to them for you, like kind of build out some sort of assistant, etc. But the possibilities really are endless. And I'll just show this one as well to just give another use case. So especially if you sell into enterprise, one of the main pains in your ass basically is figuring out who the buying committee is and finding those specific contacts. So this is an agent again with multiple sub agents, and this is even a sub agent that then in turns also has a sub agent that you just basically give a company. And it just keeps hammering away at finding the specific buyers you identified. Using Google, using LinkedIn, APIs, using Tavli, which is a tool to basically query the web in a different way than Google does, and then exa, which is also some sort of a Google alternative. But it just keeps hammering away at it keeps asking the sub agent, hey, looking for this? Hey, found this one. But it's not really what we're looking for. Is there anyone like this in the company, et cetera. And it just keeps running until it finds it. Obviously in production you'll have some limits on like if you try it a thousand times, you're probably not going to find it. So you have to stop. But it's just to show you another use case as well. The use cases in theory are really endless. I'm not sure if I can show you that one, but another one I'm building as well right now is a customer development one where you have all your meetings, all the transcripts get used by an AI agent. The AI agent kind of analyzes the full call transcript and builds out hypothesis based on the call. So for example was a call with company in industry, certain size, geography, and we talked to Persona X, Persona Y. And then it creates a hypothesis on what's true for this Persona. So for example, they currently care about this, the CFO at healthcare companies tend to talk about this, etc. And you save those hypotheses and then every New call, you reject, validate or build new hypothesis. And you just build out an AI system that just builds a crazy knowledge base on who your customer is and really develops a customer profile that you can use for training. New reps you can use for sending emails. You can connect that agent later to other agents for a copywriting agent, for a meeting, agents, whatever. Right now it's like marketing teams can use it as well when they're writing advertising. They can just interact with it and be like, hey, targeting these companies, what do they care about? Et cetera. But that's also an interesting way to use and build these agents. So yeah, I think that's it. And we'll dive into all of this in the courses. We'll start very simple. Obviously we're not going to start off with some crazy NN flows, but we'll start with a simple one and then we'll just build on top of it.

Thank you. Any question, what kind of questions come up when you're seeing these workflows? This is your chance.

Okay, wait, so I'm looking at the chat. So how long does it take to build such a workflow? Do you mean like take as in building or take as in learning the skills to be able to build? I'll just answer both, by the way. Yeah, so learning really depends, to be honest. I think Claude and ChatGPT are decently trained on the documentation of N8N. So you can get some prototypes just talking to them and then building out kind of depends on the workflow. Something like this I think took, I don't know, two days, three days building it, something along those lines. The hypothesis customer development one takes a little longer because there's a lot of stuff you just need to reason through of. Okay, what's the framework to use when doing customer development? How do we define Personas, et cetera. But yeah, I would say anything from a few hours to a week or whatever. So many landflows obviously take long, but you can get basic stuff working pretty fast. Like for example, if you would build just a CRM enrichment one, you can get that working like 15, 30 minutes to where? Just like every company that gets added to the CRM is enriched using LinkedIn data. That's like super easy workflow. Once you know anything takes 10 minutes and it's done. The other questions.

Yeah, Fahad's asking, how do you compare to other tools? When do you use specialized or purpose built tools versus N8N?

So depends what you're optimizing for. Right. I think Nadan is Really optimizing to build a systems layer beneath your go to market. So what I usually do is I usually prototype either using clay or if it's simple enough just using N8N because that will usually be faster to get from 0 to 1. And then as soon as we're like, okay, this solves our constraint, this is like we identified this metric as being the constraint. It's moving in a meaningful way. I want to build out a system so I can set it up once and then forget about it and never touch it again and it just works. That's usually or currently I'm always doing that in N8N. I'm not going to any other system to have the or any other tool to have the full system running 24. 7. There probably are other tools as well. I think it's just part preference and part just I know N8N by this point pretty well to where that's for me fast and useful to do. But yeah. So short answer, when a prototype I would do in just the tool you know best and you know, can get the job done. And then as soon as you go from prototype to really production ready, I would probably move it to nan.

What did you struggle with when you were starting to learn Nadan? What. What do you think are the most complicated bits for a very beginner to pick up?

I think the good part of me struggling with NET N is I've been using NAN for a while and I was using it before people were posting extreme workflows on nnn, so I had no reference for what advanced was in nadn. So for me, the first time I was able to just enrich LinkedIn data in such a workflow, I was crazy excited. I was like, oh shit, this is amazing. I can just do this. And I just started building. I think if you start off with knowing what's possible, you might compare yourself to the extremes or the advanced level. I had the benefit of not knowing the advanced level, so I was just excited the whole way getting there. I think initially most of it makes sense. I think initially maybe learning some sort of database basics helps a lot, but it's very boring to do. But it does help a lot to kind of learn the basics of database design. And you can just ask ChatGPT and probably I also learned when ChatGPT, I think it was GPT3, so they still sucked basically. So I had to learn a lot by just doing manual stuff. So HTTP nodes also are like. They just like if you open them, it's the JSON and Stuff seems like daunting, right? So stuff like that is like, you know, a little scarier initially, but you get used to it, like decently quick.

And you mentioned in the intro slides, a data source is typically part of your architecture, you know, so. And we see postgres. So is that most of your flows, you're. You're storing, you're scraping, you're, you're analyzing, you're using AI, then you're saving it back to postgres database and using it in another part of the workflow. Is that right?

Yeah, exactly. Yeah, yeah. I've even like a doc I could potentially share in like the advanced track in the course. But once it's like production ready, it's really built out to where you have some sort of dispatcher. It creates a task in a postgres table and then all the other tools just kind of pull the dispatcher system constantly and then they do their task and give it to the next one, et cetera. But yeah, most of it uses a data source or a trigger, and then there's a bunch of other APIs and tools that they use across it. It's kind of similar to Clay, where in Clay you have all this credit system that allows you to very easily try a bunch of tools. Any then kind of exists as an orchestrator and you'll use a bunch of different tools that exist out there via API or via. Because any then also has a library of tons of tools that you can go through and just use natively. And these are basically preset up, so you don't have to go in and build the HTTP ones. But yeah, and then postgres is very useful to have the data kind of. I always visualize this. You have the data layer, you have the systems or intelligence layer, and you have the layer that people interact with.

Stupid. But let's be real, they can't do.

Much on their own. MCP acts like a central.

All right, surely people have more questions. Call them out.

Just a easy question for Lawrence. Lawrence, Sorry, can you hear me?

Yeah, I can hear you. Okay.

Okay, sure. So just, just to take a step back as a GTM engineer, right. How often do you see like clients or people coming up, custom requests that, hey, this is great with these tools, but can you also do this with N8N? And you kind of agree for that? That's first. And second, do they also come up with things like, hey, clay is getting expensive for us. Can you build. Build this custom setup in N8N or something like that? Do you see this Kind of thing.

Yeah. So once like client facing, I only use Nadan. So as in, I build it on their infrastructure and they have their own Nadan account and we push it to their account. This, like, obviously it depends on. Like, if you work for a company, you'll probably do the same. Like, it's on their account. If you're an agency, you like might decide not to do it. But I'm like, I don't gatekeep the system. I'm a GTM engineer, I'm not an agency. So I sell systems, I don't sell leads. So Clay is like just run on my end and I pay for it from GTM6+sites. And then if Clay is proven to be successful, we build it in n immediately and put it on their infrastructure. So I rarely have those conversations previously. Like before the NNN craze, I did use Clay. I never really had the. Oh, Cray is getting expensive. I think there's a few things to just combat that as well. I think one, if you're a GTM engineer, you should have some basic dashboard. So you should kind of know what's the input, what's the output. So you should kind of know, like, we're paying X amount to get X amount. So as long as your customer acquisition cost to your lifetime value, the value you're getting from your customer makes sense. I don't think it really matters that K is more expensive. If you're paying a 400amonth subscription to get a 10x, like maybe a 10x is too much, but a 2x in the amount of leads you close every month, that's very easy to just be like, yeah, but we're paying 400 to get 2k, why would we stop paying 400 to get more? So I think it's about just making sure you measure the data. Because it might like if you're paying 400 but you're not getting any return for it, then it is too expensive. It's only not expensive if there's actually a return there. And I think as a GTM engineer, unless the company has some sort of revenue operations as well, are really into the customer acquisition LTV stuff, I would be looking at customer acquisition cost and just being like, hey, but the data shows this is the return. And I just kind of end of discussion because there's a return.

Thank you, Lawrence. That makes sense.

Perfect. Yeah, it's still cheaper than hiring 10 SDRs, for sure. So how much of these are common across your clients versus client specific? I think the abstraction of it is common. Come on, it's Halloween. Depends on what you mean by these, of course. Boys, I just bought all the eggs.

Hold on, we got some background noise there.

Yes. Okay. Yeah. So at the end I was like, how much of these are common across your client versus client specific? I'm assuming these refers to workflows by committee analysis. Yeah. So the abstraction of it is common. The implementation of it is always a little bit custom. For example, buying comedy kind of depends on the solution you're trying to sell. So you'll always be adjusting to the specifics of it. But for me also we identified our ICP as companies selling into enterprises selling a complex or deep problem they solve, like selling complex software products. So across those you'll do buying committee analysis. If you're going more towards a product LED company or like, for example, you're selling some sort of a social media tool or a content tool, you'll probably don't need to know the buying committee. If you're selling to like SME, you'll probably also don't need to know. So I'm assuming because you asked the question, you probably have more clients. I would probably focus on one type of clients if you want to be able to be as efficient as possible and reuse as much efficient as possible. And also just keep learnings across those because the way they operate is just very different as well. Okay.

Fahad has another question for using databases in workflow is if the company uses Microsoft, does the same work with Excel or not? If yes, is there a separate integration required?

Yeah, so I usually use postgres, which is like a legit database tool which is completely separate from either Google or Microsoft. They both have their postgres solutions. I use Neon, which is like a third party. It's fast, it's cheap. I would assume Excel is in here as well. Initially you might use something like that. I think one level up as well might just be using Airtable to just like Basic would then be Google Sheets or I think you can get Google Sheets for free as well by the way, that you don't have to pay for it, but maybe your enterprise and policies. Microsoft Excel would probably work fine as well. I think the next level is using something like airtable and I'm not sure what the limitations of airtable are exactly. I know with Google Sheets you can get about a million rows and then it just stops adding more rows. Postgres, you just never have too much rows. You can query it, et cetera. And there's also just a speed component because both Excel, both Google, they have like, they call it a rate limit. Like you can only get like 10 rows a second or a minute, for example. With postgres, that limit is usually way, way, way higher to where you get, get like a thousand rows every second or whatever. So airtable might be like a good middle road in there. But yeah, Excel works fine. Google sheets work fine. Most of these just all work.

You mentioned in some of your workflows you've got two agents talking to one another and one is sort of determining if the quality of the input from the other one is sufficient or not, or if it found a good answer. Can you just say more about that? How does it, how does it do that? How are you doing evals within Nadan?

Yeah, yeah, so evals is something else. But where was the. Okay, I think this one is the easiest one to show that. But in essence what it really just is in nn, it allows you to. So AI agents, they can call tools. So for example, a tool is to scrape something to use an API call within N8N. You can set up another agent as a tool call. So then it basically just calls the agent. So you basically instruct it to the main agent. You say like, hey, if you need data from LinkedIn, talk in this way to sub agent. Like say I'm looking for Persona, here's the LinkedIn URL or whatever. And then the sub agent just gets that as like the message, like the user message. And then you have the system prompts and maybe to backtrack a little bit, like AI kind of has multiple levels of prompting. In theory you have the developer prompt, the system prompt and the user prompt. The user prompt is like if you're interacting with ChatGPT, you're sending messages, that's the user prompt. The system prompt is kind of. OpenAI has made some rules like if you ask some really weird fucked up questions with ChatGPT, it will tell you can't answer. That's because OpenAI put in a system prompt telling it that, hey, if user asks some weird fucked up thing, don't answer and abort mission or whatever. So that's the system prompt. So for these sub agents, you basically have the system prompt and then you have the user message or just the message that's being sent from the main agent. And then the main agent just asks the LinkedIn sub agent. In this case, this question is like search for senior engineer, something, something. And then this one just replies with hey, this is what I found, et cetera. And then the main Agent gets that information back. And then just in this case, it thinks Persona found core algorithm leader, etc. And it goes back to the LinkedIn sub agent and it's like, hey, find me the CTO now. And again in this case couldn't find it. So replies, he couldn't find it. And if we make smaller here, it's right, it's right again for the CEO. It thinks again of like, hey, these are the ones I found, goes back foreign, et cetera. So there's just like a back and forward happening until it either finds it or you can kind of set up the max iterations or it just kind of decides that you kind of reach the maximal amount of attempts here, which is like this at 100, but you kind of set it at something reasonable to where the cost, like AI isn't that expensive, but the cost of the AI just doesn't skyrocket all of a sudden for no results. And usually 100 is a lot, but like, usually 25, like depending on the system is probably enough to where you probably won't get it on the 26th one. If you're like needing 25 interactions here.

Okay, we got another question. Let's see. How do you identify which tools or platforms to experiment with beyond your usual stack?

To be honest, those haven't changed a lot. I mean, the main shift for me was like about a year ago, this shift from clay to NADN to where NADN was the production one outside of K nad. I'm not using a lot. Right. Like you'll always. I'm a big proponent of first principle thinking and I think there's just a few things you'll always need. You'll always need a place to store data. So that's your database. You'll always need some sort of a tool that does something with the data that can be cray, can be N8N can be something else. And then you'll also always need some sort of a source of data. Like you can have the best storage and the best automation, but if nothing is coming in, then nothing is happening. Right. So I tend to reason through first principle anything kind of. It's a layer of abstraction that allows for a lot of capability. There hasn't been a lot to where I'm like, oh, I can't do this in nnn. I'm trying to think, but I don't think that's ever been the case to where I was like, oh shit. Like NNN just literally isn't capable of doing this. So I think it's the main trade off is just like the reason we use K is because it's faster compared to nnn. Right. Then it's just like, what tool do you know? What tool do you like? And which is like, which is the fastest way to get from A to B? And I just use that tool. I think, especially in GTM engineering, especially if you look at LinkedIn a lot, I think a lot of people are a little too obsessed with getting the new flashy tool and not obsessed enough with looking at GTM as a system, identifying the constraints and then just solving the constraints. Because if you go really deep in trying to, I don't know, you identify a constraint. If you just go really deep in. I just need to solve these fucking constraints. How can I make it faster? The tool will kind of reveal itself in the process of trying to solve for it. I hope that makes sense. It's almost like philosophical.

Yeah. All right, let's say take one more question before we wrap up.

I have one more, but it's a quick repeat. Yeah, sorry, Lawrence, you answered my question and I'm in an office. Someone came and interrupted me. So I didn't hear what you said. But briefly, were the deep research agents in N8N did they go deeper than just like a deep research prompt in clay gen, or is there still a level of can't find certain information?

Yeah, so I haven't used clay agent in a while, so I'm not sure how advanced it is right now. But with N8N, I'm pretty certain or confident enough that if it's. If the data is online, you should be able to get it using the agent. And if you're not getting it, that's not an issue. That's like, that's a skill issue, basically. But if it's online, you should be able to get it to where, like, if you don't get it, it's not online, it can't be found. Yeah. Awesome.

Thank you.

Great. Any other questions or are we. Are we true.

I think it's. I think we're all good. You know, you could shoot me a note or Lawrence, a note. If you have any more questions after this, I will be posting a recording of this on our YouTube channel for go to Market Engineer School. And if you are signed up for this next cohort, hope to see you. Hope to get to know you as. As we learn together and, and otherwise hope to see you in. In a future session or future course. So thank you. This is eye opening, Lorenz, and I really appreciate it and I know everybody learned one or two things at a minimum. So thank you all and have a great rest of your day.

Awesome. Yeah. Bye. Bye. Take care.

Bye. Bye.

Thanks.