So What? Getting Started With n8n AI Automation

So What? Marketing Analytics and Insights Live

airs every Thursday at 1 pm EST.

You can watch on YouTube Live. Be sure to subscribe and follow so you never miss an episode!

In this episode of So What?, you’ll learn about getting started with AI automation using the no-code platform, N8N. You’ll discover how to strategically approach AI automation by starting with your purpose and understanding the people, processes, and platforms involved before diving into performance measurement. You’ll explore the importance of clear documentation and well-defined processes, to set you up for AI automation success, and the powerful benefits of using N8N. Finally, you’ll see practical use cases for N8N and gain actionable insights to improve your marketing workflows.

Watch the video here:

Can’t see anything? Watch it on YouTube here.

In this episode you’ll learn:

• Why n8n might be a good AI automation solution
• What n8n can and can’t do
• How to get started with a practical, simple n8n AI automation use case

Transcript:

What follows is an AI-generated transcript. The transcript may contain errors and is not a substitute for listening to the episode.

Katie Robbert – 00:33
Well, hey everyone! Happy Thursday! Welcome to So What? The Marketing Analytics and Insights live show. I’m Katie, joined by Chris and John. How’s it going, guys?

John Wall – 00:41
Hey.

Christopher Penn – 00:41
Hello.

Katie Robbert – 00:42
Hello! Well, so close.

John Wall – 00:46
So close.

Katie Robbert – 00:48
This week we are talking about getting started with n8n AI automation—which is not easy to say. You have to be very articulate when you say that. n8n is—think about it. When you look at it, when I look at it, the first thing I thought was Microsoft Visio, where you could sort of drag and drop the icons to represent different parts of the process. The difference—a big difference—is that when you drag and drop the icons in n8n, something actually happens, whereas in Microsoft Visio, it’s just a representation. So today we’re going to get into what that actually means and how you marketers, analysts, operations, data scientists can start to think about integrating the n8n platform into your process. So Chris, where should we start?

Christopher Penn – 01:42
Well, I would love to run in and start talking about the technology, but Katie won’t let me do that because we have this thing called the 5P framework—purpose, people, process, platform, performance. And starting with the platform—starting with the technology—is universally a bad way to do it. It’s kind of like saying, “Hey, let’s do blenders!” And then you’re like, “Oh, yeah, we’re a steak restaurant.” Maybe not. So Katie, with an automation tool that does workflow automation, as a process expert, how would you start to put this in the context of the 5P framework?

Katie Robbert – 02:23
The first thing I would think about is my purpose—why am I doing this? Why am I adding extra steps to an already existing process? What is the purpose of having more automation in a process? And so, is it—I need to make it even faster. I need to make it less human intervention; I need to scale it, whatever. But first, you want to decide what is the purpose. Because basically, you’re adding another layer of software that has a lot of steps and could break at any given step. So you need to be clear about why you’re doing it in the first place. Once we start showing it, it looks pretty cool and I want to add it to everything. But it’s a shiny object if you treat it that way.

Katie Robbert – 03:10
So having a very clear purpose as to what you’re trying to accomplish and what specific process you are examining is going to set you off on the right foot.

Christopher Penn – 03:21
Exactly. One of the problems people have with generative AI is that generative AI is very much like an engine—and an engine is important. If your car has no engine, you’re not going anywhere. However, an engine by itself is also not particularly helpful. No one rides down the road on an engine. You ride down the road on a car. The problem that almost everyone has once they get past basic use of ChatGPT is like, “I want it to do this; I want it to do that.” Can it read my email? Can’t do this; can’t do that. And almost universally, the answer is no. We did talk about Model Context protocol on the Trust Insights podcast not too long ago. However, that has an extremely high barrier to entry.

Christopher Penn – 04:04
From a technical perspective, tools like n8n and Zapier and Make and the many others in the space effectively let you build the rest of the car around the AI engine. We named this show “Gangster with n8n AI Automation,” but it is in fact a general automation tool. You can use it entirely without AI, just to do basic automations. Again, very similar tools like Zapier with AI. However, generative AI, in particular, is all about building the rest of the car—how can I build the rest of the car and connect it to my data sources? How can I take data into AI? How can I take data out of AI and do something else with it? There are two flavors; there are two versions of n8n. There is the cloud-hosted version, which is hosted by a third-party company, n8n.

Christopher Penn – 04:55
And then, because it’s open-source software, there’s a version you can download and run on your computer. For today’s discussion, choosing between the two is going to depend on things like privacy, security, and cost. The local hosted version is free. You can run it on your laptop. It has very low overhead. The process of installing it is not fun, but it is free, and by default, it is as secure as your equipment is. So if you are doing, say, processing healthcare records, you would absolutely want a locally hosted version that is governed by your IT and HIPAA and all that stuff so that you don’t worry about your data leaking out of your network. However, that requires technical expertise.

Christopher Penn – 05:46
The cloud-hosted version is where you pay money to a company, and it’s their problem to administer all the nuts and bolts of plumbing, and you just use the tool and connect it via the various APIs. So those would be the two versions of n8n that you want to do some thinking about—going back to the 5P framework—purpose and performance—which one matters? Do you care more about cost? Do you care more about privacy? Do you care more about the skills needed to set it up? Those govern which version you’re going to choose.

Katie Robbert – 06:16
John, when you think of process automation and the work that you do on a regular basis, is there anything that if it were more automated, your life would be easier?

John Wall – 06:28
Yeah, there’s always, seems to rotate around the inbox—the big thing. Incoming messages, if they could—stuff that normally gets answered with a template, if that could get pulled—same deal with outgoing stuff, responding to stuff that comes in. I don’t have a ton of web traffic stuff that I would need it for. But email is just kind of always a thorn in the side.

Katie Robbert – 06:52
Makes sense. All right, so Chris, we know that you have to start with the purpose—why are you doing the thing? What is the question you’re trying to answer? What is the problem you’re trying to solve? The next P is people. So Chris, you mentioned that depending on the version you’re picking—local or cloud hosted—is going to vary who you need to help get this set up. If you’re doing the local version, you need technical expertise. You still want technical expertise. If you’re doing a cloud version, I’m assuming it’s not a, you know, set like right out of the box; it’s all set up. You probably still have to do some configuration. So that is something with software in general, not just generative AI that teams tend to forget.

Katie Robbert – 07:36
And why IT teams hate the rest of us so much is that we’re like, “Oh, and by the way, we just bought this new software and we need it set up yesterday. So go ahead and prioritize that over the other 10,000 things that I’ve already put on your plate because you’re the expert in this technology and privacy and governance.” So if you could go ahead and do that. So getting that squared away first and being like, “Hey, I’m going to need someone from IT, maybe a developer, maybe HR, maybe legal,” getting all of those people together ahead of actually making a decision to purchase something, they’re going to like you a little bit more. They’re still going to hate you a lot, but they might hate you a little bit less. Anything else on People?

Christopher Penn – 08:24
No, the one thing I will say is for the process part—the more that you know your processes currently and the better documented they are, the easier n8n is to use. The more opaque or undocumented your processes are, the rougher time you’re going to have because you have to—you’re going to be replicating your process with machinery.

Katie Robbert – 08:48
And I know a lot of companies struggle with this part because it feels like the hardest part of all the Ps, even though you know it shouldn’t because there’s a lot of tools and software that can help you. But what ends up happening—and we heard this from a couple of different clients—is, well, there is the way that we think it should be done, and then there’s the way that everybody else does it. So if you ask four different people, you’re gonna get four different versions of how something is done. That’s not helpful to try to automate it with something like an n8n software because then you have four different versions, or you have a version that nobody’s using. So like, well, I wouldn’t adopt this.

Christopher Penn – 09:30
Look at you. Companies have GPTs for everything in triplicate.

Katie Robbert – 09:37
So you know the purpose—why am I doing this? You know the people you’re not going to make enemies with your IT team. You have a well-documented process. Only then—even though you’re going to do it first—only then should you be choosing your platform. And then you know if n8n is the right tool for you, then go ahead and start building out in n8n.

Christopher Penn – 09:57
Right. And the performance you’re going to measure in a couple of different ways—you’re going to measure is the task getting done faster and is the task being done at a higher level of quality because there are fewer mistakes. Because the whole point of automation is to remove the variability of the human being in the process.

Katie Robbert – 10:15
Right.

Christopher Penn – 10:16
All right, so let’s go ahead and dig into getting this thing set up. Now again, there are a couple different ways. If you are going the paid route—the cloud-hosted route—pretty simple. Like every company on the planet, you just go and sign up and you swipe the credit card. If you are building an n8n company use—meaning it’s not going to be controlled by one individual but might be used by a team and you don’t have strong technical expertise—pay the money to have to use the cloud-host version, unless there are specific privacy-related things that you don’t want a third party seeing. If I click on this and I go to the GitHub version, this is the open-source version.

Christopher Penn – 11:01
If you are familiar with installing open-source software, you can simply install it. It uses NPM and npx, which is part of Node.js. If you know what those words mean, just NPX n8n@latest and you’ll be up and running quickly. If you are not familiar with those words, perhaps this is not the version.

Katie Robbert – 11:19
For you, this is not the version.

Christopher Penn – 11:22
For me, this is not the version for you. Once you get up and running, you will end up in an environment that looks kind of like this, except there won’t be any existing workflows there. You will just have a blank task area, and you can start creating a workflow. When I create a workflow, this is where you start. And it looks completely underwhelming because nothing is there yet. On the right-hand side is all of the menu items of the things that you want to do. So first, it’s going to ask you what you want to trigger the starting of this workflow? And this can be things like—it runs on a schedule; it receives something from an external system; it is pinged by an external system. There’s something that basically makes your workflow start.

Christopher Penn – 12:11
And this is the heart and soul of automation is to say, yeah, we’re going to do something on a schedule, or when something happens. For today, I’m going to choose trigger manually because obviously this is a demo, but in production, you would perhaps probably be doing it with things like—hey, when a new file is added to my Google Drive folder, or when someone puts a new line in a Google Sheet, or when I receive an email, or when someone submits a web form—those would be the triggers. After a trigger, you’ve got to do something. What is it that you’re going to do? This is where n8n is really powerful and extremely confusing for people because what can I do? And the answer is pretty much anything, which is completely unhelpful.

Christopher Penn – 12:59
And so this is where having an existing process that you know is awfully handy. I’ll give you an example of an existing process. Maybe I want to look at the podcast episodes from the Trust Insights website. So I could go into my nodes here. I could say WordPress; I want to do something in WordPress; maybe I want to get some WordPress posts. And I’ve already done the authentication here. So I’m going to say I want to get my posts; however, I want to get my posts specifically that contain In-Ear Insights. That would be an example of the kind of thing I’d want to do. If I just hit test workflow, it’s going to start churning away and it’s going to start pulling posts from our WordPress site. Once that’s done, then the question is, what do you want to do with that information?

Christopher Penn – 13:57
If you have things like the post title or the content and things, what kinds of things might you want to do? And this is where we get into the AI portion. You can say with AI, I could build myself an agent or just a simple prompt or do sentiment analysis or do all sorts of things. There’s no shortage of different AI nodes you could do here. You could also connect to all sorts of other crazy AI tools. For today, I’m just going to use a basic LLM. Now this part is saying you’re going to talk to an AI, and how you do that is going to be dependent on what it receives. I’m going to choose define, and that’s where I can put my prompt.

Christopher Penn – 14:40
I could say, “Summarize this transcript from the Trust Insights podcast,” and maybe I might have my ideal customer profile written there. “Summarize the five things my ICP might want to know about this.” If there are system instructions that you’ve built in the past, you could plug them in here. There you can see the WordPress script finished running. And so it’s then there’s In-Ear Insights and so on and so forth. Let’s go ahead and connect this. I’m going to connect this for today just to DeepSeek because this is non-private data. You absolutely positively would never ever do this with any kind of private data because DeepSeek’s privacy policies are non-existent. But I might want to say, “Summarize the post based on our ICP.” Let me pull up the Trust Insights ICP here.

Christopher Penn – 15:43
I’ll put the ICP in, and then I will take my post content and that goes in there, and so now it will read the post and it will summarize it. Now I probably—just for safety’s sake so that I don’t cost you a million dollars because you can see there are 692 posts here—I probably should say instead of return all, maybe do two.

Katie Robbert – 16:12
I appreciate that, but I think those are the things that if you’re just kind of playing around with it because I can see where someone might be like, “I’m just going to open it up and see what I can do.” Those are the kinds of toggles and switches that are going to get you into a lot of trouble if you don’t have something like the 5P framework already written out. Yeah, you can play with this all you want, but putting it actually into production could end up costing you money, but also—to your point—when you opened up the menu of widgets and connectors and things that you can do, there’s no shortage of things that you could do.

Katie Robbert – 16:51
So if you are trying to focus or trying to get something done, but you very easily get distracted, you could go down a very deep rabbit hole of getting nothing done. But then they’re like, “Well, what did you do? You just cost us $2,000 because you ran 692 blog posts.” Like, “Yeah, I was just testing it.” Like, “Great. So where’s that $2,000 coming from, John?”

Christopher Penn – 17:20
We’ll find it.

Katie Robbert – 17:20
We’ll find it.

Christopher Penn – 17:25
And so one of the pieces of advice we would give using generative AI with a tool like n8n—and we covered this in previous episodes of the Live Stream—is using a local model, using something like a Cobalt cpp that’s running Gemini 3 or something so that you can test. And when you test, you’re just hitting your machine, and yeah, you’re going to make your fans spin, but you’re not going to cost yourself a million bucks in API calls. So this is finished running. We can see there’s the input side; there is what DeepSeek came up with; and there is the output side, and it created the summary of these posts based on the ICP. Now we’ve got to do something with that because you can’t just leave it in here. It’s got to go somewhere.

Christopher Penn – 18:14
So I’m going to just say I want to write this as a file to disk, and we’re going to put this here. However, the challenge with this is that what is in the data stream here may not necessarily be all that useful. I need to transform this information into—let’s see—convert JSON. I’m going to convert JSON data to binary data, and we’re going to convert it into plain text. Let’s go ahead and remove this block here. This next step says, “Hey, the response that came back from the AI, which is in AI’s native language, which is JSON, needs to be turned into text because otherwise it’s super not helpful.” There is the text. If I test this step, it has now run, and now I can add that block to say, “Write this file to a disk again.”

Katie Robbert – 19:11
I feel like that’s where being clear on who’s involved is going to be helpful. So when you’re doing the 5P framework, if you don’t know that you’re adding the AI and you don’t know that the native language of AI is JSON and you get out a bunch of what looks like garbage, but really it’s just in a different coded language, you’re going to be sitting there trying to figure out how the heck did I screw this up, where did it go wrong? So again, I’m going to keep beating this drum, harping on this—pick an instrument that you need to do the 5P framework before getting into new software and setting it up. Because we don’t have time to waste; we don’t have budget to waste.

Katie Robbert – 19:56
I certainly don’t want to be on the receiving end of someone finding out that I accidentally spent a few thousand dollars.

Christopher Penn – 20:05
Exactly. So this is the output here. This is a file says, “Summary of this podcast episode aligned with Trust Insights ICP.” And here are the key takeaways for this ICP about what they should have learned from this episode of the podcast. Now, from here, you could turn this into other things. For example, if I wanted to—if we like that description—we might want to tie it into—let’s see—do I have? Yes, I have LinkedIn. So I could specify I want to create—turn into a LinkedIn post. And so I might say you could go in and configure and build a LinkedIn post template that would allow you to convert that podcast summary into a LinkedIn post. I have not set up my credentials and automatically post and say, “Here is a LinkedIn post that is tuned for our ideal customer profile.”

Christopher Penn – 20:54
So that might be the kind of thing you would put on a brand page. When we think about the time that a social media manager might say, “Okay, I’ve gotta watch this week’s episode. I gotta figure out what the talking points are. I gotta make a LinkedIn post for it, and stuff like that.” That could be a lot. But if I could get it to this point, perhaps I could do that exact process of wiring together and say, “Okay, now, as soon as the podcast is available, have it assemble a LinkedIn post tuned for the ICP.” That’s the critical part—we’re using the generated AI capabilities. I can make a link to post; here’s a new podcast.

Katie Robbert – 21:34
But that’s—and I think that’s—again, such a critical step because it’s not just a matter of connecting these widgets. There’s a lot of customization that goes within. So in order to create a LinkedIn post that matches the Trust Insights brand voice, we would need to give that information to that LinkedIn widget—maybe it’s a set of system instructions; maybe it’s writing samples; maybe it’s all of the above. Those things need to exist. Those knowledge blocks that we talk about week after week have to go into each of these widgets, which is why having that information handy in your library—whatever it is you’re setting up—is going to make this process, setting this up, even more efficient.

Christopher Penn – 22:20
Exactly. Let’s expand on this idea. Let’s do another basic LLM chain here. I’m going to take tied straight to my WordPress post like before, and I’m going to add DeepSeek to it as before. And we’ll use the Reasoner model because that’s better. And let’s—I’m going to give this a terrible prompt because just for example—take this post content and convert it into a Google Ad. Keep the ad to 75 words or less. And here is my post content—gonna drop—oops—drop that into the widget here. There we go. And then from there downstream, I could say, “Let’s make a Google Ad.” I have to get the campaign first, put that there, and then so on and so forth.

Christopher Penn – 23:27
So you can see kind of where this is going is if I have a post come up, I maybe want to write the ad copy for it and start getting the ads chained together. So this is a very theoretical example. I’ll show you a practical example. This is one that I use in production every single week. And what I do is I take the Fireflies transcript of the Trust Insights podcast. I have n8n read it. I have it extract out and that transcript—what comes to the Fireflies is, you know, needs some cleanup. So the AI cleans it up first. It then writes those files to disk so that I can put them into our blog post later. I could put it in WordPress directly, but I like to just check to make sure the machine’s doing what’s supposed to do.

Christopher Penn – 24:12
The second is I have from the cleanup transcript; I have it make the captions for YouTube to say, like, “In this episode of the Trust Insights podcast, Katie and Chris talk about whatever.” It spits that out in both HTML and Markdown. And then from the cleaned-up transcript, I then have it say, “Figure out the smartest thing that Katie said in 60 seconds or less because what is what I use for the social posts,” and I have it right back. This process used to take me about 30 minutes a week just to go through all these processes.

Christopher Penn – 24:40
Now this takes me—because I can say, “Here’s the transcript, go and come back later, come back in 10 minutes when all the processing is done and my files are just on my hard drive ready for me to do stuff with,” and I can stage my content much faster.

Katie Robbert – 24:57
I think the thing that’s worth pointing out though—and this is what we’re talking about with the setup and knowing your process—is within each of these widgets is a set of instructions. You can’t just drop in a widget and call it cleanup transcripts. You actually have to give it the instructions of what that looks like. How do you clean it up? Is there a specific way that it should look? “Find the smartest thing Katie said,” I don’t even know what those system instructions would include, but I would assume that there are some because—number one—it has to identify that I’m speaking and not Chris, and number two, it has to find something that’s coherent and sounds like a fully structured sentence.

Christopher Penn – 25:39
Exactly.

Katie Robbert – 25:40
Hoping.

Christopher Penn – 25:41
Exactly. So these—this is an example—the system instructions. “Three 30- to 60-second clips of Katie’s most impactful ideas ranked in descending order by importance and key insights.” And so this goes through a fairly extensive process to try and figure out what it was that you said and then return the results. And so when it runs, it’s executing those specific instructions. One of the critical things—going back to where we started today’s show—if you don’t already have the process documented out, n8n will just add an additional layer of complexity on a system that’s not already working particularly well. You need to have all of the prompts that are being used, well tested and proven before you build a system like this.

Katie Robbert – 26:32
John, you’ve been producing and editing the Marketing Over Coffee podcast for 16 years, at least. Does something like this appeal to you? Like, is editing the podcast and producing it and getting it out the door and getting your monthly newsletter that is always consistently sent every month—does this type of process appeal to you as someone who’s taking on all of those extra steps?

John Wall – 27:01
Yeah, the idea of getting the summaries is good, and we have started to do more with the transcripts. I have that stuff going into Fireflies. I’m still caught on all kinds of points. But yeah, it’s definitely—it’s obvious that this gets a ton of the grunt work out of the cycle. If you want to get more stuff on there, it’s definitely the right way to do it. You just can’t do that stuff manually; it’s too much there.

Christopher Penn – 27:27
Another example of one—let’s go ahead and leave without saving—another example is for my personal newsletter. I use Google’s Gemini; it’s a phenomenal translator with a good translation prompt. And so every week I take my newsletter that I publish in English. I have it translated into Chinese, Korean, Melee, and Latin American Spanish, and that goes on my personal website. Being able to process this takes about 15 minutes to run because Gemini thinks through a lot of stuff very heavily. But boy does this save me time because when I did this manually, it was just copying and pasting—copy paste, copy paste. That’s not a good use of my time. I don’t want to be the middleware. I want to be out of the process entirely.

Christopher Penn – 28:17
This is an example where we can take information out of HubSpot, look at our ICP, and then score the contents of a lead against our ICP and say, “How is this lead a good fit for us?” As an example—this is not in production yet because there are a bunch of bugs I’m still trying to work out—you will notice however, down here it’s using the Ollama chat model. Ollama is local AI because I do not trust our confidential data with a third-party API for something like this.

John Wall – 28:54
Like.

Christopher Penn – 28:54
Nah, you know what, I’m going to run Gemini 3 locally on my computer, and yeah, it’s going to make the fan spin, but the contents of our CRM are sort of the crown jewels. We don’t want to be handing that to a third party.

Katie Robbert – 29:08
So let me ask you this question in terms of introducing automation software such as n8n for lead scores. A lot of CRMs already have lead scoring built in; it just needs to be configured. What is the benefit of doing this way? Aside from maybe you can’t afford the functionality?

Christopher Penn – 29:30
Certainly, cost is an option. Second, most CRMs, they score on a relatively primitive set of factors. So you can go in and adjust the lead scoring if they have a known domain like this or their domain rating is like this or they visited last time or visited these pages; you give it a point for this, a point for that. When you think about like an ICP—goals, motivations, needs, pain points—and you think about the information in your CRM—the emails, the notes, the conversations, the call transcripts—none of that is taken into account by today’s lead scoring mechanisms because they can’t. It’s just the computational overhead would be ridiculous. When I do get this working, it will be taking into account all that qualitative information.

Christopher Penn – 30:17
How if I talk to a prospect on the phone, how did my language and the prospect’s language map against the known needs, pain points, goals, and motivations of our ICP? Are we having conversations that contain buying intent? Language models are really good at that. Traditional lead scoring, not good at all.

Katie Robbert – 30:37
I would love a process. Now I’m just giving you my wish list, but I would love a process that includes this, but then also takes it to the next step that says—based on all of the input calls and transcripts and knowledge blocks about Trust Insights and our services and our expertise—it actually spits out a draft of a scope of work that then John would just read over and say, “Yep, this is exactly what they asked for. Let me just serve that up to them and add in some pricing.” I feel like that—especially as a company like ours who’s small—you’re looking at the majority of the company right here trying to scale. Those are the things that take a lot of time.

Katie Robbert – 31:22
Like we can build a custom GPT or a custom model to write a scope of work, but there’s still a lot of input that needs to go into it. If you can automate from start to finish—”Here’s all the transcripts of input calls. This is what they said on the contact form. This is our services put together what they’ve asked for with what we offer”—and do that as a first draft of a scope of work. That’s a big time saver. That’s huge. I would still want John as the sales lead, as the expert, as the human intervention point. But how much time he could churn up so many more scopes of work if he wasn’t focused on writing one at a time. I’m just making your job harder, John.

John Wall – 32:07
Now I’m all up for more scopes that I’m 100% in for that one. And I mean, we’re—it’s not automated, but it is amazing how much things have changed just in the past year. Like, so much of the existing scopes come from Fireflies. We do the initial calls there, and those summaries usually end up almost verbatim in the scope as far as task list order and things like that. So yeah, it’s obvious that can really cut some time out of the process as far as just the—”Okay, here are the 10 things we agreed on; let’s get those in the scope so that we’re on the same page.”

Christopher Penn – 32:40
And so the big challenge for n8n is not can it do it? The question is, what do you want to do? So if we look even just what the input triggers are—this long list of services—here are all the things that can cause n8n to run. Comments on your blog, an ad action from Facebook ads, something from Ghost, the newsletter platform, you name it. There’s Harvest timesheets. So if maybe you’re an agency uses Harvest, for example, you might say, “Hey, when an employee triggers this thing, say, ‘Oh look, we’re running; we’re going to be overservicing a client.'” Get data of all the clients or get expenses or projects. “Oh, our time entry; we’re overservicing this client.” You can create a warning that emails your employees saying, “Hey, you’re overservicing this client. Stop it.”

Christopher Penn – 33:40
Take your pick of any of the gazillion and a half connectors that are in here. What could you do to then take an action downstream whether or not AI is involved?

Katie Robbert – 33:53
And I think that—so obviously we’re focused on AI automation in this episode, but I think that’s one of the things that people who are trying to figure out what to automate, they automatically think, “Well, it has to be AI; AI has to be involved.” And I mean, think about traditional automation; AI is not involved at all. We’ve been doing what—Robotic process—RPA.

Christopher Penn – 34:18
RPA.

Katie Robbert – 34:19
Robotic Process Automation forever. That does it before AI was available in the mainstream way. It’s really just a matter of going back to the 5P framework—what is it you’re trying to do? In the use case that I just put together about creating more scopes of work, it’s about scale. And so how do we scale? And then—but also the other unit of measure would be—are we closing more contracts? It’s not enough to just spin up more contracts. If nobody’s signing off on them, then this is a whole waste of time.

Christopher Penn – 34:59
Exactly. And so it just comes down to what is it you want to do? What is the thing that you’d like to try and tackle? Do you have a process for it? What is that process? How well documented is it, and how would you adapt the tool to do it? I saw in the comments, “Can you automate content creation using ChatGPT and post on different social media?” Yeah, of course you can, but you’ve got to have a place to start. So what is the starting content that you would use? Is it posts from Twitter as opposed from LinkedIn as opposed from Facebook that you would want to ingest and then perhaps reformat or re-spin and then publish outward? You have to come up with the process.

Christopher Penn – 35:45
Once you do, then it is open field as to what happens downstream from it.

Katie Robbert – 35:54
Could you start the trigger where I say something like, “I have an idea in Slack,” and that triggers an n8n process to tap into KDGPT, write a thought leadership post about it and then post it to the various channels? Because I’ve already built the custom model, but all I had to do was say, “I have an idea.”

Christopher Penn – 36:21
You could; you wouldn’t be using KDGPT; you’d use the Assistance API. So a part of n8n used is knowing what APIs are available and how to tie the various systems into it because they’re not the same as the consumer-facing AI pieces. But if you know it exists and you can copy and paste the contents of KDGPT into a KD assistance API, absolutely you could do that. You could have it be a—you could have you listen to a Slack channel for a specific keyword or term to invoke it, or you could even use n8n. This latest version allows you to do Model Context Protocol.

Christopher Penn – 36:59
So if you—from last week’s podcast episode—if you can get that set up, you could then invoke it from a tool like Claw Desktop and say, “Use KDGPT in this instance,” and have as part of a natural chat with your regular AI, have it kick off that task. n8n has an MCP trigger, so it can act as an MCP server, and you can invoke it right from your desktop app and kick off that process. That’s one of the things that makes it so flexible is it can tie into all of these things. However, again, you’ve got to have all the pieces assembled so that it knows what to do.

Katie Robbert – 37:40
So when should you not use n8n? So we know if you don’t have a clearly defined process, you’re not ready for automation. If you don’t have the technical expertise, you probably shouldn’t be standing up an automation tool. If you don’t have a clear purpose or performance metrics, probably don’t get started. But are there other times when using software like n8n is the wrong choice?

Christopher Penn – 38:10
There is a point after which it does not scale well. When you start getting into 2, 3, 400 node workflows, at that point, you have to ask yourself, “Would I be better off having this be a dedicated app written in a language like Python or something where it is just custom bespoke software?” I would say once you start having trouble maintaining your workflows, or once they start getting burdensome, or once execution speed really starts to matter, then look at converting your n8n workflow into actual software that can run on a server. There will be cases where there are compliance issues. You say, “Yeah, I don’t want other people being able to edit this.” And so n8n might not be the right tool. Again, bespoke custom code might be the right tool because you need to have it locked down.

Christopher Penn – 39:03
Or in a very specific instance, an example would be a tool like Snowflake where you can build right inside that ecosystem, and you’re not allowed to use connectors. There would be cases where—because so much of n8n is reliant on APIs—if you are using systems that don’t have an API or that you can’t afford the API, again, not a great choice at that point. There are probably manual processes you need to use where you could take the outputs. For example, Fireflies is a transcription tool. If you’re not paying for the business or pro plan, you only get 50 API queries a day, and so n8n could consume those really fast. And you might want to have a lower-tech solution than using the API.

Katie Robbert – 39:49
What do you think, John? Do you have some processes that you’re going to start automating?

John Wall – 39:54
Yeah, I would love to start just playing around with it because the devil and the fun part is in the details—like posting automatically to LinkedIn. I’d like to see how much of that can you do before the wrong people get angry. And also, setting up email to automatically go out the door—how much of that can be automated and what kind of stuff do you have to do? Can it just get routed through my personal accounts, or do I need to set up something more industrial? But yeah, it just seems like it’s a better place to start than Zapier. It’s more in line for the kind of stuff that we’re doing, and with the generative capabilities, there’s just a whole lot more there that it can get done.

Katie Robbert – 40:33
Well, I think with that, I would definitely still encourage you, John, to use the 5P framework responsibly, but it sounds like you’re talking more of like a proof of concept versus full scale—”I’m going to suddenly hand over the entire Marketing Over Coffee production process.” But a proof of concept is always a great way to start. It’s the experimentation; it’s the R&D. It’s how you find out whether something is actually working. But even good experiments have a structure around them. And I told you—pick an instrument; you can get your version of the 5P framework to walk you through it at our website, TrustInsights.ai 5P framework. I will never stop shouting that from the rooftops that you have to have some sort of structure, even if it is an experiment.

John Wall – 41:20
Yeah, it’s just crazy. Like, there are no classes; there’s no manual to read. It’s like you have to get in there and start messing around. Unless we do a 5P framework training on n8n, I could see that coming in the horizon.

Katie Robbert – 41:35
Well, you know, start up your automation engine and add that to the backlog.

John Wall – 41:41
Excellent. See my course soon.

Katie Robbert – 41:46
Anything else we need to know about n8n, Chris, to use it responsibly?

Christopher Penn – 41:50
Nope. I would say again, go through the 5P framework, decide which flavor of n8n you want to be using, and then, yes, start diagramming out your processes and building them. I would encourage people, as always, start small. Don’t try to boil the ocean all at once. Pick a use case that you know well. Pick something where you already have proven prompts that you know work well so that you don’t have to try to debug AI at the same time you’re trying to debug an automated workflow. And finally, I would say once you’ve got these things really well baked and they have automated triggers, you then evolve from an automation to an agent. So a lot of people will talk about agentic AI and stuff like that.

Christopher Penn – 42:34
An AI agent with a tool like n8n is just a workflow—that automation that runs without you participating in it at all. The caution there is, please don’t try to build an agent. Build a workflow first that is all manual so that you can prove the individual prompts, the individual pieces. Then build an automation from your manual workflow, then build the agent. Because you have to have a solid foundation of stuff that you know works before you go building agents. If you try to build the agent straight away, it’s going to be a bad time.

Katie Robbert – 43:08
All right, so we’re gonna have to wait a few more weeks for Agent John to make his appearance on Marketing Over Coffee.

Christopher Penn – 43:15
Exactly right. Exactly right. Any final thoughts, Kenny?

Katie Robbert – 43:19
I mean, just you know, map out your plan first, but then you know, start doing it.

Christopher Penn – 43:25
Start doing it. One other thing for the very technical—if there is not a node in n8n that serves your company’s needs or whatever you can use, have it execute Python code. So if you have developers or you are a developer and you have a piece of Python code that does a very specific thing that just your company does, you can put it into n8n, and then it will run that code for you. So maybe you’ve got some legacy COBOL system that’s left over from the 80s. As long as it has some form of data interchange that is available, you could wire it in there as well. So just because it’s not available in the interface itself doesn’t mean it can’t do it; it just some extra connectivity that you will need to have set up.

Christopher Penn – 44:12
But that’s part of what makes it such a powerful tool is anything new that comes out; if you can write code for it, it can do it.

Katie Robbert – 44:20
All right.

Christopher Penn – 44:22
All right, that’s gonna do it for this week’s episode. Thanks for tuning in, everyone, and we’ll see you on the next one. Thanks for watching today. Be sure to subscribe to our show wherever you’re watching it. For more resources and to learn more, check out the Trust Insights podcast at TrustInsights.ai/TI-podcast and a weekly email newsletter at TrustInsights.ai/newsletter. Got questions about what you saw in today’s episode? Join our free Analytics for Marketers Slack group at TrustInsights.ai/analytics-for-marketers. See you next time!

---

Need help with your marketing AI and analytics? Get help with your AI strategy » Learn generative AI for marketers with our new course » Prove the ROI of your marketing with our advanced analytics services » Just got a "silly" question you want an answer to? Ask us here »	You might also enjoy: About Trust Insights How AI is Changing Marketing Keynote In-Ear Insights: The Trust Insights Podcast Empower Your Marketing With Private Social Media Communities Trust Insights Services : Data & Analytics Help, Predictive Analytics
Get unique data, analysis, and perspectives on analytics, insights, machine learning, marketing, and AI in the weekly Trust Insights newsletter, INBOX INSIGHTS. Subscribe now for free; new issues every Wednesday! Click here to subscribe now »	Want to learn more about data, analytics, and insights? Subscribe to In-Ear Insights, the Trust Insights podcast, with new episodes every Wednesday. Click here for Google Podcasts » Click here for Apple Podcasts » Link for all other podcasting software »

---

Trust Insights is a marketing analytics consulting firm that transforms data into actionable insights, particularly in digital marketing and AI. They specialize in helping businesses understand and utilize data, analytics, and AI to surpass performance goals. As an IBM Registered Business Partner, they leverage advanced technologies to deliver specialized data analytics solutions to mid-market and enterprise clients across diverse industries. Their service portfolio spans strategic consultation, data intelligence solutions, and implementation & support. Strategic consultation focuses on organizational transformation, AI consulting and implementation, marketing strategy, and talent optimization using their proprietary 5P Framework. Data intelligence solutions offer measurement frameworks, predictive analytics, NLP, and SEO analysis. Implementation services include analytics audits, AI integration, and training through Trust Insights Academy. Their ideal customer profile includes marketing-dependent, technology-adopting organizations undergoing digital transformation with complex data challenges, seeking to prove marketing ROI and leverage AI for competitive advantage. Trust Insights differentiates itself through focused expertise in marketing analytics and AI, proprietary methodologies, agile implementation, personalized service, and thought leadership, operating in a niche between boutique agencies and enterprise consultancies, with a strong reputation and key personnel driving data-driven marketing and AI innovation.