Leveraging AI

266 | From Spreadsheets to Strategy: How AI Turns Business Data Into Decisions with Keith Moehring

• Isar Meitis, Keith Meohring • Season 1 • Episode 266

Share episode

Use Left/Right to seek, Home/End to jump to start or end. Hold shift to jump forward or backward.

0:00 | 54:39

What if your business data could tell you exactly what’s working, what’s broken, and what to do next—without dashboards, analysts, or endless spreadsheets?

Most leaders know data-driven decisions matter.
 Very few have the time, tools, or teams to analyze data consistently, deeply, and at scale.

In this episode of the Leveraging AI Podcast, Isar Meitis is joined by Keith Moehring, CEO of L2 Digital, to break down how business leaders can use AI + automation to turn raw data into clear insights—and actionable recommendationsautomatically.

Instead of chasing reports, building pivot tables, or relying on expensive BI platforms, you’ll learn how to build an AI-powered analytics assistant that pulls data from multiple sources, identifies what actually changed, explains why it happened, and emails you the insights on a schedule you choose.

This isn’t theory. It’s a practical, repeatable system that replaces hours of manual analysis with intelligent automation—while keeping humans focused on decisions, not data prep.

In this session, you’ll discover:

  • Why “more data” isn’t the answer and how AI helps surface the right insights
  • How to automate multi-source data analysis without advanced technical skills
  • The 4-step framework for building reliable AI analytics automations
  • How AI identifies deviations, root causes, and meaningful trends
  • When to use rigid automation vs. flexible AI reasoning
  • How leaders receive clean, actionable insight reports directly via email
  • Why separating “analysis” and “strategy” agents improves AI output quality
  • How this approach applies across marketing, finance, sales, HR, and operations

About Leveraging AI

If you’ve enjoyed or benefited from some of the insights of this episode, leave us a five-star review on your favorite podcast platform, and let us know what you learned, found helpful, or liked most about this show!

Isar Meitis

Hello and welcome to the Leveraging AI Podcast, the podcast that shares practical, ethical ways to leverage AI to improve efficiency, grow your business, and advance your career. This Isar Metis, your host, and we have a really awesome episode for you today. And me as a data geek, I'm really excited about it personally. Uh, but every business has multiple data points and being able to pull the data from all these different points and actually identify deviations from what expected and finding reasons for these deviations and diving into specific unique events and analyzing what's happening in them is. One of the most critical aspects of driving a successful business. Now that being said, doing that, especially doing that consistency and at scale is not easy. You either need somebody in your company who is a data analyst, or at least somebody who is an Excel whiz, uh, in order to do this. And even then doing it consistently and at scale is not easy. So that's why bigger companies have departments or teams that are doing, uh, business intelligence and they invest a lot of money in really sophisticated and advanced BI platforms, and that enables them to do some magic. But most companies do not have the budget, uh, or the right people or resources to do that. The good news, AI enables to do it. Amazingly well, AI is really, really good at analyzing data and finding needles in haystacks in really big chunks of data, and you can do it with almost no technical skills and almost for free, which is a huge benefit. So in today's episode, we're going to teach you exactly how you can set up something like this in your business, how you can go to data sources that you have access to, and how you can build the right processes around them. So AI can help you analyze the data, find what's good, bad, and ugly so you can make informed business decision, so you cango your business in the most effective way and how you can do this while at the end of the process it just sends you an email with the results on whatever cadence you want without you having to go and doing the query on your own every single time. Now, as I mentioned, I think that being able to make. Data-driven decisions as the way to drive a business is maybe the most critical aspect of making the right decisions. Executing them is a whole other problem, but making the right decisions to drive your business in the right direction. And so, as a data geek myself, I'm very excited about this. And to guide us through this process today, we have Keith Mooring, and Keith has been the CEO of L two Digital, which is a marketing and sales agency, uh, in the B2B space. He's been running it for six years, but before that, he was the VP of growth at PR 2020, uh, which is another agency he's been at driving growth, both of them in the Cleveland area. By the way, PR 2020 has some other AI ninjas like Paul Reyer or Michael put, which you may know, uh, if you listen to podcasts or in your, in the AI space. Uh, but Keith has been. Using AI combined with automation tools for a while, and every time I see what he does, it blows my mind on how effective the stuff that he builds is. And so he brings to the table 20 years of experience of running businesses and he uses his AI and automation knowledge to grow the business, which is what makes it more interesting. It's just not a geeky exercise in building automation. It actually drives business value. And as somebody who's A, an AI geek, BA data geek, and c somebody who started in grow companies, this is really the core of what I love doing. And so I'm personally very, very excited to welcome Keith to the show. Keith, welcome to Leveraging ai.

Keith Moehring

Thank you very much for having me and for the intro and for calling me a geek several times. I actually, I love it. So that's awesome.

Isar Meitis

Uh, I, I'm, again, I, I, I was very genuine. I, oh yeah. I think this is gonna be really fun and I'm looking forward to see, uh, what you prepared for us. So, stage is yours.

Keith Moehring

Yeah. No, I appreciate it. Yeah, and I, I think the, we're gonna kind of talk through is how to approach building these AI automations in a way that can surface usable insights out of data. So I'll give you an example. So every week, every Monday, I get an email in my inbox. In that email, there's kind of the main website, KPIs, sessions, users, that kind of thing. It breaks down channel performance, how each channel did, did it go up or did it go down? And then overall is the site up or down? But then it goes a level deeper. So it identifies the top one to two to three channels that were the biggest reason that we're up or that we're down. And then in each of those channels, it goes in and dives deeper. So. Channel. Okay, then what source details do we get? And then for each of those sources that had the biggest change, landing pages, and then the email summarizes all of that, it says, okay, here are some of the insights that we found based on the data of what we're seeing. And then in that same email for that specific channel, it'll also recommend next steps. So if it's like organic search, you're gonna wanna write these three pieces of content, update this other piece of content, social media, here's some posts to share based on what seems to be getting the most engagement. This thing is awesome now. I, I've done these types of reporting for years and when I would do this manually, the process of building these reports, we're talking about two hours to pull the data, process data, check it, balance it, do the calculations, then another two hours to sit there and like write up the whole thing, edit it, polish it up, and then that probably involves at least two to three people, someone else to check it and sanity check kind of all that fun stuff. And all for what, I mean, we're talking about doing this weekly. You're not unearthing, mind blowing revelations every time you do this. It's maybe one or two small insights that can help incrementally improve, uh, overall performance. But with all the data we have available, with all the tools that can connect back and forth, we're expected to make these decisions using this data. And so. Uh, the, the, the challenge then comes how much time is worth your time? And then what can we do to automate some of this process? And, and thanks to AI and automation, that report I get every Monday, I don't do anything. All I do is read it Monday morning. It pulls the data, it processes it the exact way I wanna process it, uses AI to understand what's going on, generates the report and sends it off into my inbox.

Isar Meitis

Amazing. I, I'll, I'll add one thing again just to generalize it. For those of you who are not in the marketing space, I'm doing a workshop, uh, tomorrow for a pretty large group of financial analysts from a large international corporation. They do financial analysis of large international corporations. So they have these really ugly, big Excels that have 40 columns. Uh, and with. Thousands or tens of thousands of rows of cost analysis per their projects because there are people in multiple locations with multiple price points in multiple roles working on multiple components of the project. So every time a person somewhere does a work, it adds a line to that Excel that says he is from this location, he has these, uh, roles, these capabilities, he's working under that project on these two products under et cetera, et cetera, et cetera. Like, so there's again, multiple columns for every time a work gets done and then somehow you gotta figure out. Are we ahead of budget or behind budget? And where are we behind budget and why? And stuff like that. And, and again, it's just, it's, it's impossible for, it's not impossible, but you have a team of people sitting there doing pivot tables and different macros and different analysis to try to figure out, first of all, where is the deviation? And second, like Keith said, uh, so what are the recommendations? Okay, what, what should I do next? And, and now AI knows how to just do these things and does it replace the human? Probably not because you still wanna read it and decide what you wanna do as far as the taking action on the business. But analyzing the data and getting the report, you can apply this to any aspect of your business from finance, marketing, sales, customer service, uh, hr, et cetera. Wherever you have data, you can do the same process. And that's why I love where Keith is taking this, because we're gonna. Show you the how to think about this and how to build this, and not just, okay, here's an automation that does marketing data analysis.

Keith Moehring

Mm-hmm. Yeah. And I, I totally agree. And the, the, the fun part about, so ai, these tools are mind blowing. I mean, amazing. Uh, but they do have their limitations, especially when it comes to what they connect to, what they know to go get context and all that, which is why I, I'm so bullish on AI plus automation. So with the automation tools, now you have the ability to programmatically determine exactly what you wanna go get and bring back. And then you layer that in with some additional context to the AI to then do what it does best. And so what we're gonna walk through today is kind of like the four step process that when we attack these different projects, like what we think about and how we go step by step through them to make sure that we're covering all the bases and everything's thought through, and we're using AI to, its the max potential.

Isar Meitis

Okay. Sounds awesome.

Keith Moehring

Yeah. So, uh, so I guess the kind of real quick, high level four main steps, uh, when you're approaching any of these AI automation projects. The first one is to define approach. So define your approach. So think about it in terms of there's a, the way you go about doing something or your thought process behind why you approach generating content this way, or how you go about pulling data specifically and looking at the metrics. So it's your overarching set of principles and how something should be done. And then kind of the next step, which we'll get into is the detail process. This is the step by step. Walkthrough on exactly how this stuff works. And then from there, once you have that process kind of mapped out, how do we layer AI on top of it to do what it does best? And then you get into, okay, now that we know what the process is, we know where AI is, how do we build the actual automation itself? And I'll kind of walk you through an example of how this analytics assistant that I've built, how this whole thing works. So

Isar Meitis

awesome. Let's, let's dive in. Cool.

Keith Moehring

Perfect. Okay. So in terms of, let's start with the approach. So when I'm thinking about an approach for the analytics assistant, uh, there are really four main things that we look at and I've done, and again, done these reports for 20 years now, and the, the process to surface the most tangible insights I have fairly well, or the approach I have pretty well defined. So there's really four main steps here. Um, the first one, just to kinda give you an overall example of how to think about this. So with analytics. The first one is to look at how much the change actually was. So we don't wanna look at the top level metric. We don't wanna, uh, and just look at that percentage of how much it went up or down. We want the actual number of how much it increased or decreased. So it's the diff perfect example of this is, so if you go to a Cleveland Cavalier Cavaliers game, I'm from Cleveland, um, up on the scoreboard, they have. One team score, other team score, how much time is left, what quarter it is. But right in the middle they have a number and it's called the diff, and it's either like plus 10, plus two minus three. It's the difference between the calf score and the opponent score. At the end of the day, if that number has a plus in front of it, everything's good, but as a minus, it's bad. That one number is key, is critical to the whole thing. And really what we're trying to do to uncover the most valuable insights with analytics is to uncover why that number is up or down the way it is. And so that's why, and we will get into, when I show kind of the automation, how that works. That's really a lot of what the automation is designed to do is to. Pull out that type of information. Another one, another kind of approach principle that we have is five why's. So if you've ever read like the Toyota Way, I think is where I saw it. Um, they have this, uh, five why's philosophy where you continue to ask why five times to ultimately get like the root cause of the problem. So, uh, the, if you, and if you do that, and if you look at it from an analytics perspective, traffic's up. Okay. Why organic search is up. Okay, why is organic search up? Well, Google specifically is up, okay, why is Google up? Well, traffic, so the blog is up and then you kind of work your way down to get something very tangible and very specific. That's where the real, uh, insights lie, is buried in the why, why, why, why, why. So having a process to unearth that, that level of detail is very important to this whole process thing. The other side of it too. I have two children, uh, and they're now nine and 11. And even when they were kids, like you'd walk into a room, one would be on the floor crying, and then you'd say, okay, why, why are you on the floor crying? And then you'd get some sort of explanation of how it was the other one's fault. And she did this, she did this, she did this. And then as the, uh, unbiased, uh, judge in this situation, I have to go talk to the other one. And turns out her story, while there are some parallels, is wildly different that they're both describing the same event. So the third, uh, approach, the, the kind of, the principle we have is like there are multiple sources of truth. For different channels. Perfect example with analytics is Google Analytics, or I'm sorry, with analytics. Google Analytics four can give you organic search performance. It can show you how people or organic search traffic got to the site, what they did on it, what pages they viewed. But there is also Google Search Council, which can show you the other side of that same story where how many, what rankings are, uh, average position, impressions, clicks, all that kind of stuff. You marry those two together and now you've got something very, very tangible that you can use with the reporting. So when there are opportunities to layer different technologies on top of each other to combine the data, that becomes very, very powerful. Um, and then the fourth and kind of the, the, the other one that's really very useful when it comes to analytics reporting. Is, uh, context, context matters a ton. Uh, the example, um, for one of, uh, one of the sites that I manage, it's a golf, uh, live scoring system, and they get a lot of traffic, golf related traffic. Well, on, I think it was April 13th, 2025, there was this enormous spike in, uh, in traffic on the site. And you're like, what, what, what was going on? Like, why, why did this happen? Now we're we're trying to analyze the data, see what was going on, and nothing was lining up. Well, that was also the same day that Rory McElroy won the Masters. So there was this huge, enormous spike in interest in golf that day that we couldn't do anything to replicate with, uh, with marketing, but it's context, you need it to kind of level out the story. Another good example is a lot of the sites that I manage the last week of December, traffic d. In most cases they're B2B service companies.'cause everyone's on vacation. That context matters. Otherwise the, like, you give that to the AI tool, it's gonna go deep on what happened and may come up with insights that aren't really relevant.'cause there's no traffic that week.

Isar Meitis

Yeah.

Keith Moehring

So it's like combine looking at it from a holistic view of how do you approach this problem? How do you want, how do you need to go about it to get the outcome you're looking for. That's really the goal of the approach process and taking time to really think it all through, uh, is a very, very valuable first step.

Isar Meitis

Yeah. I love what you're saying. I want to add a few small nuggets. Uh, one I think is the most critical thing is at the end of the day, you don't need data for the data. You need data that will support business decisions, right? So having more data doesn't necessarily help having the right data. The right outcome and the right analysis is what helps. So you need to figure out what business decisions you want to make and what data will help you or what analysis will help you make those decisions. The other thing, and you touched about it's a lot, knowing what happened doesn't help. You need to know why it happened. Mm-hmm. Uh, because knowing what happens is great. I'm like, okay, uh, traffic is up, traffic is down, sales are up, sales are down, customers are happy, customers are not okay. But in order to actually change something in your business, you need to know why and to need to know why. There's usually two ways to do this. One is to add more layers of quantitative data, which is what Keith is talking about. Uh, in order to get additional. Layers of views, and then you are speculating less and basing your results on real data more. The other option, which is maybe less relevant in this example, but is extremely relevant in many examples, is getting qualitative data. Qualitative data, like actual customer reviews, like actual summaries of meetings with clients like, which is stuff that before AI we couldn't analyze. It was literally impossible because you couldn't go through 3,700 reviews that make sense in them. Uh, now you can, and so qualitative data is an extremely valuable tool to know why that you couldn't analyze previously and you can analyze right now. And then the last thing that Keith said as an afterthought, but is. Mind blowing when it comes to capabilities that we didn't have before is the ability to cross pollinate, siloed data from three or four different sources with one AI tool to make sense in them. So think about most of the tools you have. E-R-P-C-R-M marketing platform, HR platform, have their own little analytics environment. They have reports, they have dashboards, but it tells you what happens in that siloed universe. If you wanna know what happened to a specific channel based on the sales in that channel, combined with the customer reviews, reviews from that channel combined with the inventory, that goes to the, there's no way to do this, but ai, if you give it access to all the data sources, is very good at doing this and, and crossing the data and knowing that, which again, is a huge benefit of doing some of the stuff that Keith is gonna show us.

Keith Moehring

For sure. Yeah. And it's, it, it, it kind of painting the full picture. And I think you outlined it perfectly. That was exactly what the goal with that was, is like, it's all about qualitative, quantitative, marrying the technology together, painting a full picture. The more context you give these AI tools, the stronger the outcome. And so combined with all of that, you're giving it some very specific system prompts. And this is where these reports become extremely valuable, uh, and, and require absolutely, like in my case, zero time, except for to read it and to implement whatever the recommendations are. Um, so step two of this whole process where you're approaching the AI and automation, uh, projects, is, uh, we've got the approach. Now we need to translate that into a process. And the process is, if we're gonna define it, is a, the step by step, the very minute kind of ex details, one after another, after another. As if you were to teach this process, you're gonna hand this entire process over to some brand new intern who's never had a job before. Um, so think about it in those types of specificity terms. So I like to give you just kind of a very high level, quick example of this, where if I was to do this for like the analytics reports, I would first have to log into Google Analytics, pull up a report, and pull the session number for the current timeframe. Then, uh, I would have to, we have to calculate diff on that. I'm gonna go in and pull that session data for the previous timeframe, and then I'm gonna calculate that diff, and then I've got the change. But now I need to understand why at a next level, kind of the first why. So now I'm gonna go back into Google search count or search, uh, Google Analytics, and then I'm gonna pull source data, and then I'm gonna do the same thing for the compare. I'm gonna calculate my diff, then the other next Y you're going landing page data step by step by step. And then at the end of that, we've got all this Google Analytics data. Well, let's use, we can use, now we have to go and kind of summarize. Exactly what's going on there. And so we're gonna write down that summary and then, okay, we've got Google Analytics four data. Now let's take Google Search Council. We're gonna do the same exact process step by step by step. And then once I've got that summary written down, okay, let's pull in some additional context. Who are we targeting? Let's pull in some persona details. Let's, if we have a log of campaign activities and content that we publish, let's pull that context in. Uh, any sort of additional, uh, website behavior context, we need to layer in, like the drop off at the end of December. December. And traffic, uh, we typically, there's, uh, traffic doesn't pick up. Sales don't pick up till March because that's when the buying seasons, that kind of context layer that all into like a document. And so now I've got that paired with my data. Now I can start to really analyze and see what's going on. I've got all the information in front of me now I need to analyze it. And then once I've got the analysis, looking at the data, okay, what are the next three things that I need to do? And usually this is like the senior level person is going through and kind of like, okay, based on this, here's what the next three or four things need to be. You can build the A AI tools when we get to that stage to be that level smart, that senior level, smart, giving that all the context and then having it write out, okay, here are the next three things that we're gonna do based on what the data is telling us. So once we got that whole, and ideally that's a list, like, uh, it's a bolded list of start here, then this, this, so you can go all the way through. The third step of this whole process is

Isar Meitis

to be, before you go there,

Keith Moehring

one

Isar Meitis

thing, this is basically a really detailed SOP, right? It's exactly like you said. Oh

Keith Moehring

yes. Yeah.

Isar Meitis

This is an SOP for an intern that has never done this work. Down to what to click, where to go, what format to open, what to get, where to put it, how to copy, what to paste, what calculation to make. Uh, the easiest way to do that, by the way. Is to either yourself or the people who do this, let them do the process. Do a screen recording with whatever tool. You can do it in Zoom, you can do it in Loom. You can do it in any other tool that records the screen. Have them narrate what they're doing. Just explain in English, I'm doing this, I'm capturing that. I'm copying it here, I'm pasting it there. Uh, do it three times just because you're not gonna get it perfect the first time ever. And then take the three recordings, drop it into your favorite AI tool to transcribe, and they will do, oh, here's the process. And then you ask it. Do you think there's any open-ended things that you're not sure about how to do this process? Say, oh, what about this, this, and that? It's gonna ask you five or six questions, either you or whoever does it, let them answer the questions. And you have your SOP, now you're ready to go to the next step that Keith is going to describe.

Keith Moehring

Yeah, no, that's, that's perfect. I, and I do that, like the video recording of the process, I must have dozens of those recordings'cause it's the easiest way to like, make sure that everything's covered. And then you write the AI tool, just feed it into that and it becomes, it's like I, you know, no additional work. You just layering in some color where it needs to be. So we've got our very detailed step-by-step SOP process doc. The next part of this is to identify where in those steps AI is, AI can take over and do some of that work. And I do think it is worth kind of weighing out the pros and cons of what is AI good at versus what is it still potentially struggle with. And so like when we're talking what AI is good at, it's good at an good ana at analysis as long as the data is kind of structured and well, well, uh, formatted. It's good at writing out report, brainstorming strategy stuff, consolidating documents into summaries and isolated in or a smaller, more concise documentation. And then there are some things where it's not necessarily, and again, this is, this may be outdated by tomorrow, but there are some parts of this that it's just not quite perfect at yet. Um. Like, for example, uh, advanced processing of data within like a, like an unstructured format. So for example, I can pull out a Google sheet with all these rows and all this session data and pull the same thing for the previous time period. And I can feed both into analytics and have it compare and contrast and find the, do the calculations of page differences and, and then sort it and spit it out on the other end. I don't specifically know exactly how it's doing it, but there is a very specific exact weight to that it should be done. And so from a predictability perspective, this is something that's better to be handled like with a, um. Like with different types of code. So you can feed both in and process'em that way and get the data out, uh, through the code and then, you know, it's reliably going to be exactly how it should be set up. So that would be one example of, I don't think I'd hand the whole data processing thing if it's an unstructured thing yet to ai, but I would take whatever that output is and I would be comfortable handing that off to AI to analyze

Isar Meitis

a hundred percent. I think it's that trade off between rigidness and flexibility.

Keith Moehring

Mm-hmm.

Isar Meitis

And there's cases you want flexibility and a thought process and analysis, and there's cases where then you gotta get 90% accuracy, which is not acceptable. And so, uh, you, you wanna do the rigid stuff as much as possible because it's consistent. And then you want to use the brain where the brain provides value and not risk.

Keith Moehring

Mm-hmm. Yeah, that's perfect. So we've got our SOP. Based on our approach. And then we've got, we've identified what steps in that whole process is ai, uh, ready. So now the next step of this whole thing would be to build the actual integration. And so there are a variety of tools you can choose from, uh, make and n eight end, or probably the top two. Uh, you can probably Zapier to an extent. I, I like make a lot, largely because of how it integrates with different Google Properties. Uh, it's a little bit simpler. There's a Dutch mess. There's a learning more bigger learning curve with N eight N than with make. So if you're just starting out, even just trying out and make and then maybe graduating up to N eight. NN eight N does offer a lot more, uh, functionality in terms of what it can do and on the technical end of it. So pick your poison. Uh, I would say I, I, like, I have most of my stuff built in make, though, largely so clients can easily understand exactly what's going on too. Um, but. Uh, and so what I can do here is do you want me to do a quick screen share and kind of share?

Isar Meitis

Yeah, absolutely. Let's dive into N8N. See how, uh, what happens behind the curtain of the Wizard of Oz.

Keith Moehring

Uh, okay. So, all right, let me know if you can see that.

Isar Meitis

Yeah, I can. By the way, for those of you who are driving or doing the dishes or something, and you cannot watch the screen right now, uh, will explain everything that's on the screen, but those of you who can, if you're on Spotify or on YouTube, then you obviously can watch.

Keith Moehring

Okay. Alright, so here is, so this, these, what is this? 6, 7, 8, uh, eight different, uh, scenarios is what May calls them. These are the different AI automation workflows. I have got eight of these things separated out. They're all specialized, they all do one specific skill very, very well. In other words, I have a automation set up to analyze organic search traffic. I have another one set up to analyze email traffic. I have another one set up to generate the report and send it off at the end of the day. So each of these, um. Assistant, like channel level assistance in the email report generator. These are like, consider them as skills. In other words, they have a very specific purpose and they do one thing. They take data, process it in a specific way and return a predictable output. The thing that ties it all together is what, like we're calling this the master agent analytics assistant. In other words, what this is designed to do is it does some additional data processing. So the first thing we do is we set up some, uh, variables. So current timeframe, compare timeframe, uh, both start and finish, uh, description of the, the target audience, description of the company, the website, that kind of thing. Uh, we have that all built in because all those other automations are going to use that information when it comes time to call them. So we build'em all here so we don't have to update'em across all those different other automation. And then what it's gonna do is it's gonna go through and pull the overall traffic for the current timeframe and then the, the previous timeframe. In this case, in this case, last week compared to the week before that. And then we're gonna do some quick math. Did we go up or did we go down? And then we're gonna look at, at it from a channel perspective. So let's dive into and see and uncover of the channels that drove traffic, which ones were the biggest reason we were up, or then which ones were also the biggest reasons we went down. And then once we get that information, then we feed it into a, it's called make code, where it allows us to run some Python script. And the idea with this is we feed those, that channel data into it. And like I mentioned before, it's taking one spreadsheet of data and another spreadsheet of data comparing it. And ultimately on the other end, returning for us the names of the specific channels that had the biggest effect on why traffic changed. So in other words, direct and organic were the biggest reasons traffic went up. So it's gonna call those out. The other thing that this is designed to do is it actually will if traffic increased by a hundred sessions. Okay. Uh, but we uncover that the top two channels, if you add those up. Traffic should have increased by 200 sessions. Well, there was something that on the other end that's probably dragging the overall traffic down. So that's what this code's also designed to do that. Now here's what I will tell you. I am not, I don't know, Python code, like I, I'm not a Python coder. I know it enough to read it and to write, like jot it some stuff in. So what I did was I just outlined exactly what I needed this program to do into Google Gemini. I said I needed to do this. It needs to process this. Here's the sources, blah, blah, blah. And on the other end of it, I got this Python code. Then, then I just had tweaked here or there and it was good to go. So you don't have to know how to code to do this stuff. You just have to know exactly how it needs to work, and then use the AI to help you create that code.

Isar Meitis

And the same thing by the way, about troubleshooting. Let's say you run it and then it does kind of what you wanted, but it not exactly what you wanted. You just go back to in this particular Gemini or whatever the poison you chose to, to do the, the coding and say, Hey, this is what's happening, but I wanted to do this instead. And then it will give you a new piece of code and you test it again. And usually within two to three iterations, it will do exactly what you wanted it to do.

Keith Moehring

Yeah, yeah. If you can even give it the error, the specific error message,

Isar Meitis

yeah.

Keith Moehring

It'll be, it'll take it and it'll find exactly where the problem was and fix it from there.

Isar Meitis

And I'll give one more small tip that I started doing in N8N and in make, I actually run N8N or make in a, uh, in like an agentic browser. So like Comet and then you open the agentic browser agent on the side and you ask it. To write the code and create the code, and it creates it straight into the right box inside of Macon and a 10. So you don't need to know anything, it just does it for you, and it sets all the parameters and connects everything correctly. Uh, and, and it just, it saves the step of copying and pasting back and forth, and it also sees the screen. So I say, okay, here's, like, look at the screen, there's an error message. Go fix it, and then it will go and fix it. So that makes your life even easier.

Keith Moehring

Yeah. It's, it's also unnerving, but yeah, for sure. Yes. So, uh, so in this, back to this one, so we've got. Our overall traffic up or down, we know the channels and we have specific data for that, and we've identified the top reasons the traffic is up or down. Then we take all of that and feed it into an, uh, an AI agent. And so the idea with these AI agents is it's almost like you're building a custom GPT, but within make, so the, you go in and you, you give it a kind of a sym, uh, system level instructions. So if I go into here, you give it a overall, here's the system prompt. Here's who you are, here's your context, here's your instructions, here's exactly what you need to do with all of this stuff. And you can feed it in and provide it as much context and detail as you want. And then the other, what's other, what's nice with this too is with these agents, you can layer in, uh, files with the additional context. So for example, website traffic, uh, patterns throughout the year, put that as a document within the context for background information. Now, it can draw on that when it sees something happen within the, so for example, December, traffic drops off in December, call that out. And then with that, within there too, you also give it tools know or skills. So you're, you're telling it that all of the, here are all the scenarios or all the automations that you have access to. And then with, with each of those, you give it a specific set, uh, set of instructions on here's when you would use this. And so like with these, like with the organic search one, for example, use this tool only when organic search is a priority channel. This tool analyzes SEO performance from Google Analytics floor. And Google Search Console. Uh, and so in other words, here's when to use it. Here's what it does. And so for each of these, I have that instruction built into it. So that's the whole system problem. That's the agent as a whole. But then if I go back into, uh, let me go back into the, uh, the automation itself. I'm also then with the, like you give it the actual prompt as part of this conversation. So in other words, that system prompt is you building the custom GPT, this is you interacting with it and this is where you're gonna feed in this, this specific scenario information. In other words, traffic's up, uh, here's the breakdown of channels, if this has happens, also consider and think about this. Here's additional context and variables and details and so on and so forth. So you build all of that into this agent.

Isar Meitis

By the way, Gabe, for those of you who don't understand exactly how this works, I'm gonna pause for a minute.

Keith Moehring

Oh sure.

Isar Meitis

The way Keith connects the dots is in all these automation tools and make is no different. You can pull data from all the previous steps. When Keith is saying in the, when he's chatting with the agent, he's saying, Hey, traffic is up. How does he know that traffic is up? He doesn't. He say, traffic is, and then there's a parameter from one of the previous steps that says whether it's up or down. There's a parameter in one of the previous steps saying which channels are either down or up. So you can use these parameters that gets updated dynamically. So when you're writing the prompt, you're using the parameter. The parameter is called up down. I'm making this up. Mm-hmm. But when it's gonna say the sentence, it will actually look at the previous steps to know whether it was up or down, and will then fill up the sentence correctly. So every time it runs, it runs slightly differently, even though you have one prompt. Just because you used parameters inside your sentences. Inside the prompt.

Keith Moehring

Mm-hmm. Yeah. This is, this is like if you're, if you're, if you know how to code, this is a, this is very familiar territory. If you're not familiar with exactly how code works, it is really just you put in a variable. If this, do this, if this, then do that. And then you slot in different information that is previously generated.

Isar Meitis

Yep.

Keith Moehring

So the, the agent now has all the context it needs, it has the whole instructions that it's been trained on how, on what to do. So now it's going to start running, and then it's going to look at the data it was provided and figure out what to do next. And this is where it really is a true agent because it makes the next decision. Now we've given it various tight parameters, but it, it makes the next decision. And what it's going to do is based on the traffic and the top channels, it's gonna go call one of these other scenarios that we've built. So I'm gonna open up the o uh, organic search version of this, and now this. Automation really goes through step by step, that process we defined earlier. So it's going to go in here and it's gonna pull context, it's gonna pull information about the persona from our Google Drive. We have a document with the persona detailed in it. It's gonna pull that information in. So now I have that available to me throughout the rest of this automation. And then it's gonna go through the process of pulling data from Google Analytics, uh, organic search data, current timeframe, last timeframe. Do the math. It's gonna go next level down. My first why source level data pull current per pull, past timeframes. Do the math. What are the top sources? And then it's going to use that as a filter for the next y. So say organic, say Google was the top reason that organic search is up. Then it's gonna say, okay, what were the top landing pages from Google? And then it's gonna compare current and past and do the math on that. And then once we get all of that down and boil down, now we have a detailed list of, okay, organic search was up. Thanks largely to Google and these three landing pages. Now we're gonna hand that information off to uh, uh, the first AI. Part of this whole this, this specific channel, and it's gonna summarize and analyze it. So it's gonna look at it, okay, what does it mean? Uh, dissect it a little bit, summarize what happened, and then we're gonna save that information for later, excuse me. And then we're gonna do the same thing, but with Google Search Console. So we're gonna go in and pull data, process it, go a layer level, deeper process that go another level deeper. Uh, all pulling just out of Google Search Console. We're not comparing it or doing anything with Google Analytics yet because we gotta keep it isolated in the channel until we can summarize it in more broader terms that can then be compared and contrasted. So we do that. We use AI again to take all that data, analyze it, what does it mean, summarize it, and then save that information.

Isar Meitis

Again, just to clarify to people, when Keith is saying, give it to AI to analyze, it just sends a prompt plus the data to. GPT or Gemini or clo, it doesn't matter. Basically saying, here's what I want you to analyze. 1, 2, 3, 4, 5. This is what I'm trying to figure out. Here's some additional information. Here are the actual pieces of data that I've collected, uh, for you to analyze. And then it just analyzes it in the way you define it. In the prompt.

Keith Moehring

Yeah. So like, I'll give you a quick example of kind of what the Google Search Council want is like. You give it a role, you're a lead marketing data analyst for X company and put in the company description, uh, here's your mission, here's what you're designed to do. Here's all the data you need. Here's your analysis rules, here's what to think about, here's how to look at it, where here's what we're trying to come out with, and then here's what your output needs to look like. It needs to have a recap, it needs to have a diagnosis. And then the primary drivers, any other observation, and then some very specific things on what not to return. So very much like how you would use chat PT or Gemini or whatever system you're using. Um, but it's all kind of baked into one and we've refined this over and over and over to get, make sure that the output is exactly what we are looking for.

Isar Meitis

Yeah.

Keith Moehring

So again, there not, there, there is no, there is definitely a level of trial and error with this as you build these. Uh, but what's nice is you can read the review report and then jump in and tweak this, tweak that, and then all of a sudden you're starting to get what you really want.

Isar Meitis

Yeah.

Keith Moehring

So we, we've got, in this case, we've got. GA four data. We got Google Search Council data. Now it's time to do the full organic channel analysis. And so same thing here. We're gonna send it off to chat GPT with a very specific set of instructions, a role, here's your goal, here's your inputs, here's what we're looking for. Here's how to analyze and review the data you've been provided. Here's some context of how it was gathered. And then ultimately, here's the output. Here's what we're looking for. We're looking for a recap, uh, why it changed, what it means, and then some constraints. Once we have that, the next step of this is we have another AI tool designed to be a strategist. We do, in this case, when we tried this and it, it didn't work out well where we tried to make the analyst and strategist into one AI in one AI call. And the problem with that is if you give it too many roles and too many tasks simultaneously. The reports start to get, things start to get a little convoluted in there. So give it one specific task, one specific, uh, operation, and then create a second one and then feed that into the next one. With the context it needs and the instructions it needs, the outputs become a lot more, uh, consistent and predictable. So the strategist though, it's designed to take that information in this case, uh, review it, and then your job is to come up with three new content ideas for us to write based on how organic search performed. And then we take that strategist mentality and then social Okay. Come up with. Three new social posts to write referral traffic. Uh, what are ways to further link building to drive more of this kind of quality traffic, that kind of thing. Channel specific recommendations based on the analyst data that was generated. So it goes through that. It'll identify what channels to, to process, and then once it's gone through all of those steps, it's identified, okay, the, the top channels and it's analyzed those. We got a summary of all of that. The final step is the agent will then call the analyst email report. And this one is designed to take the full report, format it in a way that will work with Gmail.'cause it comes out of Gmail. Um. We also format some additional data that we're gonna layer in here. And then the email gets sent off through an, uh, Gmail integration and it comes from an email that I've got set up specifically for this role. And then, uh, it'll reply back to the agent saying, I'm done. The agent knows as soon as that happens to shut down and it's over. So that's really, I mean, that's, it's, it's, it is, it seems like a lot, but if you look at it and compare it to kind of that process document that we had outlined, it is almost word for word operation line. Operation line, operation line. So, um, and it's, and it's fun. I mean, this is, this is the stuff I really enjoy doing is like getting into the weeds with this and walk, watching it all come alive.'cause it is cool when it actually starts generating reports like, oh man, I didn't even, I didn't even think of that. Or I would've never dug that deep to figure out that's what the problem was.

Isar Meitis

I, I love this. I, I want to touch on a few important points. One is a quick summary, right? We touched on a lot of details in the weeds. Let's go back to 30,000 foot. We're trying to figure out a problem in the business, right? And we have data sources that can give us information about it, but nobody has the time, the bandwidth, uh, or, or, or you do have the time, but you'd rather invest the time in other things. Like Keith said, he used to do it on his own. Uh, what you can do right now is you can go to the data sources, whatever they are, again, E-R-P-C-R-M, accounting system, uh, marketing platforms, third party solutions like Google Analytics, et cetera, et cetera. And you can get the data. You then explain to the AI what to look for in the data. Or you do some initial math before ai, uh, to give you some additional, uh, data points. You give it to AI with a set of instructions, say. This is how you look through this data. These are the things you're looking for. These are the cues. This is how you do the math, uh, this is what it may mean. And then you collect all of those analysis and you create a summary report with another ai. And that's another very, very important point that I agree with a hundred percent. You want to be very granular with the tasks you give every ai Let it focus on something and do that one thing very, very well. And kinda like think of the jack of all trades, but the master of none. You don't want that, you want the masters of, of every single trade to work on that specific aspect of the work. And then one thing to just put it all together, which that's its special trade. It knows how to put it all together. And so if you think about it in this granular way and you follow the steps of the process, like he said, you will end up with extremely detailed, well analyzed, easy to read report and easy to read is, is. Your choice, right? You will tell it how to write the report, how detailed, how long in what format. Like one of the things that I'm doing for the workshop that I'm doing this week, it actually generates, I have three different agents generating three different reports. One of them generates an Excel file with multiple tabs that are basically subsets of the data that are answering different questions. Another one is generating a PowerPoint presentation, and the third one is generating a written, uh, report that is the longest, most detailed and with the most amount of explanations kind of report. Uh, but you, you can build it in whatever, whatever way you want. The other thing that I will say that that is true as of right now, the day we're recording this, which may not be accurate, the time this comes out, but, but I'll say that anyway. Right now, the tool that is the best at formatting really amazing documents, reports, Excel files, PowerPoints, and so on, is clawed. It is. Mind blowingly good at doing that. And if you give it, like Keith said, here's what I want you to do, but you also give it, here's the structure, formatting guidelines, brand guidelines, and so on. It, it will give you a document you don't have to touch and you can send to your clients. It's that good with your logo in the header and the footer with page numbers, with the right colors, with like literally everything you need. And so that's just, if you want icing, icing on the cake, uh, to be able to create the reports just to also look better and not just have the right content. Uh, but overall I absolutely love this process. I think it's brilliant. And I think you touched on a lot of really, really great points. Um, any final words kind of summary, words of wisdom after doing this several times?

Keith Moehring

Yeah, so like two points. Uh, real quick, the, what you mentioned about the beautiful documents and all that stuff, the other thing that is kind of an underappreciated, uh, uh, flexibility thing that you can do with these, these make automations is you can create templates and insert variables within those templates. So if I have a PowerPoint where I want the analysis dropped into this specific section, I just have to put two curly cues and name it. And then within make, I can identify that specific text and drop that block right there. So I could drop this number here, this number here, and then it's all formatted, everything looks perfect, exactly how you want it,'cause it just drops the text into the template. Um, and then the other, the other quick note is this could just be the start of a larger process. So you have three new content ideas that it generates. Use automation to feed those into an editorial calendar. And then when those are fed in, another automation takes over and starts writing out the content and then uploads it directly. You see, I mean, you could start tying these things together in ways that you are automating not only the report, but Monday morning. I also have five new blog posts that I can review and publish within 10 minutes that are driven by the data that the analytics report's coming from. So that's where this stuff really becomes truly like magical is when you pull that stuff all together.

Isar Meitis

Yeah. Thank you for bringing this up. I, what I tell people, when you start looking at all these tools today with agents and automations and ai, and the combination of all of them is the way you need to think about it is the bottlenecks in your business. What is the biggest bottleneck or bottlenecks that you have right now that is preventing your business from growing? And go and solve that with an AI automation, which will lead to another bottleneck. Now, okay, now, now I know what to write about. Now I need to write blog posts. Now I know what I'm short of in inventory. How do I make like what, whatever the thing is, do you now create another bottleneck? Go and solve that. And in some cases, you get to the points where AI doesn't know how to solve it, or summation doesn't know how to solve it. I'll give you a great example. A lot of my clients are in the smart home integration space. They help home builders, build smart homes, uh, AV systems, smart curtains, you know, uh, air filters that can send stuff, like all this kind of stuff. You, you actually need a, a bunch of people in, in a house, on a construction site, actually doing the work. AI doesn't know how to solve that, but it knows how to help you, how to hire better people. It knows how to help you to build better procedures for these people so they can do more jobs in a certain amount of time. It knows how to help you, how to make sure that they actually have the right equipment on the truck before they leave the, uh, the warehouse to go to the job site to make sure they don't have to come back. Like even the stuff that it can solve, there's a lot of touch points that AI together with automation can help you solve. So, as a very high level summary to this really amazing session, think about bottlenecks in your business and start solving for these one by one by one with these AI automations. Uh, and then you can just grow the throughput of your business without increasing the resources that you have right now.

Keith Moehring

Mm-hmm.

Isar Meitis

Keith, this was spectacular as expected. I really love every time I see what you're doing, if people wanna work with you, hire you, follow you, learn more about what you do, what are the best ways to do that?

Keith Moehring

Best way is, uh, through the website. It's, uh, just l two digital.com. Uh, you can also find me on LinkedIn. I'd love to connect with whomever. It's just at Keith Mooring, uh, on LinkedIn. So, uh, those are two, the, the two primary ways. So I, yeah, I'd love to connect with whomever.

Isar Meitis

Awesome. Thank you so much. This was fantastic. Uh, keep on sharing this amazing stuff that you do. I really appreciate you.

Keith Moehring

Thank you. I appreciate it.

Head Shot

Isar Meitis

Host