So What? How to Extract Insights from Qualitative Data With AI

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? The Trust Insights weekly livestream, you’ll learn how to extract insights from qualitative data with AI. Discover how to identify patterns and categories from large bodies of text using generative AI. Understand how to apply these insights to validate your marketing strategies and improve product offerings. Learn how to leverage qualitative data with AI to make informed decisions for your business.

Watch the video here:

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

In this episode you’ll learn:

• The challenges of finding and extracting qualitative data (and how generative AI can help)
• What methods are appropriate for qualitative data analysis
• How to use generative AI to build robust qualitative data reporting

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 – 00:25
Well, hey, everyone. Happy Thursday. Welcome to So What? The Marketing Analytics and Insights Live Show. I am Katie, holding up Chris and John.

John Wall – 00:35
There you go. Yeah, and down.

Christopher Penn – 00:37
Exactly. And down. She’s back after kayaking over Niagara Falls.

Katie Robbert – 00:41
I’m back. The biggest thing that we both got from vacation was cold, so I’m still trying to eradicate that. Today, we’re talking about how to extract insights from qualitative data with AI. Now, you might immediately think, “Oh, so we’re talking about market research data.” That’s one of many kinds of qualitative data. Analyzing qualitative data is a specialty up until now, and I say it’s a specialty because it’s not as easy as putting everything into a spreadsheet and writing a formula and making a chart out of it. That’s what you do with quantitative data.

Your numeric data, your qualitative data is your data that is more freeform. It’s sentences, it’s feedback, it’s opinions, it’s survey responses, and that tends to be very challenging if it’s customer service feedback.

Katie Robbert – 01:46
And the responses are not normalized in a way that everybody is responding the exact same way so that you can have one-to-one responses and then just sort of cluster them. So there’s a lot of different things that you can do with qualitative data. What we want to do today is show you some examples of how to extract insights from qualitative data using generative AI. Chris, where should we start?

Christopher Penn – 02:15
Well, I suppose we should start with—huge surprise—the 5P framework. You can find out more about that in our new AI strategy course, the Trust Insights AI Strategy Course. Katie talks about that for about two hours. Why would we want to do this? What’s our purpose, Katie?

Katie Robbert – 02:37
So, our purpose? It can go a couple of different directions, but for the sake of the show, let’s say the purpose is we want to understand what our customers or our audience is saying about us on a social platform. Or we want to understand the pain points of our customers—what they are reporting in our customer service portal—so that we can fix them, those kinds of things. Essentially, we want to understand what people are saying so that we can resolve it, so that we can take some action with it.

Christopher Penn – 03:18
Got it. Okay, let’s do this. Let’s take a very concrete problem. We want to sell more of our AI strategy course. We want to sell more of this. So, how would we go about finding information about this? We know what the course is about. We know it is meant for leaders or people who want to be leaders in their companies around AI and setting strategy. Logically, we should probably understand who the “who” that might be and where they might be talking about this.

Now, the good news is we’ve done a lot of this work already, so the second P, people, has largely been solved. We have solved that with our ICPs that we talked about in a past episode of the live stream when we built a sales playbook that contains our different ICPs.

Christopher Penn – 04:15
So, our first step in this process would be to say, “Where would our ICP be asking about this?” Let’s go over into Google Gemini, and I’m going to attach our sales playbook, which is this. We just want this one here. And we’re going to say something pretty straightforward. We’re going to say, “Based on our sales playbook and the ideal customer profiles in it, we want to sell more of our AI strategy course, which is called the AI-Ready Strategist. It’s targeted toward leaders and people who want to be leaders in generative AI at their companies. To do this, we need to understand where these people might be talking about their AI challenges. What forums on Reddit, for example, would be places where our ICP might be having those conversations? Make a list.” Now, from this, we’re going to start with our ICP.

Christopher Penn – 05:17
We’re going to say, “Let’s just see if we can, if there even is any credible place on Reddit where this might be occurring.”

Katie Robbert – 05:25
And so, John, as our chief statistician and head of business development—AKA you’re the sales guy—if I said to you, John, “We need to sell more of the AI strategy course,” where do we start? AI aside, where would you, as the human, start to look?

John Wall – 05:46
That’s a huge challenge. It’s the kind of unconscious, incompetent thing of the people that know it, don’t know that we’re out there. So, there’s a huge awareness gap. It’s easy. Normally, the way we would go with that is just, let’s just start with the internal list. We would go with the people that we already have had some kind of contact and already know who we are. That’s kind of the easy hunting.

John Wall – 06:08
So, yeah, if there’s any way to do some kind of mining of what’s out there on Reddit or LinkedIn or whatever to find some new possible leads, that’s a great way to do some prospecting. Anything we could get some leverage on via automation would be fantastic because we have found no easy answer to that problem, aside from, “Let’s continue to do content like this and speaking and stay out in front of everyone.” Our reputation does the work. But, yeah, we’ve never found a way to kind of go mining for who those people are that are trying to get to this stuff.

Christopher Penn – 06:43
Yep. So, we have a list of like nine different subreddits. For those who are not familiar with Reddit, it has individual channels inside of it. If you’re familiar with Slack and you see channels in Slack, Reddit has channels—they’re called subreddits. These are each self-contained communities that are moderated by different people. It’s all people-based. Some subreddits are great, some subreddits are garbage fires, dumpsters on fire. It all depends on the people who are moderating each one. But the big ones tend to be moderated pretty well. That’s why they’re big, because they are able to sustain their audience.

Christopher Penn – 07:23
If you are a registered developer with Reddit, which you can do for free, you can get an API key that will allow you to download data from Reddit for free within reasonable limits. You can’t just download all of Reddit, and your computer would explode if you tried. But you can say, “Give me the last 90 days of posts and comments from the Artificial Intelligence subreddit or the Marketing subreddit.” Our next step would be to go and grab that data. Now, this is the hard part, because you have to be good enough at generative AI to have Gemini or ChatGPT or Claude, or whoever you use, write you the code to do this. You don’t have to code it yourself, but you have to be able to articulate.

Christopher Penn – 08:21
I want you to build me an app in Python that will use my Reddit API key to download the posts and comments from the subreddit and put it in a form I can use.

Katie Robbert – 08:32
Is this where you would use a tool like Google Colab to write the code?

Christopher Penn – 08:39
You could. You can do it straight up in regular Gemini or ChatGPT as well.

Katie Robbert – 08:45
Okay.

Christopher Penn – 08:46
Or you could do it in a coding environment. All the major services have dedicated coding tools. Gemini’s is called Code Assist. OpenAI’s is called Codex. Claude’s is called Claude Code. Not particularly imaginative, but there you have it. All of them could. You could do that, and even in regular Gemini, if I was to start a new chat here.

Katie Robbert – 09:06
So, it doesn’t matter where you’re doing it, as long as you know what instructions to give.

Christopher Penn – 09:12
Exactly. Let’s do a very quick example. Today, we’re going to write a Python script in Python 3.11 that’s going to be a command-line application. It’s going to address the Reddit API. We’re going to need a way to store our Reddit credentials and our API key to address the Reddit API. At the top of the script, we’re going to have configuration variables for things like which subreddit we want to analyze and how many days of posts and comments we want to extract. We also want to store this information in a couple of different formats. We want to store it in a SQLite database so that we don’t have to keep re-grabbing the same information over again. We also want to store it in JSON files to make it easy for generative AI to read.

Christopher Penn – 09:51
We also want to have the ability to pick up where it left off, so it’s not trying to re-crawl all the time. It has to obey backoffs and 429 errors using exponential backoffs, so that we don’t get banned for hitting the API too hard in the Canvas. First, develop a product requirements document for how we might build this app. Based on that very large foaming at the mouth, we want to build a PRD, a product requirements document, because you generally shouldn’t start coding without one.

Katie Robbert – 10:26
You shouldn’t. But as we know, people do.

John Wall – 10:28
People do.

Christopher Penn – 10:29
Exactly.

Katie Robbert – 10:30
I just want to point out, for those who aren’t aware, a lot of these generative AI tools now have the ability for speech-to-text, and that’s what Chris is doing. So Chris, rather than having us watch him type and backspace and fix spelling errors—which is what you would actually be watching me do—if you can see in the, “What do you want to build?” that chat dialogue, there’s now a microphone, and you can click on that and do what Chris is doing, which is Chris’s favorite thing, and just yell into the void.

Christopher Penn – 11:04
Exactly. Now that we’ve done this requirements document, we can now say in a new canvas, “Build the Python script,” assuming the requirements document is good, and we skipped over a whole bunch on purpose, like going back and forth, gathering requirements properly and stuff, which we’ve done on previous episodes.

Katie Robbert – 11:25
Yeah, human intervention, making sure that things are correct. “What did I miss? What about testing?” Those kinds of things should all be in there.

Christopher Penn – 11:32
Exactly. So now it’s going to try and build the actual Python code for us to do this task. Now, at this point, we’re going to leave this behind because it’s not a great use of time to just watch it code. But this is where you would start, and you would go back and forth with the AI, making corrections and things. Ultimately, what you should end up with at the end of the process is something that looks like a big folder of stuff. Each of these folders, files that you see on screen here, is a database of Reddit posts. So, let’s open up one of these here, let’s see, where are we? Let’s open up online learning, actually online courses. This is a database. Here are the comments from different Redditors, here are the posts from different Redditors.

Christopher Penn – 12:21
And now we have qualitative data. We have a lot of qualitative data in here. But what we’d want to do next is figure out, well, what does it say?

Katie Robbert – 12:32
It’s a great question. Well, I think. No, but jokes aside, I feel like that’s where a lot of people get stopped, because it is. It can be so much information. Trying to read—it’s one thing to read through, it’s another thing to read through it and also identify patterns and categories. That’s where generative AI is going to be really helpful, because you can read through a big body of text, which is essentially what you have. But then to make sense of it and to do something with it is the difficult part.

Christopher Penn – 13:07
Exactly. So, in my folder, because, remember, I said I want JSON exports of this data, which means it’s something that a machine can read. If we look, if I just pop open one of these, you can see there are the Reddit comments, IDs, titles, authors, and it’s all very nicely structured so that in AI, a large language model can read the language and see all the metadata in here as well, but not get confused about, “Is this person’s name? Is it a comment they made? What is it?” By having this very structured data format, generative AI is like, “Great, I know what to do with this. I know how to read this.”

Katie Robbert – 13:46
And I feel like that’s a pro tip you’ve given before: how to structure the data correctly. I believe, if I’m not mistaken, Chris, you gave that in the Trust Insights newsletter in one of the data diaries, which you can subscribe at TrustInsights.ai newsletter, because structuring the data correctly is more than half the battle.

Christopher Penn – 14:13
Exactly. And so, the best. All these files, you’ll note, these files are like a megabyte each. These are fairly large files. If I were to zip into one, it’s about 150,000 words each file. That’s a lot. That is going to be more than any one general chat window is going to be able to handle. However, the place that can handle this is NotebookLM. NotebookLM is ideally suited for having, being able to process these large plain text files, because that’s what a JSON file is, just a plain text file. It’s nothing fancy. So if we were to go in here, and based on—we remember what Gemini said about the forums and things that our target market might be using—they might be, for example, interested in the marketing subreddit.

Christopher Penn – 15:07
Because a lot of our audience is marketers for AI stuff. Specifically, they might be very interested in the artificial intelligence forums. So, I’m going to add the artificial intelligence forums in these. Just more. Whoops, I messed up. There we go. Let’s add that.

Katie Robbert – 15:28
Try that again and remind me, remind us: there’s a limit to the number of files that you can—or sources, rather—because it could be any kind of file sources that you can include in a NotebookLM notebook.

Christopher Penn – 15:47
That’s correct. For the free version, it is 50 sources at a time. For the paid version, including the Google Workspace versions, 300.

Katie Robbert – 15:55
Oh, for some reason, I thought it was like 20.

Christopher Penn – 16:00
A long time ago. It was 20. Back in the bad old days, when it was like, “Oh, we’re just trying to figure out if this thing even works.”

Katie Robbert – 16:07
Gotcha.

Christopher Penn – 16:08
Now that it’s an actual production product, it is considerably more generous. The other thing we probably want to have in here as a source, just for reference, would be our sales playbook. That would make logical sense. If we want to be able to have a conversation about what our ICP is, what they care about, we want to have that background data in here. So, we can now go ahead.

Katie Robbert – 16:32
I have a question, though. If this is for selling the course, wouldn’t you have background information about the course in here as well? Or, no. Like, I guess I’m—maybe I’m jumping ahead. Let me know that.

Christopher Penn – 16:47
No, that’s perfectly fine. If you wanted to have the course information in here as well, that would be good. I would say maybe we should add that a little bit later on because we want to try and get as much of the original words of the people first before I gotcha. But we’re going to condition—we’ll definitely do some prompt conditioning along those lines. So, our next step would be to say we want to understand from the Reddit forum feedback that we’ve loaded in here what people are talking about in terms of their frustrations about AI strategy, particularly AI and marketing.

Christopher Penn – 17:27
But AI in business in general, based on the conversations that I’ve provided and the ideal customer profiles in the sales playbook that I’ve provided, explain what the frustrations of leaders in organizations are about AI, specifically about AI strategy, and what they’re doing to address those frustrations. We’ll give you a format that looks like this. I’ve got this first part of the prompt here, and then we’re going to give it a very simple outline. The outline is basically just “Make it look like this: What is the frustration? Why is this a frustration? How is somebody working around it?” This is now going to go into all the sources we’ve uploaded and try to pull out the qualitative insights from what real people are saying.

Christopher Penn – 18:21
Now, one of the cautions we have to have here: the people who are on Reddit may not be our ICP. They may not be our customers. So, we have to take that feedback with a grain of salt.

Katie Robbert – 18:34
Well, because we don’t actually know who these people are. A lot of people are using handles. A lot of people like to incite conversations that are derisive, or they like to cause problems. So, I would say I think it’s a good starting point. I don’t think it’s something that we would use as the end-all, be-all. I wouldn’t say to John, “Okay, I have a list of pain points that we need to address with our services, go out and start selling these exact things.” But I think it’s good directionally.

Christopher Penn – 19:08
And that is exactly where we’re going to go, because you’ve identified a pretty clear direction that you want to head. We want to know if it’s our ICP. So, here comes the list of frustrations, and let’s take a look at these really quickly, and let me hide our side panels so that we have a bit more viewing room here. “Lack of clear AI strategy and overwhelm.” “Leaders often overwhelmed by the rapid pace of AI frequency.” “High frustration.” “Difficulty demonstrating AI ROI and measuring impact.” “AI reliability, accuracy, context, and ethical concerns.” “Skill gaps and talent upskilling in AI data.” “Internal processing efficiencies and conflicting stakeholder feedback.” So, those are the things that it identified from the conversations. Like, “Yeah, this is what people are pissed at about their AI strategy.”

Christopher Penn – 20:01
Now, what we would do is go back to our regular Gemini, attach our sales playbook, and say from the following qualitative information that we received from Reddit forums, “Analyze from the perspective of our ideal customer profile interested in buying our AI strategy course called the AI Ready Strategist. Re-rank all of the complaints in the most likely probability that they would be relevant to our ICP, because what people say on Reddit may not reflect what our ICP actually cares about. Present your results in the exact same format as the form results, but re-scored and re-ranked by our ICP’s relevance.” So, we’re going to put that in, and then we just copy and paste from NotebookLM. You might say, “Well, why can’t you do this in NotebookLM at this point?” We don’t want all those other conversations in there.

Christopher Penn – 21:03
We don’t want the context window being polluted by all this conversation. So, we’re now just taking that, that extract plus the original sales playbook, and have it play that back. Thinking about now, with the data being much more heavily weighted towards our sales playbook, this is now the time for barbecue recipes.

Katie Robbert – 21:24
Yeah. Well, John, I always—when we’re talking about the sales playbook and sales in general, my first question is always, “Does this jive with the process that you use or that you were taught? How jarring is it to be using generative AI for your prospecting, when, for the past however many years that you’ve been doing it, it just hasn’t been an option?”

John Wall – 21:53
Yeah. Well, that’s the thing. This is completely revolutionary. It’s just a matter of time before people figure out how to really take full advantage of it. This is the kind of thing that a C-level person would just read 10 articles and then make up their whole opinion on how it’s all going to go. The fact that you’re digesting hundreds of thousands of posts and actually getting an accurate picture of what the world is like—that’s a whole different world, because yes, so many sales decisions in the past, as far as prospecting, people won’t admit it, but it’s just going solely on gut. They’re just taking a random sample of five and going from there.

John Wall – 22:34
And so that’s why there are so many crashes and burns. Yeah, that process. And so this can—yeah—can change everything.

Christopher Penn – 22:43
So we have our results. Now we have from our ICP, Strategic Sarah, which is one of the ICPs in our sales playbook, the re-ranked frequency by the pain points that our ICP would have: lack of clear AI strategy and overwhelm.

Katie Robbert – 22:57
You know, we have a course for.

Christopher Penn – 22:58
That difficulty demonstrating ROI, reliability, accuracy and context, and ethical concerns, skill gaps and upselling, upskilling, and internal process inefficiencies. It’s interesting that those are in a slightly different order, but it makes sense because this is our ICP that’s talking to us now. So now we can go to the actual course page, right? If we go to our AI Ready—

Katie Robbert – 23:26
Strategist page. I will say I also, in case you are curious, Chris, for easy use, I created a NotebookLM about the course, which is in the client services, our internal account, to get all of this information.

Christopher Penn – 23:48
Excellent. So, I’m just going to take the publicly available data, which is what’s on the outside here. I’m going to feed this in, going to say, “Here are the contents of the course landing page. How well does this landing page address the frustrations of Strategic Sarah as outlined? If Strategic Sarah read this landing page, rate it on a scale of 0 to 10 about how relevant it would be to her frustrations and how compelling it would be for her to take this course.” So, we’ll put that in, and we’ll put in the landing page text, and we’re going to see what strategic sales. Wow, that’s really.

Katie Robbert – 24:30
Well, I didn’t know I was getting graded today, man. But no. So here’s the thing, and this is what generative AI doesn’t pick up on: I would actually argue, “Oh, I got a nine out of ten. Forget it. None of my arguments are valid.” I was getting so ready to go on the defensive and defend my choices, but it’s a nine out of ten. We’re good. What I was going to say, in all honesty, is what generative AI or what you couldn’t pick up on and put in—but we could get—was the transcript from the overview video that’s on the landing page. That explains a lot of these pieces. That’s what’s not in here, but we could add that in.

Christopher Penn – 25:14
Exactly. So it says, “Overall rank, relevance, 9 out of 10. Compelling, 9 out of 10. How the landing page addresses it: lack of clear AI strategy, the TRIPS framework, difficulty administering ROI, the ROI calculator, AI reliability, the 6E framework, skill gaps, driving adoption, internal process efficiency, the 5P framework, and stuff.” So what this tells us is from qualitative conversations that we picked up from Reddit—no influence of our own—through a NotebookLM distillate through our ICP to our landing page, there’s very little we need to do to make changes, which answers a critical question for us as marketers: if you have any product that’s not selling, there are a million reasons why it might not be. We’ve all had those conversations: “Is it the economy? Is it price too high?”

Christopher Penn – 26:11
Is the value not clear? What this tells us? This knocks some stuff off the table. This says the course is relevant. It’s clear to this person who is our ICP that we validated through data in the past. The material in the course is not out of alignment with what they want. So now we can take that whole lingering doubt, like, “Was this any good?” It’s off the table. It is good. It’s a very clear fit.

Katie Robbert – 26:38
Thank goodness. I really—I was gearing myself up to get all kinds of defensive. So, I’m glad that I don’t have to do that now. That takes a lot of energy.

Christopher Penn – 26:49
It does. Now here’s the question: what if it’s price? How would we know?

Katie Robbert – 27:00
John, sales guy.

John Wall – 27:03
Takes a hard line on that. The only way you do that is your wins and losses—actual people with actual money. You can do studies and figure out what the odds are out there, or compare it to other things that are out there and sold, but there’s no substitute for actually beating on it. So I’m interested if Chris has got a take on this, if he’s got an angle that we can run with this.

Katie Robbert – 27:26
Oh, I’m sure he does.

Christopher Penn – 27:28
And of course I do.

Katie Robbert – 27:29
Of course you do. And, okay, so is this where I get defensive?

Christopher Penn – 27:33
No, this is where we go back to the well, and we go back to NotebookLM. Let’s start a new notebook. Now, instead of having the marketing stuff, let’s take in stuff from the online courses forums for people who make courses, the e-learning forums for the people who are designing courses and things. We’re going to start with a very simple question. We’re going to say, “Based on the Reddit conversations from these different forums, what are course creators discussing and debating about pricing? How do they set pricing? How do they determine what a fair market value for their course is, and what strategies do they seem to follow?”

Christopher Penn – 28:25
So let’s start with that and let’s just see what the conversations are, because this is about 90 days worth of conversations—that’s what I have my script set to download—because I feel like anything older than that, given the economy and the macro perspective, things change really fast. So, we want to figure out, “Is it a case where we’re out of alignment with pricing, with what the rest of the industry is doing?” And if so, “Can we justify it?” We see “impact of price on student commitment and completion rates.” “A theory that a course priced too cheaply might lead to lower student commitment compared to more expensive degrees.” Let’s make this a little bit bigger here. “Low payouts from marketplaces like Udemy, where one user reported making an average of $4 a sale and questioning the product’s worth.”

Christopher Penn – 29:16
“Pre-price hikes and hidden fees from SaaS platforms.” “Perceived pricing plans.” “The overall expense of high-tier LMS solutions.” “How they set pricing, focusing on business outcomes, market research, and competitor analysis.” “Direct user engagement.” “Cost of platform features.” “Perceived value for content presentation.” “Tiered offers and bundling.” “Leveraging free content and trials.” “Platform choices and hosting models.” “Diverse marketing and sales approaches.” “Enhancing engagement to justify value,” and stuff. So we have a variety of different ways to look at this issue from what other course creators are saying. Let’s go hit copy, go back, and hit paste, and say, “Let’s say we want to sell more of our course. It is currently priced at a premium value of $1499-1000.” I knew it.

Katie Robbert – 30:05
Premium.

Christopher Penn – 30:09
Based on this, we know that there are certain segments of the audience that will not be able to afford this. What we want to do is understand Strategic Sarah and what she might be willing to pay, and then some of the feedback from other course creators about the challenges they’ve run into in course pricing, and try to reconcile the two perspectives for the purposes of understanding how we price this course to maximize the overall revenue generated from it. So we put in that prompt, then we add in our Reddit conversations, and we see what generative AI comes up with to see if there’s a middle ground.

Katie Robbert – 30:51
And so I can tell you a little bit of my thinking about why it’s priced at $14.99, because I’ve heard, “Oh, that’s a lot of money,” or, “Is it fluffy, and am I going to get something for it?” If you are new, if you haven’t heard of this course, in the course there are over 20 downloads, and by that, it’s frameworks, exercises, checklists, templates, plus there’s about eight hours of instruction walking through with exercises. If I did basic, super basic, back-of-the-envelope math, and I said if I take one download, like the ROI calculator, and I sell that for $99—which is a really inexpensive price—or even like $59…

Katie Robbert – 31:42
If I take $59 as a price point, or $99 as a price point for each individual download, multiply that by how many downloads, that’s way more than the price point of the course. So, in theory, you’re actually getting everything bundled at a discount with instruction versus buying everything individually. So that was my logic for coming up with the pricing, in addition to Chris, the some of the research that you did internally in terms of course pricing, those kinds of things. It wasn’t an arbitrary, “I’m just going to sell it for this much.” There was thought that went into why it’s priced the way it is.

Christopher Penn – 32:28
Exactly. So, Gemini says, “Based on the profile of Strategic Sarah and the answers to other course creators, the current price point is well-positioned for the core target audience. It’s insufficient on its own to maximize total revenue. Take a tiered pricing structure.” Strategic Sarah’s perspective: “Why it works: it’s priced based on business outcomes.” “What course creators recommend for B2B sales: for a corporate budget sub-$2,000 course, it’s a standard professional development expense that often doesn’t require complex proof like a $10,000 consulting engagement, which we also do.” “Perception of quality: the price premium signals a premium strategic product. A cheaper course might be perceived as tactical rather than strategic, making it less attractive to her. The price for Strategic Sarah is not a barrier. Do not lower the price for this ICP.”

Katie Robbert – 33:15
Thank goodness.

Christopher Penn – 33:17
Now, when we look at the course creators’ tiered offering and bundling, this is the key to unlocking new revenue: different access levels or extension courses. So, the AI Ready Strategist—this is the flagship product—the Strategic AI Toolkit, which essentially is taking just pieces of the course and unbundling it into its own thing, or having accelerators and bundles, like, “Hey, get five seats for the price of four,” etc. So, it goes through and says, “Here’s what we know: the market is working. In the market, based on Reddit conversations, but reconciled with the fact that this is squarely targeted at our ICP, which makes sense.”

Katie Robbert – 33:59
And so, if you go to that second option of unbundling and rebundling things, that’s really interesting. The way I had been looking at it was I would take the individual modules in the chapters and record, re-record those as mini-courses, mini-lessons. Actually, this is much in terms of overhead and resources; this is way easier, and something I could probably put together tomorrow. So, if we want to sell just the toolkit, we could absolutely do that. Just for context, for those who aren’t aware, this all started with our AI toolkit, which is a white paper download. But it doesn’t go into the details of the templates that this would. This Strategic AI toolkit—that’s actually really smart. I like that. And stay tuned, because that’s coming.

Christopher Penn – 34:59
So, I have a question about that because I’ve been wondering about this.

Katie Robbert – 35:02
Yeah.

Christopher Penn – 35:03
Here is a copy of our lead magnet, the AI Ready Marketer, which is a precursor to the course that contains a lot of these materials that you targeted for the Strategic AI Toolkit. “Should we stop giving this away as a lead magnet? Instead, charge a much lower price for it, because it does contain substantially similar content to the actual course.” Let’s see what it says.

Katie Robbert – 35:35
I would argue that this is still valid because it doesn’t actually create the templates. It just talks through the frameworks themselves, whereas the 399 bundle would actually be the templates themselves. In the free download, you have to do the work. In the 399, I’ve already done the work for you.

Christopher Penn – 35:59
I’m wondering, though, and this is just my personal curiosity, whether we should or not. It says no, you should absolutely not stop giving away the AI Marketing Strategy because a free lead magnet’s strategic value. Charging for this kit would be like a high-end restaurant starting to charge for its menu. I think I love that analogy. The value of the lead is greater than a small sale to best qualify. Someone who downloads a 26-page PDF is not a casual browser. They are likely your ICP. It builds brand trust and authority. So, the free kit gives them the “what.” The paid toolkit gives them the “how.” Reposition the free Mac lead magnet. Keep it exactly as it is, call it the DIY toolkit. Define the paid product, which is the accelerator pack. And then that leads into the upsell for the full course.

Katie Robbert – 36:43
I like it. I could also see if somebody downloads or somebody purchases the—

Christopher Penn – 36:50
The accelerator pack.

Katie Robbert – 36:51
The accelerator pack. And then they’re like, “Okay, I want—I still want the course.” They could get a bit of a discount on the course because they got the accelerator pack. But, I mean, those are all things that we could work out in the background. But I hadn’t—I was thinking along these lines, but this is a different angle, and I really like this because it actually takes a lot of the burden off of me to recreate the wheel exactly.

Christopher Penn – 37:18
So, putting the course aside for a minute, what we’ve done today is identified from our ICP where the qualitative data that we might want lives. Then we’ve extracted it using code generated by generative AI to build the tooling. We need to actually grab the data and pull it in. We then have generative AI process that data. We put it into a system like NotebookLM, which is like a library to extract insights from the data, because the data is just too large for a regular AI model to handle. We take those insights, we put them back into regular Gemini against our ideal customer profile, and say, “Validate these insights: Are they real? Are they fake? Do they not align with what we know to be true about our ideal customer?”

Christopher Penn – 38:09
It came back and said, “Yes, these insights are relevant, but in a different order, because of the nature of the Strategic Sarah,” as she’s called in the playbook. Then we say, “Here’s the product we want to optimize based on the insights from conversations. Is this product aligned with the marketplace?” That’s where we got the tiered bundling and stuff like that. Then we went back, and we validated more pieces of our marketing strategy against both our ICP and the real-world conversations. What’s useful about this approach is that it is grounded in reality. We are not asking generative AI to make things up. We’re saying, “Here’s real data from real people, and here is our ICP from real people and real things, and here’s our real actual course with all the real information.”

Christopher Penn – 39:02
Create strategies, tactics, execution methods, and measurements of success.

Katie Robbert – 39:10
And if you want to get the course, go to Trust Insights AI Strategy Course. I recorded it before I got sick, so you don’t have to suffer through that.

Christopher Penn – 39:21
Exactly. So, that’s one of many ways to extract qualitative data. Now, there are other—there are so many other ways and so many other data sources. Let’s say you’re a company that maybe you’re a B2C company, and you have a restaurant, right? For an example, let me pull out Ostanya Barakata in Kiev, which is a very popular restaurant. It is half restaurant, half museum. You’ll notice on here that there are Google reviews. These are all the different reviews that have happened on the site. Guess what you can’t do from Google Business?

Katie Robbert – 40:02
Export the review data.

Christopher Penn – 40:04
Exactly. It sucks. It’s awful. You can’t do that. But what you can do is you can turn on your screen capture utility. I’m going to turn on—I’m using Camtasia Snagit. I’m just going to hit “Screen Record” and I’m going to click “Read,” and so on and so forth. I will just do a page or two of this. Not a ton, but can look here and read this. Most relevant views. Did I miss any there? No, no. Okay, let’s hit stop. Now, I’ve got qualitative data. So, I’m going to take this. I’m going to use Google’s AI Studio, just because I know it’s very reliable about this sort of thing. I’m going to take this video file and I’m going to drag and drop it straight into Gemini. I need to drag to my desktop first. There we go. And desktop.

Christopher Penn – 41:11
There’s my file. I’m going to say, “From this video, create a transcript of the reviews. The transcript must contain the reviewer’s name, the reviewer’s star rating, the estimated date of the review based on today’s date. Today’s date is September 11, 2025. The review text and the review sentiment. Produce your results in JSON format with each of these fields as a key and the appropriate corresponding value.” Let’s see. Did my thing go to sleep? Did it? It went to sleep. Let’s try it again. It turns out that the application I wrote to do that—if it doesn’t fall asleep, but eventually consumes all the available memory on my computer—this is a slight memory I haven’t fixed yet. “Transcribe the following reviews from this video.”

Christopher Penn – 42:02
The reviews include the author’s name, the review text, the date inferred and the star rating, and the sentiment of the review. “You’ll have to infer the sentiment and calculate the estimated date of the review based on today’s date. Today’s date is September 11, 2025. Return your results in JSON format with each key and value corresponding to the fields I specified,” and let’s go ahead and hit run.

Katie Robbert – 42:29
I think Google Reviews is a really good use case. The one that always brings to mind for me is if you have a customer service like support system, you get a lot of emails back, even if you don’t. We get emails back about the newsletter. We get emails from people who just want to tell us off. We get emails from people who want to tell us things that we’re doing well. All of that is qualitative data that we should be able to do something with, and I think that’s where we get stuck as well. It all just lives in my inbox. So, what do I do with it?

Christopher Penn – 43:06
Exactly. So, here we can see it has gone from a video of me scrolling into a JSON file that we can now work within generative AI. If we didn’t want to do that, we could say, “Now reformat the JSON as a CSV.” From here, we’re going to take that, reorder the data, and it’s going to put it into a CSV format, which you can then take into your Excel spreadsheet or your Google Sheets or the system of your choice. I like to start going with JSON first because it tends to perform better and have fewer errors. But you can try to skip straight to CSV files. It’s generally better to start with JSON first. And so now I’ve got my CSV file. I could just download this sucker and start working with it.

Christopher Penn – 43:57
If you are a SaaS company, and you care very much about Capterra, and maybe, let’s say you work for—let’s just pick ZipRecruiter. So, ZipRecruiter is on here. They’ve got these reviews: 4.2 stars, 11,000 reviews. You would follow the exact same process and just scroll through the pages of reviews and grab that data. Anything you can see as a human being, you can extract data from. If I was a product marketing manager, I didn’t have access to expensive social media management tools or review management tools. This is at least the way I would get to that data.

Katie Robbert – 44:44
What do you think, John? Where are you going to start, man?

John Wall – 44:47
I’m going to go get my free API key and pull down all the garbage I can from Reddit. I didn’t realize that it was just that easy. So, what are the limits? Is it like a gig a day or something?

Christopher Penn – 45:03
It’s about 10,000 posts and comments per day, per subreddit.

John Wall – 45:09
Per subreddit, really?

Christopher Penn – 45:10
Per subreddit.

John Wall – 45:11
Oh, wow. So that’s—

Christopher Penn – 45:12
But I think you’re limited, like 150 or 200 subreddits a day.

John Wall – 45:15
Oh, okay.

Christopher Penn – 45:16
All right.

John Wall – 45:16
So, and then what? Does it hit a paid tier or do they just shut you off?

Christopher Penn – 45:20
And they just shut you off. If you don’t respect the back-off after a certain number of retries, then your key gets canceled.

John Wall – 45:28
You’ll get completely—

Katie Robbert – 45:30
But I would imagine that’s a process that you could probably automate as well. Like, once a day, go pull some of the previous days, something so that you’re not hitting those limits.

Christopher Penn – 45:43
Yeah. The one we have for Trust Insights clients—because we do this for a few clients—runs every Monday and Friday. So, it just catches up for the last four days.

Katie Robbert – 45:51
That’s not bad. I like it.

Christopher Penn – 45:54
Yeah. So, that is qualitative data. Again, if you can see it, gen AI can process it. Then it’s up to you to figure out how to calibrate it against reality and ultimately how to make use of it to do something productive.

Katie Robbert – 46:08
So, I think this goes back to what we’ve talked about on previous episodes of the podcast and live stream: AI does not replace critical thinking. What you did, Chris, was piece together a lot of different things to come up with a full story as the human. Because you use critical thinking of, “All right, let me think about what I’m trying to do. Let me think about where that data could be, see it.” And so, those like you using your critical thinking skills—that’s sort of why we keep saying critical thinking is what you really need to refine versus worrying that AI is going to do the stuff.

Christopher Penn – 46:47
Exactly. And being thoughtful and methodical in your process.

Katie Robbert – 46:53
Yeah.

Christopher Penn – 46:54
All right. That is going to do it for this week’s show. Thanks for tuning in, and we’ll talk to you all 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 Trust Insights AI/TI Podcast and our weekly email newsletter at Trust Insights AI Newsletter. Got questions about what you saw in today’s episode? Join our free Analytics for Marketers Slack Group at Trust Insights 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: The Trust Insights Leadership Team The Woefully Incomplete Book of Generative AI Empower Your Marketing With Private Social Media Communities Contact Trust Insights Disclosure of AI and Protection of Copyright
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.