livestream header

So What? Can AI solve this problem?

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 week’s episode of So What? we focus on Core Web Vitals in Search Console. We walk through which marketing challenges AI can and cannot solve and how to determine if AI is the right choice. Catch the replay here:

So What? Can AI solve this problem?


In this episode you’ll learn: 

  • Which marketing challenges AI can solve
  • Which ones AI can’t solve
  • How to determine if AI is the right choice

Upcoming Episodes:

  • TBD

Have a question or topic you’d like to see us cover? Reach out here:

AI-Generated Transcript:

Unknown Speaker 0:26
Boy howdy, happy Thursday. Welcome to so what the marketing analytics and insights live show. I’m Katie, joined by Chris and John. Welcome fellows.

Unknown Speaker 0:36
Nice to have everybody back. I know last week I was like, so

Unknown Speaker 0:44
this week, we’re talking about can AI solve this problem. And so that’s what we’re gonna be talking about, you know, there are different challenges that marketers run into a lot of common problems or specific problems. And one of the questions that we get a lot, or that we see discussed a lot is, you know, the use of the appropriate uses of AI, you know, in those situations. So today, you know, if you’re with us live, feel free to drop your questions in the comments, but basically, we’re gonna be covering if AI is an appropriate solution for various marketing challenges.

Unknown Speaker 1:22
So Chris, where would you like to start? Look at this pig.

Unknown Speaker 1:26
Okay, that was really random.

Unknown Speaker 1:29
So this is the output from a model called stable defusion, which is an image generation piece of software.

Unknown Speaker 1:38
What it does, you give it a text prompt, I typed into this thing, a pig standing in a pile of money in a library, because I was writing a piece earlier on, on politics stuff, and it generated this image.

Unknown Speaker 1:53
I use this for a blog post. So one of the very interesting problems that AI promises to solve and and can obviously, realistically do it with this with this lovely image is that for some types of content generation, it’s a really good fit for stuff that’s not super high risk, not super high value production, where you would probably just have used, you know, random clipart from the internet or something previously, now you can have genuine custom custom built stuff that is tuned, like setting up this photoshoot would be a really kind of a messy pain, right to rent a pig, get a whole bunch of money, and then rent a library and do this, whereas a machine has just done this for me. And it did it in about six or seven minutes. So one of the I think most interesting applications of AI technology now is in the generation of new content that we can use to augment our existing content marketing capabilities. So to go back to the question, Does AI solve this problem? The problem being solved in this context is you want a specific image, that you don’t have time to take your or generate yourself, and you don’t want to run into copyright issues. That’s exactly right. The the problem I want to solve is I want some custom artwork for some content that is unique, that is decent quality, that is fast. And that is relatively inexpensive.

Unknown Speaker 3:26
Okay, so that’s one very specific use case.

Unknown Speaker 3:32
You know, and so, you know, sort of the other side of that is, why wouldn’t someone just use some of the, you know, royalty free images? Is it because that, you know, everybody’s using the same image of a laptop with a cup of coffee with a plant.

Unknown Speaker 3:50
Exactly. And everyone’s you know, got that same awkward image of, you know, three people are smiling and shaking hands or whatever, or stuff like that. There’s that there is also the aspect that, depending on the company you’re dealing with, you don’t know if the rights have actually been cleared for that image, right. The only rights that you are guaranteed to be clear on are either ones that have been generated by a machine that’s under your control. Or you have taken the photo or drawn or made the art yourself, right, if you take the photo with your camera in your living room, you know, without a doubt, who owns that copyright.

Unknown Speaker 4:26
What do you think, John? Yeah, for this kind of stuff. I mean, it definitely is a easy solution. You know, we see a lot of people going to stock art piles and, you know, hopefully this will finally be the death of clipart. We can see the you know, see crappy garbage animations dry up and die up, ultimately, for good. So yeah, but Chris, how about though? I mean, how long did it take you to get that running? And I know you strung it all together, like, how long do you think it’s going to take before stuff like this becomes more commercially available? Well, it’s commercially available now. Open

Unknown Speaker 5:00
AI and their models have, you can sign up for their service and pay like 11 cents an image to have a generator for you. And you just have to be clear about the prompt. The version I used is, as expected completely unsuitable for 99%.

Unknown Speaker 5:19
Yeah, asked that question. Exactly. So you spin up your Google colab instance, you import the notebook, you run all the libraries, provision the instance, set up the GPU, and then you’re ready to start typing stuff in. Now, here’s the trade off, once Trust Insights pays a $9, a month fee for Google colab for Google colab Pro, because I use it for a lot of very heavy machine learning processes,

Unknown Speaker 5:43
including image generation and text and stuff like that. There if the fee to generate all these images is essentially zero, right? Because we’re paying for the instance anyway. And I can use it as many times as I want.

Unknown Speaker 5:58
If I was the average marketer, who did not have machine learning technical skills, the 11 cents per image from from open AI is reasonable, that is a reasonable thing. Because if you think about it, how much would it cost you to have a designer design a custom image for every blog post, right? That could that’s at least $1, that five, or 10, or $25, right? Or you end up doing a template like we do with a lot of our stuff, which where we had a professional designer build a template for us. And then we just remixed the words on it, but it’s essentially the same image over and over and over again, right? This allows you some level of customization, while still being able to have you know, stuff. And it’s, it’s not just pictures of silly pigs, right? This is one I said, I want an image of a seaside town in Denmark. And, again, for a piece of blog content for social media posts or something, this is perfectly good enough, I sent this to a friend of mine who lives in Aalborg. And she was like, Yeah, I mean, that pretty much looks like two thirds of Denmark’s towns, it is not any one town in general, like you will not find this on Google Maps anywhere. But it’s close enough that if you were, say, a travel blog company, or travel company in general, this would be an easy way to source images, we have a travel client, and you know, they’re constantly running into issues of trying to, you know, images and image sizes and things, well, just have the machine make it for you, and exactly the specifications you want. And then you don’t have to worry about oh, we actually put up a 44 megabyte photo that our our field photographer put up and now our website has been penalized by Google’s core web files. Now is that can you dial it in by size and resolution to Can you specify what you actually want? You can? Oh, that’s cool. Okay, so that’s one challenge. I was gonna say. So it sounds like we’ve covered, you know, unique images for content or unique content, you know, visually represented. So that’s definitely one challenge where AI is an appropriate use. So we had asked our analytics for marketers community, it’s our free slot community AI slash analytics marketers, if you’re interested in joining, totally free. We asked them what some of their marketing challenges are. And what we wanted to do is see, you know, is AI an appropriate solution?

Unknown Speaker 8:19
So it’s funny, the very first response that we got was talent.

Unknown Speaker 8:25
And so I feel like we need to pick it apart a little bit. So the hiring process, AI, probably not the best use, unless you’re calling through resumes. You know, we can talk through sort of the pros and cons of that.

Unknown Speaker 8:42
But AI might be an appropriate solution, depending on the job roles that you’re trying to fill.

Unknown Speaker 8:51
So Chris, I can see you, you have a lot of thoughts on this. John, I’m sure you have some thoughts, too. So, Chris, what do we got?

Unknown Speaker 8:59
I would say the appropriate use of AI in that context is to improve the productivity and remove tasks from your existing workforce to free people up to work on higher value tasks, right? If you’ve got like, you and I used to work at this PR firm, and this is one poor sucker who

Unknown Speaker 9:15
her job all day long was literally copying and pasting links from Google searches into a spreadsheet like that is clearly a task that you even need AI for that doesn’t look straight up regular procedural programming will solve that problem, but AI will help you do a better job of filtering it, right. That is a task that AI can 100% solve, and then that person there 40 hours a week, you know, struggling not to claw their own eyeballs out, they can be put towards a better task so you can get more efficiency out of your existing workforce on the hiring front.

Unknown Speaker 9:47
AI machines in general are always trained on existing data and existing data has biases. Amazon found this out the hard way in 2018 when they deployed a service internally to

Unknown Speaker 10:00
filtered through LinkedIn profiles to, you know, to narrow down hiring pools and the service immediately stopped recommending profiles or women just flat out stop hiring women like well, I can guess which data set you trained your data on. And it took them a little while to find this enough so that it made the news like it got out and and made the news. So anything where you’re dealing with sensitive protected information, anything you’re dealing with protected classes of any kind, age, gender, sexual orientation, veteran status disability, you want to be real careful about what you’re doing with that, and, and just not turning the data over to machines? Because that’s, that’s where very large lawsuits lay.

Unknown Speaker 10:40
All right, John, what do you got?

Unknown Speaker 10:43
Yeah, there’s, you know, I’m just kind of interested in some of the applications as far as intelligent advertising, you know, stuff that existing advertising teams can use on all the different platforms that they don’t have to do any work, it’s, they’re just taking advantage of Better Place Facebook ads, we saw, I think, recently, with Twitter, their ads, for trying to drive traffic to Apps has become more and more automated. So those are the easy hits that you know, I’m watching and require no technical knowledge whatsoever, the entire team, never mind, you know, just the person doing it. So that’s always the easy lifts that I’m looking for.

Unknown Speaker 11:22
And then yet another one that just is on the horizon that I’ve been watching and waiting for the kind of take a bigger role is all the internationalization stuff, you know, so much of translating websites and content into other languages and other cultures or whatever. Having all that stuff automated, I mean that,

Unknown Speaker 11:41
it to be honest, I haven’t even kept up with that field in the past year. So I don’t know if that’s continuing to move in if there’s stuff off the shelf. Now, that’s decent. But that’s just another one that I have on my wish list. It’s very decent. At the spark conferences here, I want to spoken Montenegro, the native language, there’s Serbian, and one of the things is great built right into PowerPoint, like you don’t have to use any external tools. As long as your PowerPoints got a microphone. And it’s got the internet connection, you can not only have automatically subtitle your talk as you’re speaking on stage, but real time translate. So as I was talking in English to this audience, PowerPoint was translating what I was saying into Serbian into Serbian closed captions right on screen, that level of accessibility stuff is so cool. And it’s it, like I said, it’s built right in us zero technical skill whatsoever, you just got to find the button in PowerPoint to turn that on. And you are good to go in the language of your choosing.

Unknown Speaker 12:38
Alright, so it sounds like the consensus about using AI for finding new talent is, you know, dependent on what kind of talent you’re trying to find what kind of tasks you know, John, as you mentioned, advertising, there may be uses of AI that you can implement, which would then mean, you wouldn’t need to add more of a headcount in terms of hiring the people, probably not the best use of AI, you probably still want that human

Unknown Speaker 13:07
sort of judgment, you know, which that comes with its own set of issues. But AI being programmed by humans, is probably not the best use of AI in that in that context.

Unknown Speaker 13:20
The one exception, I will say there’s, we did do some work for client a few years ago, where we took 17,000 of their phone call interviews with their recruiters did, had them all transcribed and then did some language processing to see what recruiters and candidates were talking about versus what was the language in their job ads. And what we found was that everything in their job ads was none of what the candidates cared about, like the candidates were, oh, no, we care about, you know, starting cents per mile, my home on the weekends and all this stuff. And when we gave that feedback back to the client said, hey, you need to change your job ads, you just talk about these things, because that’s what all your recruiters candidates are talking about with you on the phone. They did. And they saw a 40% increase in applications like literally overnight.

Unknown Speaker 14:07
So AI can certainly help you focus in on what you’re hiring for. But again, sort of the people themselves, I think the conclusion that we’re coming to is that AI is not the appropriate solution for the actual hiring process.

Unknown Speaker 14:23
Alright, so the next marketing challenge that we were given was gaps in the data. And so thinking about your, you know, sales and marketing funnel from awareness all the way to purchase or enrollment in this case.

Unknown Speaker 14:38
You know, this is I think this is a common challenge for a lot of people because their data lives in a bunch of different systems. And depending on how you’re collecting the data, you may not be collecting, you know, contact information right from the get go. You may be collecting it, you know, five or six steps down the line so you then can’t

Unknown Speaker 15:00
connect it back to the original. How did they find out about us?

Unknown Speaker 15:05
So laid on me, Chris, where does AI fit into this one?

Unknown Speaker 15:10
In this case? Well, there’s two things that issue one is that there’s not enough data. So in this particular instance, from our analytics remarketing slack group, the person in question doesn’t have enough data to work with yet.

Unknown Speaker 15:26
And they need to collect more of that. So that’s one of the major challenges. But one of the outcomes there that is possible without individual data is a tactic called Marketing Mix modeling. This is using software, machine learning powered software, that, and we’ve talked about this a bunch of times on the show in the past that essentially ingest all the data that you have about your marketing and your outcomes and all the activities you’re doing. And then performs essentially a regression analysis that says these are the channels that have the biggest impact on the outcome. It’s very similar to attribution analysis, but not quite the same thing. Because attribution analysis requires some more granular data marketing mix modeling can be done with more summary level data, and you can still get valid conclusions out of it. And so in this case, if you have the data, and it’s even if it’s a bunch of heterogenous, systems that are all over the place, you can unify that data into essentially what’s just one really big spreadsheet, and then build a marketing mix model from that, so that you can say, okay, these are, these are all the inputs, here’s the output, tell me which of these inputs matters the most.

Unknown Speaker 16:43
So I am going to respectfully disagree with you.

Unknown Speaker 16:48
AI is the appropriate use here. When I see this problem, I see this as a people and a process problem, not a technology problem, because what you’re describing still doesn’t answer the question of where did this person come from? What did their individual journey look like? And so to me, that’s a

Unknown Speaker 17:08
process issue that there weren’t clearly defined requirements from the get go to say, this is how we plan to track John Wall from finding out about us to enrolling in our program.

Unknown Speaker 17:22
AI can’t fix that AI can’t go backward in time and say, Okay, let me just go trace this person down all over the internet and figure out everything they’ve ever done, and then only pull out what’s relevant to your specific, you know, product. So I respectfully disagree that AI is the solution here. I think it’s a people in process problem.

Unknown Speaker 17:48

Unknown Speaker 17:50
well, the big thing with that is it kind of depends on what situation are you in? I mean, obviously, if you want to cure the disease, yeah, you’ve got to go back and figure out what the people process problem is, if this is some kind of one time thing, where you’re just trying to backfill data, which, you know, you screwed something up, and somebody threw away, you know, one of the spreadsheets, you need to complete the data, then AI is your thing.

Unknown Speaker 18:13
But, and, you know, kind of squarely in the middle of that is using AI for data hygiene, you know, you could come to a point where you say, this is the kind of stuff we’re going to try and get, but most of the time, we won’t get it. So let’s use AI to supplement it. But yeah, always get it from the people if you can first obviously, you know, you may be 98% accurate, but it’s still cheaper to get it up front.

Unknown Speaker 18:36
Yeah. Now, to your point, Katie, getting from people is going to be vitally important because what we’re seeing happen now and the pace is increasing is data blindness. So a C and IL, which is the French Data Protection Agency, as part of the overall European Commission, essentially just ruled two separate rulings, one, which says you may not use Google Analytics in the EU anymore, period, like so. Dope, you’re done. So there, that is a you know, Google has obviously appealed that, but it is not looking particularly good for for any marketing company collecting data.

Unknown Speaker 19:19
Without, you know, very, very specific privacy parameters around it. And the second thing that the European Commission itself ruled is that any PII, any personally identifiable information is now reclassed within the entire EU as sensitive protected information, which means that you it has the same protections on it as if it was, you know, someone’s sexual orientation. So just having someone’s IP address or email is treated with the same level of sensitivity as someone’s gender, or someone’s ethnicity, which is a really big deal. Because as you go up in levels from personally identifiable to sensitive, protected to protected health information, and so on and so forth.

Unknown Speaker 20:00
The legal repercussions increase. And the the stringency of your own data protection increases, which means that we have to be looking at AI models to work with less and less data less and less personally identifiable data until essentially, we’re going to have almost none at all, except for the outcomes that we can see. And the the activities that we’ve done. So AI is not the answer to this, as you point out, this is not the answer in this situation. But it will very is very quickly going to have to be the default solution in a privacy friendly environment, if you want to avoid the European Commission, you know, kicking you out of your Well, and, you know, I don’t want to get too off track, but we’re also, you know, talking about the quote, unquote, cookieless future, where marketers who aren’t creating things that

Unknown Speaker 20:55
are convincing people to give some of that personally identifiable information are basically going to be screwed, you know, so if you’re just putting something out there, you know, a, you know, half written blog post, and you’re like, oh, I don’t know, I can’t figure out why nobody cares about this thing. Why aren’t people signing up for it? Like, that’s a you problem, that’s not, you know, your audience problem, you have not done your job as a marketer. And again, that’s not something that AI is going to fix AI can’t fix your crappy content. Now, we could go down the road of saying AI can generate content, but Google just rolled out.

Unknown Speaker 21:34
And I apologize. If I’m misquoting Chris, you can feel free to correct me, you know, the human basically what they’re saying is like, they want human generated content, not AI generated content. And so that further sort of, if you’re relying on AI to generate your content, and you’re relying on, you know, retargeting pixels and cookies to collect information, like you’re screwed.

Unknown Speaker 22:01
Yeah, the helpful content updates

Unknown Speaker 22:04
include John I was talking about earlier this week as well. Google has made it very clear that the content you create should be helpful, it should add value, right? They targeted specific examples like Roundup blog posts that don’t add anything new, you know, lists of lists and stuff like that, essentially, it’s all the mediocre content. And a lot of marketers been cranking out crap for a while. Where this is going to get very interesting is the language generation models are very sophisticated, extremely heavy, very, very powerful. And what they create is getting harder and harder to distinguish from human written content, particularly when we look at things that like human content farms generate. So if you have a human here that is paid, you know, minimum wage, they hate their job, and whatever, they just phoned it in a face roll on the keyboard Monday through Friday. And that’s their level of quality. And you have a machine with a language generation model that creates mediocre content. It’s okay, it’s mediocre. It’s better than the human content. So even though this is the human content, the machines creating better stuff. And so as these models improve, and the newest model, the DaVinci model from open AI, I was testing it the other day. It’s not bad, right? We’ve so we’ve gone from Wow, that’s terrible. That’s a word salad machine. Thanks a lot to mediocre like, Okay, this is the exact same blog post I’ve seen written on 400 other blogs, too. Okay, that’s

Unknown Speaker 23:34
reasonable, that’s useful. That’s helpful, you know, where’s the bar of excellence for what AI congenic is getting higher and higher, which means that Google itself will have a harder time distinguishing if the content is truly helpful. Does it matter whether human or machine wrote it? The signals they look at are things like your bounce rate and clicks and stuff? Like that’s one of the reasons why we all had to install Google Analytics.

Unknown Speaker 24:02
They can see the behaviors of users and if users are interacting with AI generated content at this with the same performance metrics, a better performance metrics than humans one ones, they’re gonna have a very, very hard time detecting it. So I would say the moral of the story there is

Unknown Speaker 24:19
yes, your your content has to be helpful. content has to be good. Whether machine generates for human generates or machine does the first draft on human polishes, or whatever the case may be.

Unknown Speaker 24:30
Ultimately, you have to produce quality, which goes back to what you were saying earlier, Katie, which is you’ve got to give people a reason to at least click right you see a picture of a pig standing on money. Okay, machine generated that but that’s still mildly amusing. That might be enough to get the click even if it’s completely machine generated.

Unknown Speaker 24:48
What do you think, John?

Unknown Speaker 24:51
Yeah, it’s,

Unknown Speaker 24:52
you know, there’s no saving crappy content. I think that’s the biggest point of all that if you can come up with all these tools in here

Unknown Speaker 25:00
have marginal generated, copy. And it is good for certain situations, if you’re just trying to get tonnage of reporting stuff out there. And one thing we had talked about earlier this week, there’s definitely going to be a full on war over product reviews, you know, because this kind of technology can be used to generate product reviews. So you can get that, you know, 2004, or five star reviews that you’re looking for.

Unknown Speaker 25:23
But, yeah, none of it is substitution for, you know, really knowing what you’re doing and having having the pillar content be stuff that’s remarkable and unique is it’s, you know, there’s going to be way too much, you know, eight out of 10 content out there.

Unknown Speaker 25:41
So we actually got a comment, which reinforces what we’re saying. So Scott Sweeney over from LinkedIn said, AI content created by natural language generation is best when designed and edited with humans, which is absolutely correct. The way that I look at systems that generate content for you, there still needs to be that human intervention, it’s the same thing with creating any kind of AI, the humans need to decide what it’s going to do. The humans need to decide whether or not it’s acceptable. And the humans need to be the ones to put the final polish on. It’s the same thing with your image, Chris, you the human said, this is the image that I specifically want. Now, I’m guessing this is just a wild guess. But I’m guessing that there are going to be systems that will read your content, and decide what kind of an image is appropriate. And that’s fine. But without that human QA, you may end up with really bonkers images that have nothing to do because AI won’t be able to detect nuance and sarcasm and those kinds of things. And you may end up with images that are wildly inappropriate for what you were trying to do. And so if you aren’t, then as a human Q weighing it, then you’re still right back where you started, where you’re putting out crappy content. Exactly. 100%, right. And believe me, in the first few trial runs of the new software was like, Whoa, that’s interesting, because when you download the version to run for colab, itself, it doesn’t have any of the safeties that are built into the online version to like, oh, that’s, that’s exciting. But here’s another example of the so this is the DaVinci. Mala. Katie, this is your cold open from the newsletter yesterday, I just put it in here. And I told the this model, do a TLDR. Write a quick summary. And, you know, that’s essentially what just did there. It says, Okay, if I summarize this entire post, it comes down to most problems, start with people. So if you’re having trouble with something, start with who made it, right. You didn’t say those exact words. But that is essentially what it distills down to. That’s pretty good, though. from a, from a content perspective. You know, one of the things that we’ve been talking about with Trust Insights is how do we create those featured snippets for for our blog posts that go into a little SEO tool on our site? That’s not a bad little summary, right? That could that could go in there. So there’s all these opportunities. So what’s interesting is I had a sentence that was essentially that in the post, and so it didn’t, I guess, and that’s where I’m like, well, it didn’t generate anything new. It didn’t summarize it. It literally just grabbed a sentence. From the post itself. It wasn’t that exact sentence. But it was basically that.

Unknown Speaker 28:22
Yep. So here’s

Unknown Speaker 28:24
a different poll from it. Yeah, it’s doing it’s just doing abstractive summarization, it’s attempting to boil it down to just a couple of sentences.

Unknown Speaker 28:34
I think, you know, to Scott’s point and to what you’re saying,

Unknown Speaker 28:38
the human is the conductor of the orchestra now, right? So we have all these tools can do these things, but they don’t do it by themselves. And nor do you want them to you want instead for us to be guiding them? Well, you know, I can see how this summarization tool is actually going to be pretty helpful. Because if you write a post, and then you ask for that summary from AI, if AI can’t figure out what the heck this post is about, humans probably can’t figure out what the heck this post is about to. So I can see it as a really good QA part of your QA process.

Unknown Speaker 29:18
For writing your content, which goes back to John’s point is like you can’t fix crappy content, like you have to do the work. You can’t get by with just sort of the bare minimum. And so using AI to support but not replace, I think is what we’re all saying

Unknown Speaker 29:35
100 The thing with that, though, that I dread is, you know, anything where if it’s not your algorithm, you want to check this stuff before it goes out the door. Because you know, you have no idea what kind of bias or weirdness is in there if you didn’t write it, so human touch is critical before it goes out the door. Yeah, for all you know, there’s somewhere along the line some angry engineer who knew nobody would be checking the code and said

Unknown Speaker 30:00
Anytime someone says marketing, replace it with a Not Safe For Work word. And because you’re just using, you know, whatever you bought out of the box, you have no way to fix it. So definitely, buyer beware.

Unknown Speaker 30:15
All right, so we’ve covered? Well, we kind of got a little bit off topic. But basically, the question that we were covering was, you know, following someone through their user journey,

Unknown Speaker 30:27
AI can fix some of that as privacy gets more strict, you’ll be able to do some general generalizations about your marketing mix. But following one specific individual through their journey is going to get harder and harder and harder. Unless you are creating things that are compelling enough that someone says, yes, you can have all you want to know about me, because you have created something that I cannot live without.

Unknown Speaker 30:54
So the next marketing challenge is definitely an interesting one.

Unknown Speaker 30:58
One of our community members said, Google Analytics 4, everything about it is a challenge.

Unknown Speaker 31:06
And I don’t disagree with that. Google, in case you missed it, for some reason, Google announced earlier this year that Google Analytics 4, was going to become the standard piece of software as of July 1 2023. So if you have not installed it yet you are behind, you will not have a full year’s worth of data.

Unknown Speaker 31:28
Go back to any of our old shows, you can find out where we rant and rave about, you know, Google Analytics 4 versus use of Universal Analytics, you can take our Google Analytics 4 course, where we help you solve all of these problems, do it sooner rather than later. Again, time is of the essence. So all of that being said,

Unknown Speaker 31:47
artificial intelligence, does that help with Google Analytics, 4 being a pain in the butt?

Unknown Speaker 31:57
Chris, you mute it.

Unknown Speaker 31:59
It doesn’t help with being a pain in the butt, but does help with processing the data that Google Analytics gives us.

Unknown Speaker 32:07
Remember, that is a this is a big architecture change. Google Analytics used to be sort of a Swiss army knife, collect your data, analyze it, report on it, and it does only one of those things. Now, it’s an analysis tool. It’s also really, really good engine for processing data. But a lot of the things that we used to value about like, page value, gone, right, it’s not in there anymore. So you have to process yourself. One of the things I was working on last week that finally got working after beating my head against the wall for a couple of months on it, is we have this thing called most valuable pages, which with the old Universal Analytics was a relatively straightforward thing, what pages, you know, help nudge people towards conversion. And the Google Analytics 3 Universal Analytics model, data model was really good at that. It all of the data that we relied on is no longer there in those formats and Google Analytics 4, so we had to totally re engineer thing, but we finally got it working. Unfortunately, it requires a much more advanced machine learning model to process it, and it takes a lot longer to but machine learning can absolutely take data out of Google Analytics 4 and slice it and dice it until you get to a point where yeah, now we can have something we can take action on. So

Unknown Speaker 33:17
from the Trust Insights website, I said, run this classification model, tell me which pages nudge people towards conversion. And we basically compare user journeys where people did convert versus user controllers where people didn’t convert and look at the pages that they did in each one and say, Okay, what’s the difference? And so for, for our website, for the last I think, was last 90 days.

Unknown Speaker 33:40
The LinkedIn job hunting webinar was a big one the five ways AI is changing marketing, Keynote, the fundamentals of marketing analytics. Some of the pages that didn’t do as much like the the homepage of the podcast didn’t really do as much. So from a use case perspective, there are absolutely ways that Google Analytics 4 data can be used.

Unknown Speaker 34:06
AI can use on this data to make it useful, because that’s one of the biggest handicaps with GA four is that the analysis that’s in the product doesn’t answer questions that your average market is user story wants the answer,

Unknown Speaker 34:23
which I think is valid. But I do think that applying AI is almost getting ahead of the problem, which is the whole interface and experience with Google Analytics 4 is changed. And they’ve pulled apart the products to be Tag Manager, Google Analytics, Data Studio, etc, etc. And so AI can’t fix that, necessarily, AI won’t be able to set up your system exactly the way you need it to be set up with all of the different toggles and settings and everything. So to get

Unknown Speaker 35:00
up to the point where you can use AI to analyze the data is like step five. Step one is you first have to set up the different systems. So AI can’t fix that. That’s still a people in process problem. Exactly. And it’s a people process problems also platform problem in that Google has made this platform change for their benefit, and not ours. And because we’re not paying for it, we don’t really get a vote into, into making any changes to it, we just kind of have to live with it or switch up and change to a different product.

Unknown Speaker 35:36
So Google Analytics 4, that’s a people in process problem. You can apply AI once you have it set up and you’re collecting data, but not until,

Unknown Speaker 35:45
John, as we’re starting to wrap up, what are some of the other marketing challenges? You see, as you’re talking to people in your network, as you’re talking to prospects? That, you know, we’re not sure where AI quite fits in?

Unknown Speaker 36:01
Yeah, I think on the SEO front, there’s a bunch of places where we’ve looked at stuff, whether it’s analyzing topics to understand who the leader in the space is who you should be emulating, trying to, you know, take back ground that you’ve lost. And what content of yours is doing? Well, a lot of that kind of stuff is well suited to machine learning analysis, because you know, just looking at your own traffic is not enough, you need to have a rough idea of how the competition is doing and how topics that aren’t on your keyword lists, or that you’re not getting traffic for are working. So

Unknown Speaker 36:33
yeah, that’s a lot of the, you know, low hanging fruit or whatever business analogy you want to use. But if you want to get in and generate some data and make some things happen, I think that’s the best place to start for that kind of stuff. And, you know, of course, we do that for clients as far as traffic analysis and helping them do content calendaring, you know, figure out not only what to talk about, what when’s the best time to talk about it, you know, you don’t want to be creating content, about, you know, getting ready for the beach in November, you want to time that stuff for April, May, June. And that there’s, there are trends and things to follow for virtually every kind of topic like that.

Unknown Speaker 37:14
So yeah, AI squish mellows

Unknown Speaker 37:18
pumpkin spice for everyone. So AI is a definitely a good use case for things like SEO, analyzing your SEO, and again, so I feel like we keep circling back to content being the thing that kind of ties all of these, you know, questions to these challenges, which is true, because content comes in a lot of different forums, content, you know, video, audio written, you know, images, whatever, AI can do a lot of that. However, you still need human intervention in order to ensure that the content is good quality, that the content aligns with, you know, what you want it to do what your brand wants to do, what your audience wants to see. So AI is,

Unknown Speaker 38:09
at least from where I’ve said, AI shouldn’t be front and center. AI is a supporting role. AI is a, you know, secondary character, because it’s not going to fix all of these problems for you. If you don’t have your people process and platform squared away. First. It’s just an appliance, you can still use chef, you still need ingredients, you still need a recipe, those things don’t change as the appliances have gotten better. AI is good at three things right? Classification, aka, I’ve got a bunch of stuff to help them make make sense of regression, which is I’ve got a set of numbers, what numbers are like it, and generation, Hey, make this thing. Those are the three fundamental tasks that AI is as you the good at. So as we look at our marketing, we have to say, is this a regression problem, a classification problem is a degeneration problem. Or if it’s none of those, then that’s probably not a good use case for AI. Right? Hiring people is not one of those three problems, right? When you look at configuring Google Analytics, 4, it’s not one of those three problems. Creating content, it is content calendar forecasting. Absolutely. identifying trends and social media. Yes, that’s regression. What number is like this number? So if we think about

Unknown Speaker 39:18
our our problems as one of those three categories, we can very quickly bucket and say, yes, there’s this potential for AI to help you or no, it’s it’s it’s not the right tool. It’s right and you’re making steak and all you and you’ve pulled out the blender, put that away. That’s not what it’s for.

Unknown Speaker 39:35
Yeah, that just sounds gross, quite honestly.

Unknown Speaker 39:39
You know, I don’t know if anyone else caught it. But John, I did see your camera glitch for a hot second. So I’m pretty sure that we just expose that John is in fact, an AI avatar.

Unknown Speaker 39:51
And so instead of actually hiring John Wall, we created an AI avatar of John Wall, so that John Wall wouldn’t have to sit in these livestream.

Unknown Speaker 40:00
seems Max Headroom 3.0. That’s what I’m doing.

Unknown Speaker 40:05
Here’s a throwback.

Unknown Speaker 40:07
It’s funny, because that’s what I was thinking of when I saw it, too. So I think we’re all aligned. Yes, yes. Yes, yes.

Unknown Speaker 40:16
Any final thoughts on is AI can AI fix this problem?

Unknown Speaker 40:25
There’s definitely applications, you know, don’t go out and just buy the first thing that says AI on it, though, that’s a bad idea.

Unknown Speaker 40:32
Chris, no, it that is 100% true. Understand the kind of problem that you’re you’re facing first. And then you can make those determinations really, really quickly, whether it’s the right tool or not, but understand that the technology is changing so fast, and the scope of what it can do is increasing so quickly, that you should reevaluate that list of problems, maybe once a quarter and say, Oh, well, this is a problem that AI now can solve.

Unknown Speaker 41:03
I think that makes sense. I would add to that, you know, um, you know, AI is not going to fix a broken process. It’s not going to fix non existent process. It’s not going to fix, you know, your inability to collect the right data. So these things you all have to have that human intervention. First, people define what AI is going to do, not the other way around

Unknown Speaker 41:28
until Skynet.

Unknown Speaker 41:33
See you next week.

Unknown Speaker 41:38
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 AI podcast, and a weekly email newsletter at trust Got questions about what you saw in today’s episode. Join our free analytics for markers slack group at trust for marketers, see you next time.

Transcribed by

Need help with your marketing AI and analytics?

You might also enjoy:

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.

One thought on “So What? Can AI solve this problem?

Leave a Reply

Your email address will not be published. Required fields are marked *

Pin It on Pinterest

Share This