So What? Marketing Analytics and Insights Live
airs every Thursday at 1 pm EST.
In this week’s episode of So What? we focus on determining a trend. We walk through how to conduct trend analysis, a trend versus an anomaly, and what actions to take. Catch the replay here:
In this episode you’ll learn:
- The mathematics behind trend detection
- Two different measures to apply to any analytics to determine a trend
- How to know when to act on a trend
- How to research efficiently – 6/10/2021
Have a question or topic you’d like to see us cover? Reach out here: https://www.trustinsights.ai/resources/so-what-the-marketing-analytics-and-insights-show/
Christopher Penn 0:19
Alright folks, welcome to the marketing analytics insights live show. Katie is off this week. So we’re going to behave irresponsibly. Now, John, what are we doing this week?
John Wall 0:31
Today we are talking about extending on last week, we were talking about trends, what’s going on. So we’ve got the mathematics behind trend detection, you’ve got some methods to measure and then also how to know when to act when to actually put this data to use. So hopefully, we’re going to deliver the formulas and the so what that Katie has demanded in lack of her presence. So here we go. Fair enough.
Christopher Penn 0:52
All right. So let’s start with this because I think it’s important. What is a trend to you? What is a trend? Yeah,
John Wall 0:59
people talk about the site Geist, I love that term, because it’s inside baseball. But it’s just every once in a while certain topics gain some heat and some traction. And a lot of them can be black swans, where you just, you know, something explodes out of nowhere, some random meme. But then there’s other trends that are cyclical, or have, you know, repeat over time or when certain things happen. And that’s the kind of stuff that you want to jump on. Because there’s many times where, you know, if you live and die by corporate goals, or hitting certain targets, if you can find trends that drive that stuff, you can make things completely predictable. And that’s where you’d really love to be.
Christopher Penn 1:34
Yeah, let’s look at a few different terms. Because I think we should probably clarify some of this, we’ll go ahead and flip over to our art easel got the got the whiteboard gone today. Let’s do some, some basic art here. If you were to look in your analytics and see something that looks like this, what would you call that?
John Wall 1:52
You know, first, it would be a spike, I’d call that. So I that would be really, that’s my economics term would call that a spike.
Christopher Penn 2:00
Exactly. Mathematically, or statistically, we call that an anomaly. Something weird happened here, right? And then we go right back to where we were. Right. So we start here. And then if you actually would just sort of just pretend this didn’t exist here, essentially, you kind of just the same thing, right? You start and end at the same place. So that’s our first thing. anomalies, you always have anomalies. And it can be anything like you’re a bot attacking your website, it could have been popular post on Reddit that sent you 1000 visitors to your landing page, it could be any number of things could be an ad that ran really well. So that’s our first part as economically. Next, what would you what would we call this?
John Wall 2:42
Oh, that’s an upward trend. Yeah, that’s the one of the two Holy Grail. Exactly.
Christopher Penn 2:47
So that’s the upward trend. So a trend is essentially in this case, either an upwards or downwards, decline, the answer decline in the data. So we have this, imagine the mathematics here we have it, we have this sort of average here. And then we start to deviate from that average, more and more and more. And a trend can be you know, it doesn’t, it doesn’t necessarily need to stick like that. So let’s do let’s draw this, here’s our trend, and back. So this section here would be a trend, right? where something has happened, we’ve got a big difference. If you have that average line, this kind of a big difference for a while and then we start headed back towards the average. That’s that’s a trend. Now, that’s something that’s important is math, the mathematically important. And then there’s a third category. And this is I think, is the holy grail, which is looks like this. What would you call that?
John Wall 3:42
That a new plateau is the you know, usually how we describe that you jumping up to the next level?
Christopher Penn 3:47
Exactly. So in in mathematical terms, this is a breakout, right? So something it trended for a while, and then it’s stuck. Right? So I hopefully this is your website traffic like this is the hope of every marketer is you run a big campaign, you get this big boost, you’re happy. And then the audience sticks around they find so much value in your like, yeah, like they they continue to give you the thumbs up. So that’s what we want to talk about today. Today, we’re talking about trends, but the goal of our marketing the so what is to find trends that lead to a breakout that find trends that lead to that sustained level of growth, because it’s not much fun if after a campaign is done, go back to the to the beginning, right that’s that’s no fun. What we want to do is we want to figure out how do we get how can we see things that will eventually lead to breakouts and the bad news there is there really is no easy way to know that right? So those things only happens in retrospective you can’t tell in advance that this is going to be the thing. There are very very few companies that have ever done that. The only way I can think of off the top of my head that’s had three breakouts. is Apple right with fresh with the personal computer. Then with the smart the graphical interface, the Macintosh, and then with the smartphones are like the three is four because they had that they had the iPod, the political musical player. Not many companies, most companies are lucky if they get one breakout much for. So it really is something special. So how do we determine this? How do we know that what to find these things? Well, let’s pull up our good old friend, Microsoft Excel, or the spreadsheet software of your choice. You don’t have to use Excel, just know that most people do. The first thing you need for any kind of trend detection is the data itself. So I pulled some data here from Google Trends, which is if you’ve have not used Google Trends, I don’t know what you’ve been doing for the last few years. But it’s it’s one of the best public data sources for just general trend things. So let’s pick a let’s pick on have a little bit of fun here. Let’s pick on the NFT, the non fungible token, just from eyeballing, you don’t even have to use any math yet. What would you classify this so far? John, I, you
John Wall 6:05
better have already short sold, because those tokens are getting to be worth less every day, it looks like?
Christopher Penn 6:13
Well, so these are searches. These are not actual dollar values. And again, this is data can be applied to justify anything. But we want to get data, it has two things. It has a numerical value, which Google Trends provides you. And it has a date, those are the two most critical pieces, the for trend analysis, if you don’t have those two things, you really can’t do trend analysis, at least you can’t do it in any sensible way. Now, as we were just showing, first thing you want to do is just do a bit of explore exploration. So I’m gonna put in a chart here. Oops, get rid of that. So let’s do I will use clubhouse to everyone’s, well, I don’t know, punching bag, let me know exactly favorite punching bag. And obviously, clubhouse is a search term that has existed long before the app that we know. But we can see right away that there was so this background noise and then big spikes, you know, coming in here last year, and then this year for for NF T’s. So our first step, always take a look at the data. Just see if there’s even a there there. If there’s not even there, there was just a flatline or something then you know, you probably don’t need to do a whole lot of extra exploration. In this case, it looks like there’s something that’s happening. So our next step trends, remember we said are increasing or decreasing. So the thing we want to look at is how does what is the rate of change? So let’s do which which one do I do more John NF T’s or cliffhouse?
John Wall 7:42
Oh, let’s look at the NF Ts. That’s definitely more
Christopher Penn 7:45
more fact. So I do call column called change here. And I’m going to do good old fashioned new minus old divided by old, that’s our rate of change, and populate that down. And we have divided by zeros, that’s fine. It’s not fine, but it’s fine. ignoring those divide by zeros for the longest time here. You know, in fact, I’m going to do this for anything that is eight zero just to avoid. Let’s do advanced replace. If it’s zero, I want to replace it with point 001. Just so that we have something there. Okay, close. And now we have our graph, that’s not really all that helpful. Is it? Yeah, let’s do that. Okay, so for the longest time, we have there was no search volume for NF T’s right? Just wasn’t and then you start to see that rate of change really started to crop up like okay, there’s, you know, something happening there. And then you get to the, you know, some pretty big changes like two and a half 2.5x changes, and then going back down, so we can already see that there is something they’re going to didn’t even have anything till zero to 30. So I’m gonna go ahead and just drag this down. They’re just to make the graph a little bit cleaner. And now our rate of change becomes zero. Now, the challenge with this is because we were talking about, you know, is something increasing or decreasing. Anytime you’re working on a day level data, it does get kind of messy gets very noisy because you have fluctuations happening all the time. So let’s make a seven day average here. And it’s gonna roll a seven day average of our change.
And take away our initial chart and put our 70 average chart up instead. And there we see our rate of change. Generally speaking, when you, you know, when we’re talking about like, should you pay attention to a trend? Once it starts that ascent? Yeah, you probably want to be paying attention to it, you know, you go in here to 14%, change date, you know, rolling, rolling average of 28%, change, and so on and so forth. So you start to see this increase, until it really explodes until you get you know, you’re well over 100% change here. And then you take the roller coaster ride all the way back down. So you almost have two trends here. At the same time, you know, if you split it down the middle, so that’s one of the easiest way to detect a trend is just do a moving average. And see, is there substantial increases is or is it flatlining? Because if you were to take a this exact same data, and actually let’s do this for clubhouse, because clubhouse is a good example of this, I’m just going to do every new minus old
Unknown Speaker 11:01
I buy old.
Christopher Penn 11:05
And do this am I going to do a few days worth here? Because Because we know that clubhouse didn’t actually exist as a search tailor. An important thing for from, at least for marketers for while I’m going to plot this out here are 70 average. Let’s go ahead and insert our plot. And what we see is the ups and down up down volatility kind of bouncing up and down, but it’s hovering around that zero line keeps coming, coming back to zero eventually. That’s an example of something where there isn’t a trend, right? Well, you got this up and down noise, there’s no trend, there’s nothing detectable saying like, yeah, there’s big changes from day to day, there really isn’t. It’s just you know, some relatively minor noise. Now, if we take that, and extend this all the way down. Now, up and down, up and down, up and down noise until you get into a little bit later on then you start getting Okay, now we’re we’re starting to see some real stickiness to that increase. So it’s not it doesn’t stay news back to you know, anywhere near zero for a while, when you’ve got your trend that kind of averages out to zero hovering around that zero line that’s it pulls away from that stays pulling away from that for a while. Now you’ve got a trend, and you’ve got something like okay, there’s a trend there. So imagine doing this with say like your open rates for your email newsletters for your your Twitter posts, for views on your YouTube videos, what you’re looking for, is for it to start pulling away from the zero line on a regular basis. Like here, you can see these are these are fairly close to zero, and then you start saying that, Okay, now we’re above 5% pulling away, now we’re above 10 1520 25%, just pulling away and staying pulled away. That’s a trend. So from a mathematical perspective, that’s a relatively easy way to figure out these trends. So here’s the next challenge. When would you take action? John, when would you like say, you know, I gotta, you know, I got some money, I’m ready to invest in, you know, an NF T of I don’t know, Bugs Bunny with a helicopter and etc. When would you make that decision? Yeah, that’s,
John Wall 13:17
you know, unfortunately, hindsight will always give you better, you know, results. I mean, anybody here would be like, okay, let’s see that. You’ve kind of got that first surge, you know, if you had your choice, you get at the lowest point before that first surge, right? I mean, that would be where you can make the most profit. But of course, at that point, you’d be sitting there saying, Oh, my God, it’s flowing through the floor here, you know, we’re not going to do that. Now, you do have kind of a whipsaw there, which is interesting, you know, it went up and then went back down again. And so that, you know, it’s a question of like, how brave Are you are, you know, when do you want to jump in a lot of formulas I see, well do you know, you take a second larger moving average, and when it gets above a certain point over the moving average, you say, okay, that’s when we’re going to jump in and dive, because I think you would definitely want to win that second peak, went above the first peak or was up near that first peak, that’s, you know, when you’re moving averages should be yelling at you to buy in. Exactly. So
Christopher Penn 14:13
what you’re describing was called a moving average convergence divergence indicator. And all it really is, is a shorter term average difference by a longer term average. So I’m going to put together a 30 day moving average here, and let’s go down to sell 31. to average, right you are rate of change. And now if we plot these two things together, let’s make this nice and big so that everyone can see it. The orange line is the longer term moving average, the blue line is zero. literature moving average. And what you were saying is exactly right, you’re looking for when the blue line pulls away from the orange line and stays pulling away from it like because there’s lots of points here where there’s not much of a difference between the two. And it’s bouncing back and forth. And then this is the first point where there’s a lot of area under that blue line that that wasn’t there previously. And then even more so to the point where the orange line pulls up. So the the buy point for you really would have been probably about here, right, we see a big change in the orange line, the orange line is trying to pull up, and then the blue line pulls away from it again. So that’s the divergence, we see the short term average, you know, things are popping things up, there’s interest in this thing. That’s when you buy and so you catch all this upside. And then the big question, of course, is when do you sell? You probably would, you know, ideally, this be the point where you don’t lose any more. So you buy in here and you sell here, you’ve still made a decent Delta right? Now, this is a seven to 30 day moving average, you can change those two windows to be whatever is appropriate for the trend you’re trying to analyze, is it, you know, email open rates? Well, you might have to have like a 90 day average, if you send emails, you know, only once a week, a 90 day average is probably going to be about right. If your sales cycle is two years, you can need a lot of data. And your your moving average might be like a five year moving average. If your sales cycle is that cripplingly long, but a moving average convergence divergence indicator is exactly the way to go. Because if you look at the the variances, look at the differences and say, okay, hey, something’s pulling away here. Now, yeah, obviously, if if this has been something you’ve been keeping an eye on for a while, this would have been the point to buy in, and then sell off around here, when you first start to know some big percentage change differences. Okay, like, let’s sell off maybe here. But even if you were to follow the, what we’ve outlined so far, you still would have done pretty well.
John Wall 16:54
Yeah, you could have taken advantage of it. The other thing with that is you do have to layer on just intelligence, like whatever, you know, like if you realize that, hey, a new product dropped here at this point, that’s something that could drive your decision making factor the next time the cycle comes around, or it’s, hey, we lost a CEO or you know, whatever, any kind of Black Swan events can can help that. But those those moving averages are Yeah, are just trends, it’s just you’re picking up more population. And so that’s the easy stuff to spot. Exactly.
Christopher Penn 17:24
So let’s look at another data series, one that is a lot more mundane and a lot less buzzy than NF T’s and clubhouse, let’s look at TSA passengers, people who’ve walked through a TSA checkpoint, we’re gonna do the exact same things, we’re gonna put in our rate of change here first, so and if you want to get this ad go to tsa.gov, you can get this data from the Transportation Security Administration. So new minus old, divided by old. And what we’re looking for is passenger changes through TSA checkpoints, let’s do our seven day average. Go down to here, like a nice seven day moving average, or percentage shader passengers, let’s do our 30 day average. Self, that’d be too.
Now let’s go ahead and plot our thanks. Because what you were just talking about is is so important. This is we’ll play the game called spot the pandemic, right? What you’re seeing here is that the the blue line is the 70 moving average number passes through the checkpoint, we see that took a nosedive last year, and then you have the 30 day average fall following it falling upon it. And then both them coming back a little bit towards equilibrium. But what’s interesting here is before the pandemic, you can see we’re kind of hovering around the zero line, you know, it was there was growth, but it was huge. And then after this, you know, when we started the recovery period, you’re seeing that there’s a lot more space under both lines than there was prior to now those big spikes downward, those are holidays, right, those are Christmas and Thanksgiving, several people traveled a lot the days before, and then didn’t travel on those days again. So again, it’s a, it’s a good reason to do that. It’s kind of smoothing. But we’re at a point now where as we start returning to baseline, were this here at the end, it looks very much like this here at the beginning. We can say with some level of confidence, hey, it looks like things might be getting back to normal. And if you were to chart out the actual passenger throughput itself, let’s put this up. So it’s visible here. We do indeed. See, yeah, we’ve got that trend. It’s a very clear trend that’s happening here. And right here at this endpoint. You can see we’re kind of even with you know, the sort of the bottom ended the pre pandemic days. So we really are at a point where our moving average convergence divergence indicator is saying things may be closer to normal than we think if we got past just coming through. So now the question is, what do you do with that? Well, obviously, if, if you were selling, if you’re selling vacation packages, you probably should have started your marketing here. Right. And you definitely should have your marketing going here. So this is a this would have been about a month ago that we sign that wrap, but you can definitely see just what the basic trend line here year is, you need you need it to be to get going. There’s some software out there that can do a lot of this this trend analysis for you, and Savior, you you don’t have to do an Excel. One of my favorites for for doing that is a piece of software called tableau. Tableau software is next is now owned by Salesforce, right?
John Wall 21:02
Yeah, yeah, part of their, I don’t know if it’s part of their digital 360. But yeah, it is definitely one of their clouds.
Christopher Penn 21:10
So you could take this exact same data, and try to put what are called trend lines onto these charts. Now, trend lines can be a little bit tricky. Most of the time you want to be using while you want to look at the data and then trying to find a reasonably close fit. So let’s go ahead and pull up this. This year, there’s our cmcsa data. And I’m going to change the trend lines to what comes out of the box. So let’s say out of the box, it comes to the linear model, right? This is not super helpful, right? That line being drawn down to the right doesn’t really tell a good story of what’s actually happening. So we want to find what are called polynomial lines, or curved lines and try and fit a line to it as as reasonably a pot as possible without you know, going completely off the rails. So really, about that third or fourth order polynomial can see fits the curve pretty well. And we can see here at that point where the curve line changes inflection, right? Again, this is a point where you’re buying like, yeah, I might want to start buying in here, start doing something here. By the time you get to here, what’s going up and up and up. Like, yeah, you really have to have been, you know, doing something getting getting ahead of your marketing.
John Wall 22:30
Yes, that’s just like classic can train investing, you know, it’s like when things are really bad, that’s the time to go out and buy because you could ride that whole way up. If you wait for that curve to turn, you’re still what you’re losing about almost five whole months worth of upswing there.
Christopher Penn 22:44
Exactly, exactly. A more practical example, for a lot of retailers, if you were to put in gift guide here. This is pretty straightforward, though. This is from from last year. This point here that inflection point in the trend is where you probably should have been doing something right. So at this point here, October 25 of 2020. If you wanted to make the holiday season merry for your marketing, that would been the point to get into your your marketing into markets so that you take advantage of people searching for gift guides here. If you work in public relations, for example, and you know that it takes what three months for a magazine to accept a pitch? Well, if it’s October, then you had better be pitching for the you know, the October issue back here in June. Right. So for like a PR perspective, if you’re not pitching Gift Guide stuff. Now, for 2021. The boats leaving?
John Wall 23:41
Yeah, you’re not going to be in the magazine.
Christopher Penn 23:44
Exactly. The boats leaving the dock right now today, you need to be you need to be audit, if you want to hit this exact same curve for 2021. And this is again, one of those things where if I zoom up to five years, there is a obvious it. Exactly. It’s a pulse. It’s a retail pulse. Right. So if you know when that’s going to be happening, you know how to get in front of it. One that’s a lot more challenging, because it moves is Prime Day right now you can see here, there used to be semi regular than last year as a huge delay. And now it’s coming back. I think it’s coming back. What a couple of weeks, right? Yeah, I think July at some point so bad. They’re trying to get back to that like half Christmas date. Exactly. So in the past Prime Day sort of peaked on July 14, and it was a month beforehand that it had done really well. So each of these times, you know Prime Day here was an October 11 and September 26 was where that inflection point. So again, if you know from past experience, if you’ve got seasonality to a trend, you know when to get out of it, you’ve got in this case, you know prime days needs about a month worth of advanced time, a If you want to hit it and you know, like one of the things Katie always says, if you are not, don’t have plans written out and scenarios written out, this would be the time to do it so that if Amazon says, hey, it’s probably the day in a month, you’re like, Okay, pull out the Prime Day emergency binder and execute all the marketing.
John Wall 25:15
Yeah, this amazed there. It looks like if they really did bite them hard that they went that late with it. I mean, they lost two whole years of momentum over that, which is insane, given how much e commerce exploded during the same period.
Christopher Penn 25:31
Exactly. Yeah. You would think in a year when everyone’s sitting at home? Yeah, yeah. It would have done well, but a you know, it’s it’s like the old joke who always say when when was Seinfeld on? You know, what day and time and people still remember, Thursdays? Thursday
John Wall 25:48
nights, he must see TV? Yeah.
Christopher Penn 25:51
And why do we know this 25 years later, because he was regular, predictable and valuable. And when we’re dealing with trends, you have a if you have things regular and principle valuable, and then you break the predictable, you’re in trouble.
John Wall 26:05
You lose the momentum.
Christopher Penn 26:07
Exactly. So we’re gonna leave you with this thought, do this kind of analysis in your web analytics, do it for like, traffic on your website? And and the reason you want to do that, is that you want to know, what are sort of the extremes? Like how much does something change when you have an anomaly, when you have a breakout when you have a trend happening? What extreme is that? You know, so if we were to look here, let’s go ahead and just go into basics of acquisition here, as I’m using Google Analytics for when we want to do this in Google Analytics for for us, the our big swings tend to be about a 200% Difference Day to day, you know, in one day, and seeing a lot more traffic coming in the next day. We want that anomaly and trend detection knowledge, though, what is the biggest percentage change, because what we’re going to do is go into let’s go back to home here, in your Google Analytics, four Academy, you can do some three, but the software does a much better job of infor go to insights going to go to view all insights. And now we’re going to create a new automated insight. And you can see there, there’s already some anomaly detection built in here. But these are Google’s settings. So I want to create a new one, evaluated daily, I’m going to do users, total users, and I want it to be a percentage increase more than 200%. On a previous day, and this is our own anomaly and trend section. This is the call your lawyer because things are looking bad or good. Exactly. Well, so this is a percent increase of more than 200%. So I know, I know, I want to see, I want to know, it’s fine. If Google tells me, like, Hey, we think there’s an anomaly. But I know from our own our own data, as in background, I want to know that specific thing, tell me when that happens, because it’s important. We also do have another version, called Yo, who broke this, and who broke it is, you know, when you have a 90% decrease from the previous day, you know, it’s like the tracking code went missing. This is the so what of trend analysis, when you have the ability to look at trends, digest the data. And then you set up alerts to say like, I want to know when we see some big spikes that we can get when so we’re not looking back, and we going, Hey, we got a lot of traffic last week, maybe we didn’t do anything with it.
John Wall 28:46
You know, staying in front of it isn’t that’s, that’s it. And if you can be there, as it’s happening, you can actually even stoke the fire, if it’s something that you could get involved with it, you know, you throw a discount on the pile or whatever.
Christopher Penn 28:58
Exactly. One of the things you can do is you could set for hourly. This is really game, you know, really, really tricky. If you want to do that our link is like a 500% increase hourly compared to the previous hour. That would tell you like, hey, something’s really hitting us hard. The caches, you should do the math first. Because if your site has one visitor at 6am and 10 at 7am gonna get swamped with alerts, you
John Wall 29:23
can get tripped with that all the time. Yeah, exactly.
Christopher Penn 29:24
Exactly. But yeah, you can do you can take action on these trends. And that is the so what is when you detect a trend, or even an anomaly that you think might become a trend, have your software tell you hey, raise my hand here. You might want to take a look at this. You might want to see what’s going on, you know, is a bot hitting your site. It would be a good thing to know if our customers hitting your site. I don’t know so no use traffic but certainly like conversions like if you’re an e commerce company, and you don’t have like a 5x detector in your you know, sales like Turn it on the day. Because you should know, hey, we made 5x more sales in the, you know, this hour they did the previous hour or whatever the biggest jump is that you have in the past that will tell you Yeah, we need to, we need to wake up and pay attention to what’s going on the website. do more of that. Yeah. So that is determining what’s trends. We talked about the mathematics, the change, 70 changes, changes or changes, the moving average convergence, divergence indicator, and how to take action on it, either for things that have high seasonality and cyclicality measure the inflection points of trends in the past and figure out what is your lead time that you need to get in front of a trend? for things that you know are going to happen? I think so you don’t know what are going to happen. have built into systems like Google Analytics for, hey, I’m raising my hand. Something’s happening. Please do something about this. Any any parting thoughts on trend management job?
John Wall 30:58
No. Again, it’s the same as every week, go grab some data and go dig in and find it. Because you know, unless you have these alerts set up, you’re again going to be doing that two weeks later, looking at the numbers saying Geez, I wonder what happened right there. And that’s you’re not in the driver’s seat if that’s the case.
Christopher Penn 31:12
Exactly. Alright, folks, if you’ve got questions and stuff, head on over to the slack who have asked otherwise, we will see you all next week. And let’s roll our 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 but Trust Insights podcast at Trust insights.ai slash t AI podcast, and a weekly email newsletter at Trust insights.ai slash newsletter. got questions about what you saw on today’s episode. Join our free analytics for markers slack group at Trust insights.ai slash analytics for marketers. See you next time.
Transcribed by https://otter.ai
Need help with your marketing data 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, Data in the Headlights. Subscribe now for free; new issues every Wednesday!
Want to learn more about data, analytics, and insights? Subscribe to In-Ear Insights, the Trust Insights podcast, with new 10-minute or less episodes every week.