When you discover important insights in your marketing attribution data, it can be challenging to explain them to other people within your organization. When you’re working with multiple channels, touchpoints, and models, the data complexity can easily overwhelm even data-savvy audiences. In this session, we’ll look at how data visualization techniques and tactics can help you communicate your attribution insights more effectively so they resonate with key stakeholders and lead to better marketing performance.
Brent Dykes (00:16):
All right. Welcome to this presentation on how data visualization can amplify your attribution insights. My name is Brent Dykes and I'm Senior Director of Insights and Data Storytelling at Blast Analytics. So if we look at the history of marketing to really understand where we've come from, we have to go back to John Wanamaker, who was an American retailer. He was one of the first marketers or advertisers who really seized this opportunity to advertise his products, his retail products at his store. And he actually, I think he was accredited with having the first half page or full page ad. He also hired the first copywriter. So he was really innovative and he had this statement. I think a lot of marketers have heard this statement before—half the money I spend on advertising is wasted. The trouble is, I don't know which half. Well, if we fast forward to today, we don't have a problem with data, but now maybe half of my money I spent on marketing is wasted.
Brent Dykes (01:18):
The trouble is, I don't know if my attribution model is right. And so this is the challenge that we have today. And if we think about analytics, I'm going back to my book that I wrote about eight or nine years ago, I had this path that you see here, it's called the analytics path to value. And what it basically is, is you collect all your data. You basically put that into reports. You analyze that data, you find insights, and then you use those insights to influence decisions that lead to action and then value. However, what a lot of the time happened, and I didn't actually realize this until later, is that often when we have these insights, we don't, they don't influence the decision. And part of that is because we haven't communicated them effectively. And so one of the key things that I've discovered is that we need to tell data stories.
Brent Dykes (02:09):
We need to take those insights that we've uncovered and then share them with other people through effective data communication, often the form of a data story. And then we can drive the decision and action and value from the data. So why do we need data visualization, right? Maybe you're comfortable using spreadsheets to analyze your data. Well, let me give you this example here. So what does this data table reveal? If you look at the pairs of X and Y coordinates, there's four pairs here. And if you look at them, they're actually, they have the same mean, they have the same, Y and X mean. The variance is similar. The R and R squared are very similar. So the summary statistics are the same. And so you might notice a few interesting things in the table, but it's only when we visualize the data.
Brent Dykes (03:00):
And this actually goes back to a statistician who was looking at how other statisticians weren't really using a data visualization. It was more like a toy. It wasn't really a tool. And he wanted to show that the data visualization actually can shed light into anomalies and different patterns in the data that if we just focus on the summary statistics, we won't really grasp the full picture. And so visuals really do matter. Now there's actually two parts to visualization. The first part is where we go into the data. We were exploring the data, we're trying to find these insights. And so that's the exploratory side of visualization. And then once we find an insight, we then need to communicate it with other people. And that's where we may need to change the visualizations that helped us to find the insight into something that's going to communicate a little bit easier and a little bit better for other people who haven't spent days or hours or weeks in the data like we have.
Brent Dykes (04:05):
So one of the things we need to understand when we're looking at visualization is really understanding human perception. And I like this quote from Colin Ware who said, if we can understand how perception works, our knowledge can be translated into rules for displaying information. I want to share some of those rules with you today. So there's the gestalt principles. And these come to us back from the previous turn of a century, a German psychologist, they are looking at human perception and how we group things using different rules or principles. And so if you look at these here, proximity, similarity, closure, figure-ground, continuity, closure, all of these are examples of how we group information. There's also something called preattentive attributes. And I look at this as how we highlight a particular element within a group of items. What makes something stand out? And so here we can see color position, alignment, size, intensity, orientation, shape, motion, and so on. All of these help us to really pinpoint or see a particular data point and help it stand out.
Brent Dykes (05:16):
Now, this is a great quote from Edward Tufty, who's a data visualization expert. And he basically said, we're always trying to compare things. And it really comes down to making and showing smart comparisons. And so that's how we're going to enable the audience. And sometimes when we think about visualization and data storytelling, we may think, oh, we need to have these really complex visualizations. When actually, when it comes down to the communication side, actually the charts can be very simple, especially because what we're trying to do is we're trying to facilitate comparisons. And so there's a number of, there's actually, in the mid eighties, they did some analysis of how different methods, graphical methods, could be more precise in facilitating comparisons that are more accurate. And so a couple of statisticians looked at making this more objective rather than subjective in how we use data visualizations.
Brent Dykes (06:20):
And they did the study and they found that position along the common scale or underlying scale was the best or the easiest for people to do comparisons. And then as we move from left to right, we have lengths, direction, angle, area, and then we go to curvature volume, shading, color saturation, where we can see more high level patterns, maybe in a map, but in general, we're going to see on the left, those are going to be the most effective for very accurate comparisons. Now, if we think about attribution, it's all about comparisons, right? We have a whole bunch of different attribution models that we might be comparing. There may be different marketing channels that we're looking at to see which is effective. And then we break that down further by looking at potentially the type of content that we're putting out there, how the marketing content response, or how different segments are responding to that content or through the channels.
Brent Dykes (07:18):
And then the individual touchpoints that may be going on and doing a lot of comparisons, a lot of analysis in this. So today I'm going to look at some different tips and tricks for data visualization. First area I'm gonna look at is attribution data in tables, charts, and diagrams. And so here, we've got kind of a basic table here. We're looking at different marketing channels, and we're looking at the different models and we're making comparisons. And we can look at this data in this table and we can see maybe some different patterns, different things that are happening, but what if we add color to it? And all of a sudden, now we can see those peaks and those valleys to see really which ones stand out here using some conditional formatting. Now, another point I want to make too, is that because we're sharing this information with other people who may not be as familiar with these different data models, these attribution models, it might be good to include a thumbnail. So you'll notice above each of these labels here, I basically got a little thumbnail that indicates what that model, how it kind of basically functions. And so, as you're sharing this information with other people, they can quickly grasp what that model means.
Brent Dykes (08:32):
Now if we're to visualize this same information, we could have a chart like this, where we've got basically the different attribution shares by the different marketing channels by the different models. And we see, what do we see here? We see a collision of colors, right? It's very hard to really make sense of this data because there's so much going on. And so I like to call this the visualization kaleidoscope effect, where, because we have so many different data points, it could be channels, it could be campaigns, it could be segments. You know, we have a lot of variety. And so we assign different colors to each of these. And then we get this kaleidoscope effect where it's hard for us to maybe see the insights or even share those insights. And if we use gray scale, which is something that's very powerful, we can kind of pull the colors.
Brent Dykes (09:21):
You know, we can highlight in this case, the email and push everything else to the background, but still, we still have a problem with this visualization because there's no baseline, there's no common scale. So it's very hard for somebody to really even judge, which of these models here actually depends the most, or has the highest attribution share for email. Now we could use other charts like a pie chart and could have multiple pie charts here. But again, we're running into the same kaleidoscope problem. We're running into the same baseline problem, because if we want it to look at display in this case here, they're all positioned at different angles. And so it's very hard and very hard for us to know. And so we have to label things. And again, if we have to constantly label visualizations because they're not communicating effectively, we have a problem.
Brent Dykes (10:10):
Now we can maybe fix this a little bit by having establishing a common baseline, which would be 12 o'clock position with these pie charts. But still it's not as effective as it could be. Now we might want to try a bar chart or like a column chart in this case, or maybe we might want to do a horizontal chart in the good thing about bar charts. Going back to what we shared about the different graphical methods of which ones are more precise. In this case here, we can easily see which of these is basically contributing the most, or has the highest attribution share for the different models. Now what we might want to do instead to really kind of break away from the kaleidoscope effect and still have that baseline that we need to kind of evaluate these, this is using a panel bar chart. And so in this case here, each of the models here has their own distinct color, obviously separating them. And we can look up and down to kind of see where most of the attribution is coming from. And this might be an effective way to kind of break out the content, not overwhelm somebody with too many colors in the same place, but still communicate something effectively.
Brent Dykes (11:31):
Now there's something interesting about, and unique to how attribution, you have these conversion paths or customer journeys, where you're looking at how people interacted with different channels or campaigns. And so in this case here, the editor default, you know, say, take Google analytics. You're going to see something like this, where you see these different channels and how you interact with them. Well, maybe what you might want to do is add an icon again, to help people to understand and quickly absorb the information. And so adding an icon or image can really help to help people orient. And then going back to the kaleidoscope effect, if we're communicating something, maybe we strip out all the color and we just use color selectively or strategically where it really matters. Now, another thing we can do in this case here is we can take out all the words and just use icons, especially when some of these conversion paths might be quite long and you're trying to tie together multiple of these and you can get lost and it can get very complex. And so, again, we've stripped out the color. If we had the color, it would look like this. And again, we might have that kaleidoscope effect that kind of gets in the way of our messages.
Brent Dykes (12:47):
All right, so we talked about some of the different options that we have, or some of the different ways that we can visualize the data. Now, we're going to look at some other data chart, design considerations. So what we might want to do is obviously limit how much information we're sharing at one time, rather than looking at five or six or seven or eight different models at the same time. I mean, you can do that on the exploratory side, but when it comes to actually communicating something, you may want to just look at the two models that you really considered as maybe being ones that you want to compare against each other. Maybe you might go as high as three models, but once you go more than three variables, it gets kind of harder to follow and more complicated.
Brent Dykes (13:34):
The other thing we can do is we can isolate trends to kind of reduce the noise. And so in this case here, we're looking at row S by campaign and you have what sometimes can be called a spaghetti chart, but you have all these lines that are being trended over time. And sometimes it might be better to actually separate out these into their own charts and because they have the similar scale it's very easy for us to kind of compare them and make see which ones are higher, lower, how they're trending without the overlapping, which can create noise for some some people in the audience. Now, in this case here, if you're using the bar chart, what are we doing when we're looking at, when we're doing the comparison, what we're looking at, the ends, or the heights or the lengths of those bars, and we're comparing them to see, you know, what's the difference between the lengths of the heights.
Brent Dykes (14:28):
And so going back to the graphical methods, that is an effective way, but there might be other ways that we could do this and still show some key points here. So in this case here, we have something called a dumbbell chart. And so in this case here, we're just isolating the differences between the two end points. And so we can see here quickly, you know, looking at last touch and linear, which one was higher for each another option is something called a slope graph where you basically have the two models. And we're now looking at the slope of the lines to kind of see, okay, did that, was that higher or lower for one of these channels? And so we can quickly look at this and see which one was higher or lower, depending on which, which model it was. And so that's another option.
Brent Dykes (15:15):
So there are options out there rather than just bar chart, bar chart, bar chart.You can use some of these other ones that may serve a purpose and help you to communicate your information more clearly. Now, the last area I want to talk about is really about some of these other data storytelling considerations that really tied to communicating your insights. So going back to the slope graph chart, one of the things I did earlier with the email is you use color. So if I was making a point about our organic search, I put that in red and I push everything else to gray scale. So it's there for context, but really what I want the audience, and I want you to focus on, is the difference between organic search, whether it's last touch or linear, or maybe I want to highlight a different point. Maybe I want to highlight the difference between email and linear.
Brent Dykes (16:04):
The next thing you can do is, often what you're doing with your analysis is you might be drilling into the data. So you might start at the marketing channel, drill into the content type and then further drills. And so one of the things we want to do is show people how we're drilling down and give them the context of where we're going. So in this case here, we're going to highlight, okay, we're going to drill into email. Now we're using one of these gestalt principles of enclosure, right? So I'm highlighting this area. I'm also using color as a preattentive with the blue compared to the gray. And then we're going to drill into that, and now we're going to drill into, okay, so we're looking at email, we're looking at linear and where did those leads come from?
Brent Dykes (16:46):
And so we see, oh, okay. The vast majority of them came from webinar as opposed to blog or white paper focused content. And then if we drill into that further, we might go into this. Okay, well, we knew that webinars were really effective, which webinars drove most of those leads. And then we see, oh, okay, it's this content around a difference between ROAS and ROI. And in this case here, what we may want to do is, as we look at this content, we might say, you know what, do we really need all this additional content? Do we need to list out the information around the top blogs or the white papers? In some cases, yes. Maybe that's valuable information to share, but if it would just be noise to get in the way of your message or confuse your audience or lead to questions that you don't want to go down, then maybe you remove it. And so your strategy may be just focusing really clearly on just the specific content—webinar content—that was effective.
Brent Dykes (17:44):
Also, one last thing we can do that I want to highlight is using text. And so we've already used color to kind of highlight what we want you to focus on, but as you can see here, what we've done is rather than using just a generic title for this, we can actually go into specifics and we can reinforce the key point that we want the audience to grasp out of this. So we're showing here that 63% of the leads came from webinar emails for the linear model. And then, of those, 85% of them came from this row as ROI webinar. And so, as you can see here, the text reinforces with the color to really guide an audience into your insights and summarize what we've covered here. You may want to ditch the defaults and you want to basically look at maybe using different charts to communicate your insights more clearly. Sometimes we get different charts out of tools and we try and repurpose them for communication. And sometimes that actually leads to problems. And we really aren't communicating as clearly what could be a key insight with those charts. We want to use color strategically. We want to make sure that we're highlighting where we want the audience to focus. And so if we're looking at referrals in this case, affiliate referrals typically happening earlier in the stage, that's where we want to focus the attention.
Brent Dykes (19:10):
We may want to remove the noise, right? So that might be separating data, that might be removing unnecessary, additional data or context that gets in the way of your message. And then lastly, you want to highlight where you want the audience to focus. So that could be through bold fonts, that could be through text, that could be through color. You have a lot of different options there, and they all tied to those preattentive attributes that really draw our attention to key points. So I want to leave you with one last quote. This actually comes from Stephen Few. If you're familiar with data visualization, he's an expert in this area, and he talks about numbers, having an important story to tell. And so you have these insights that you've gotten from your attribution data, and now it's up to you to give them a clear convincing voice, and I'm kind of a Spiderman geek. And so with this great power comes great responsibility, but I would also add great opportunity. And so I would encourage you, if you'd like to learn more about how to tell stories with your data, please pick up my book, Effective Data Storytelling, which was recently published. It's available on Amazon and other booksellers. And I look forward to any questions that you may have about data visualization or data storytelling.