Whether its analyzing digital advertising trends for Skai’s Quarterly Trends Reports, finding data points for the Client Services team to share with interested customers, or fueling our content with compelling stats, Chris Costello—or “Coz” as he’s known to his friends and colleagues—has been deriving insights from Skai’s massive advertising dataset since arriving here in 2014.
What’s been keeping him glued to his computer monitor for the last few months has been a constant analysis of Search, Social, and Ecommerce advertising data to help make sense of how advertisers and consumers are reacting to the global COVID-19 pandemic. Drawn from an annualized data set spanning nearly $7B in marketer spend, over 750 billion ad impressions, and 15 billion clicks, Coz has been fielding questions from marketers, financial analysts, industry publications, and Skai customers that want into what he’s seeing and—more importantly—what he might be able to predict in the near future.
With an analyst career spanning nearly 30 years, Coz has seen it all—almost. In these unprecedented times during the pandemic, he’s been tasked with answering almost unanswerable questions. How bad is it getting? Are consumers clicking ads again? Are there signs of recovery? When will things get back to normal?
To check out Coz’s work in this area, visit Skai’s COVID-19 Marketing Resource Center for his most up-to-date research and analysis. He’s also appeared in numerous publications and webinars recently as the entire industry copes with this global issue.
Given the unique nature of his work over the last few months, we decided to sit down with him and get more insight into what it’s like to be an analyst during this time.
Five questions with Chris “Coz” Costello on COVID-19 advertising research
What is different for you, or for the field of research and analytics in general, in terms of how you’re analyzing data during this crisis?
Coz. Personally, I can say unambiguously that I’m asking a lot more of Skai data. That means I have to adopt a higher standard when it comes to things like data quality and data hygiene. Imperfections or sources of error that you might otherwise overlook can be a bigger headache when there’s more at stake, so I find myself looking at assumptions I have about the data, making sure they’re correct, and thinking about how, with some tweaking, they might inform a better analysis.
The same goes for KPIs. Maybe you’ve known they could be better and have been putting off fixing them, and this would be a really good time to actually do that. I’ve said over and over that digital advertising needs to get back to fundamental principles—right time, right place, right message—and this is pretty much the same thing for analytics. Is your data as accurate as it could be? Are you looking at the right metrics? “Close enough” isn’t going to be good enough during a pandemic.
How much of your analysis is automated, via dashboards and other data processing or visualization tools?
Coz. My workflow is typically messing around with large datasets in more basic tools like Excel, and then once I’ve got a better idea of what I’m trying to do, I’ll try to build as much of that as I can in Tableau, so it’s more easily repeatable, scalable and shareable.
But that only goes so far, mostly because I’ve been using Excel for, like, twenty-five years now and Tableau for less than five. So I may do the basic views in Tableau and then dump summary-level data as a crosstab to get things to really sit the way I want in Excel. It may take a bit more time, but one thing I like about this hybrid approach is that you’re not relying on the tools to do all of the work. When you’re a bit more hands-on with the data itself, pulling it apart and putting it back together, it ends up being the equivalent of how writing something out longhand is better for memory than, say, typing. If your goal is understanding the data, I feel like there’s no real substitute for getting your hands dirty, and I’m not just saying that out of self-preservation in an era of machine learning. At least I think I’m not. I like having the robots on hand to do stuff! Just not all of it.
In the Skai Q1 2020 Quarterly Trends Report, there is a “Signs of Recovery” section. How did you approach this analysis?
Coz. There’s a real danger right now of being overly reactive to transient spikes in the data, in either direction. So while week-over-week trends are important, they are very short-term, and you don’t want to mistake a spike for a trend. Including a medium-term and long-term view of those trends gives a bit more context.
Last week may have been up versus the previous week, but how did it compare to the average over the last four weeks? What about month-over-month, or a four-week period over the previous four weeks? We look at all of those, first for aggregate spending in a category, and then for other metrics that might tell us something useful. Impressions in the paid search and ecommerce channels reflect product searches and product views in a category, while CPM in paid social will increase as advertisers see more value in getting in front of particular audiences.
At the end of the day—or more specifically, at the end of every week—it’s a combination of short-, medium- and long-term views across spending and other KPIs that start to draw a picture of where the recovery might be more sustained, or where it may have turned things around more recently.
In the Quarterly Trends Report, you projected what the Q1 2020 channel growth might have been without the impact of the pandemic. What approach did you take to figure that out?
Coz. The first approach was to simply compare spending from March of last year to both January and February of last year, and then apply those two ratios to January and February data from this year, since spending was humming along at a normal clip at that point. Then we got a bit more granular and looked at the year-over-year growth by week for the first two months of the year, which gives you more data points to consider. So, it’s a more stable trend and a more targeted calculation.
Applying that year-over-year growth trend to the specific weeks in March that were affected gave a bit of a tighter, more conservative estimate. So that allowed us to give a range for Q1 growth because it seems kinda silly to die on the hill of a specific number when you’re making an educated guess.
From an analytical perspective, how relevant is historical data in a COVID-19 world?
Coz. Well, it still provides the context of what “normal,” or what used to be “normal” looks like. So you can look at how you are trending as an individual advertiser, or as a channel in aggregate, and at least get some sense of where you are, rather than getting too distracted by the very short term trend.
Even in the middle of March, when digital ad spend was starting to decline, we were able to look at the same period last year and see that at least some of that decline seemed like a normal slowdown that we would see in late Q1 and early Q2. So if the impact of the virus was even just a little less than it looked like initially, that’s something that may help understand how the normal levers and normal seasonality may affect the recovery.
Check out Skai’s COVID-19 research hub
Visit Kenshoo.com/Covid-19-Resources/ to check out more of Coz’s analysis and the most up-to-date information.