The Super Bowl is the largest annual live televised event in the U.S., with viewership averaging 162.8 million over the past five years. Advertising rates for a 30-second advertisement during the upcoming Super Bowl broadcast are reportedly $4.5 million, dwarfing regular 30-second ad rates for non-sporting events (in 2014, spots for AMC’s “The Walking Dead” sold for $623,000). Advertisers have developed a range of methods for measuring ad success including views, social media interaction, YouTube hits, meme spinoffs … the list could go on.
If you are an advertiser, you’d want your advertisement to be successful, but wouldn’t you also want it to be effective? An ad can be remarkably successful, but not necessarily effective. Other ads can be effective, but not successful. The key is defining what success is and what effectiveness is.
An ad might be very entertaining and hit all of the digital standards for success; however, it might not help the business model of the organization behind the ad: Did sales increase? Did the company’s market value change? Did the ad extend the availability of the market served? How fast did this happen? Immediate? Delayed? Did it “stick” — in other words, did the ad transcend the moment to become part of the culture?
The question is, how do we analyze the synergy between success and effectiveness?
Let’s start with success, since advertisers are already measuring the raw digital impact of social streams.
One common media success tool is trend analysis. But what if we wanted to add some complexity like a “distance in time significance factor”— looking at ad performance over time or around a game event (during a timeout, for instance). We might also want to examine the social media content for sentiment (is it good, bad, indifferent?), and we also might look at sentiment not only by key word, but also in context (a phrase like “That ad was very bad” can mean either “I liked it” or ”It was terrible,” based on the individual, the thread of comments, the message as a whole and the larger community response).
Effectiveness in sales and branding is a bit different, and the impact longer term. It takes time for ad viewers to go to the store to buy soft drinks. And it takes a number of repeat visits to the store to gauge changes in sales. Even in a digital sales world people buy over time, and repeat customers show a change in behavior.
To measure the historical effectiveness of ads on product sales and brand recognition, we could look for ad ”characteristics” leading to a clustering of types, and a clustering algorithm iteratively worked and refined could be used to find ads that were most effective. However, for a business the proof of ad effectiveness is in the financials. And brand recognition is more a trending measure. For example, we might see some impact to things like stock price in a publicly held company, which might be an indicator of a future anticipated financial impact; but the mechanics are that people still need to engage in some sort of action, like buy stock or write a forecast.
Building the characteristics of a successful ad from a historical perspective, I propose, is not meaningful. People and cultures change. Expectations change. Population behavior is fairly predictable. However, the triggers of response change dynamically. For example, in 1967 (Super Bowl I) chauvinistic ads may not have created consumer backlash, but today the possibility of backlash would be tremendous (and might actually make an ad more successful – even with negative sentiment).
An algorithm that measures a successful ad must be a near-real time activity, calculating success across a continuous feed, dynamically weighing the benefits of gross digital hits against the driving forces of positive and negative sentiment. Lots of great thinkers and products do these types of analyses individually: context, sentiment, trending, impacts on sales and brand.
The tough part is to marry an empirical/historical discovery analysis measuring effectiveness with near-real time success measures. Let’s take a crack at it.
For historical effectiveness, we’ll use an empirical approach – leveraging data collected from ad viewers wearing specialized sensors (there are several groups that take movies and video to ”reproduce” data). We could add additional data types (e.g. weather) where at any given time, and for any given set of variables and conditions, a static, empirical (or close) data set exists. This model for a historical view of ads should have a lot of dimensionality and encompass as many factors as possible.
It turns out that whole-graph algorithms – algorithms run across the whole of the graph, across all of the data in the database – are good for pattern and cluster ”discovery” on empirical datasets. Data and queries run though lots of different models. For example, weather in New England after a Super Bowl could clearly impact sales of beer in Boston … especially if New England is in that Super Bowl. With the right tools and compute capability we could build queries on data that takes this type of thinking and applies it with some form of ”universality.” We would test to see if weather ‘mattered’ or whether ads with a certain set of characteristics are more or less sensitive to weather. Or whether certain products were more or less sensitive to weather?
Taking it one step further, we could start adding success measures, applying standard digital captures (some are mentioned above), layer in some dynamic calculations to determine the accuracy of standard data sets (like demographics), run a continuously updating stream of data, and finally calculate ‘ad success’.
OK, so in a few paragraphs I’ve described a process that should take a five-page white paper just for the concept. But, the idea is what matters: Near-real-time data not only measures the success of ads as they are viewed, but also updates the way in which success is measured.
We can find what is effective and rectify the various values an ad can have to an organization or company. We can do the same in terms of success as it impacts the population. For an advertisement, the marriage of business impact “effectiveness” with cultural “success” is where synergy – effective advancing success and vice versa – is to be found (and to my mind, where you will also find the best ad).
To learn more about how Cray can help with analytics such as this, join us at the upcoming Strata+Hadoop World in San Jose, Feb. 18-20. Reserve your seat using code Cray20 and save 20 percent on passes.
The post Super Bowl XLIX | Improve Effectiveness of Your Ad with Graph Analytics appeared first on Cray Blog.