I hate to be the bearer of bad news, but marketing’s data honeymoon is over.
For the last few years, every time a higher-up asked for a progress report, marketers have been able to point to a growing list of easy-to-count metrics: increases in Retweets, Pins, Likes, upvotes, and impressions; lower costs-per-click; improved engagement; and so on. It was grand while it lasted.
But bosses have started demanding something better. The aforementioned metrics might be easy to track but they don’t easily convert into effective strategies.
It doesn’t matter if you’re a marketing content manager trying to sell your director on a new strategy or a CMO trying to convince your CFO and CEO to expand your budget. Up and down the marketing spectrum, we all face increasing pressure to make data-driven decisions— and that no longer means merely tallying vanity metrics that seem (but have not been proved to be) important. Today, if you want to persuade the bosses, you need stories built from cleaned datasets and intelligently-validated correlations, not just the basic counts that are so easy to find in Twitter, Facebook or Google Analytics.
How can marketers rise to this challenge? Here are three tips to get started.
Too often, marketers assume bigger is better. More page views? The content must be doing a better job. More people viewed a product video? Sales will probably increase. Cost-per-click is lower on LinkedIn than Twitter? We should flood LinkedIn with our message.
But here’s the problem: Almost none of marketing’s favorite metrics are self-evident in their value, especially by themselves.
Hollywood analysts love to discuss how many trailer views upcoming films have accrued, for example. Unfortunately, if you’re a studio marketing exec tasked with maximizing opening weekend grosses, trailer view counts don’t help much, as we’ve previously noted. This metric is at best a weak indicator of future ticket sales. In fact, at Quantifind, where we’ve worked on several of the biggest movie campaigns in recent years, we’ve found that whether a user tweets about needing a babysitter is a much stronger signal than whether someone tweets about a new preview.
Likewise, Netflix recently announced that The Ridiculous Six, its new comedy western starring Adam Sandler, set a company record for first-month viewers. As a PR move, Netflix was probably wise to promote this accomplishment, which surely lured in viewers who were on the fence. At the same time, it’s not clear how this data should inform the company’s longer-term strategies. Are all those Ridiculous Six viewings going to translate into subscriber renewals or new customers? What was the film’s true bottom-line value?
Even relatively advanced metrics are less valuable than they appear. As alluded to above, many marketers invest heavily in cost-per-click metrics. This sort of analysis is more useful than looking at clicks alone— but is a cheap click preferable to an expensive one? All things being equal, sure, but in the real world, the cost of a click is often less important than who clicks and what they do afterward.
The takeaway is this: Even the most ostensibly persuasive metrics are incomplete. If you can’t correlate your metric to core KPIs, such as revenue or user growth, the metric isn’t as valuable as it appears.
Social media is a deeply polluted data source, which means engagement metrics built from the entire dataset are full of garbage. If you’re going to make strategic decisions based on social data, the data must be thoroughly cleaned— which is a far cry from the lightly filtered or unfiltered results you’ll find in a social platform’s analytics dashboard.
Here’s an example. Last month, Coca Cola launched a new marketing slogan, dropping “Open Happiness” for “Taste the Feeling.” Some subset of consumers’ social media responses can help the company to understand if this switch was the right call— but to find the meaningful data, Coke will need to throw out a lot of junk.
And make no mistake, there’s a lot of junk. Below, you’ll find some of the top results when I queried “Coke” on Twitter. These aren’t cherry-picked examples— they’re the ones that topped the feed, and also generally representative of the hodgepodge of “Coke” results one can find on any given day.
Obviously, many of the tweets have nothing to do with actual Coke products. Even among the tweets that are actually on-topic, only a fraction offers insight into how consumers feel about being Coke customers. The company can’t just assume an increase in Coke references or Coke-related retweets is some kind of proxy for the new slogan’s success.
To get a more accurate idea, Coke would need to filter the raw data to include only content that obviously comes from real people offering their organic, unaided opinions about Coke as a product. That means no marketing messages or spam, no song lyrics, no references to “coke” that involve drugs instead of beverages, etc. This sort of data-cleaning is difficult because it demands sophisticated data science that goes far beyond tallying up a few metrics— but it’s also the only way to ensure your insights reflect real people talking about a real product, rather than the cacophony of online noise.
If your data hasn't been cleaned, there’s no use even looking for correlations. But even after cleaning, analysis isn’t simply a matter of layering graphs from one dataset atop graphs from another. The surest way not to build an intelligent correlation is to assume that just because Twitter engagement and revenue both increased, your social strategy somehow moved the financial needle.
True, if a line measuring social trends and a line measuring financial records are on the same xy-axis, you’ll probably spot a few ostensibly compelling correlations. Sometimes the lines will move up and down together, giving the impression of causal relationships— but finding truly validated correlations often isn't so simple. When you’re dealing with massive datasets, you’ll inevitably find some relationships that seem persuasive but are in fact spurious. For example, you might have seen the chart that shows a seemingly strong, decade-long correlation between the number of films featuring Nicolas Cage and the number of people who drowned by falling into a swimming pool.
Instead of just looking at overlaid charts, marketers should instead invest in rigorously validated correlations— that is, in using correlations to build mathematical models, and then testing those models against past data in order to assess their predictive and explanatory power. If consumer signals and sales data seem to strongly correlate throughout November and December, for example, does the correlation reveal actionable information about sales drivers, or does it speak more to obvious seasonal lifts?
How do I define the value of a metric or combination of metrics?
How is my data cleaned before deeper analysis?
How do I validate correlations for future value?