Calculating the Optimal Emotional Language for Email Success
"The reaction to, and response of, an email message has the same power to conjure up an emotional response as does the spoken word, and the response or behavior we influence is the action, or lack-there-of, seen in our customers. But how do you know what to say to create this emotional reaction, specifically in your email marketing communications?"
By David Atlas, CMO/Persado, and Julia Spano, Director, Marketing/Persado
Joy: I’m Joy, this is Sadness, that’s Anger, this is Disgust
Joy: And that’s Fear
Fear: Ahhh! Look out!
Like all of us, Riley, in Pixar’s new film, Inside Out, is guided through her daily life by her emotions. As we communicate with those around us the emotions that we evoke in others cause a reaction. That reaction triggers a response, which, in turn, drives a behavior in those with whom we communicate.
The reaction to, and response of, an email message has the same power to conjure up an emotional response as does the spoken word, and the response or behavior we influence is the action, or lack-there-of, seen in our customers. But how do you know what to say to create this emotional reaction, specifically in your email marketing communications? Unlike other aspects of email marketing where there is data available to figure out whom to send a message to, and when to send a message, the actual “message” part – the writing – is something we create without data, with no statistical insight on which emotions drive response, no mathematical basis for choosing the words we do.
As marketers, we must begin to make conscious decisions about the choice of language that we employ in our direct response campaigns because words have the power to persuade. The power of persuasive language is evident in the elasticity of message response observed. At Persado, we approach language construction mathematically, and so are able to calculate the exact range of elasticity – the difference between the statistically worst performing possible message and the statistically best performing possible message. What we’ve seen across 10,000 campaigns spanning hundreds of millions of email subscribers is that there can be up to an 800% elasticity in performance. Traditional copywriting techniques will reveal that guess A is better than guess B, but does nothing to tell you where your winner is in the statistical range of possible performance.
Turns out words matter. A lot. With machine generated copy outperforming copy created the traditional way, by a writer without algorithmic assistance, the average lift in performance observed, as compared to the control message, is 75%, across opens, click-throughs, and conversions.
Having machine-generated the language for over 1 billion email marketing messages, and analyzed language’s efficacy at a word-level, Persado sits on a treasure trove of linguistic data – data we are previewing here, for the first time ever. Using computational linguistics, we break down the language of email into four types:
- Descriptive Language. These are words that that express the finite number of features or objectively observable facts about the product or offer – “available in all sizes,” “zipper in the back,” “1% cash back,” etc.
- Emotional Language. These are the words that have some emotional valence and are grammatically attached to the person you intend to drive a response from, the reader of the email. Phrases like “just for you” are categorized by Persado in a one emotion type (exclusivity) whereas phrases like “don’t miss out!” are categorized as another emotion type (FOMO – yes, we call that an emotion).
- Functional Language. Functional language includes words on buttons and the like intended to actually navigate an email reader through an email somewhere else, such as to a landing page.
- Formatting. Not technically “language,” formatting choices include the textual and graphical choices we make - boldface, capitalization (and types of capitalization like InterCaps), the use of emoticons: these elements are also an aspect of marketing copy.
Organizing language this way it is possible to determine not only the overall elasticity in performance, but also the relative contribution of types of language and of specific language choices, to a campaign or a set of campaigns over time. What we have found is that emotional language drives approximately 70% of the overall response rate observed in an email marketing message. Pixar is on to something.
At one level it shouldn’t’ be surprising to hear that consumer behavior is more driven by emotion than rationality. Certainly this is a topic many in the broad area of marketing have studied. What has not existed before however was a way to mathematically calculate the impact of emotions on digital direct response marketing.
Persado organizes emotional language into 19 categories: positive emotions like excitement, achievement, or luck; neutral emotions such as curiosity; and negative emotions including anxiety and guilt. Organized within each of these categories are tens of thousands of words and clauses that represent those 19 emotions. Technically this is called an “ontology of emotions” in artificial intelligence, and it is necessary to organize language into a finitely structured thing if we are going to successfully calculate the impact of various types and instances of emotional language.
Looking at the data organized by channel, we can present the most persuasive emotional language used in email. Obviously this is interesting only in aggregate; a given vertical will perform differently, and a given brand will perform differently, and everything evolves over time, with most persuasive emotions waning and waxing in impact. “Your mileage may vary,” in other words.
Overall, however, we see the top 3 performing types of emotional language for subject lines are achievement, desirability and reassurance, whereas the top 3 emotions for email body are luck, achievement and intimacy, respectively.
Examples of High Performing Emotions in Subject Line
It’s time to get your reward from us - save 30% off site-wide!
Congratulations, apply free shipping to everything in your cart.
You made our list for an awesome deal. Get a huge discount on online orders.
We’ve Chosen You to Save an Extra 20%. Start Shopping.
Our Way of Saying Thanks - Get 50% Off Your Next Order
Because you’re amazing: Shop our favorite summer styles and save big.
Examples of High Performing Emotions in Email Body
You’ve really scored. Holiday deals on thousands of items.
This is a win. Get 50% off your favorite items.
Aren’t you lucky. Grab these awesome weekend savings.
Wow, today’s your lucky day. Upgrade now at a discounted price.
Dear Card Member: Save on Gift Cards for Friends and Family this Holiday Season
Hey there, we have a special offer today only. Let the shopping begin.
By the way, a caveat: we really have no idea what the actual emotion is at all. Indeed, we don’t really even know why what we are showing is so – merely that, mathematically, this is what works to generate the greatest impact. All we really know is the mathematics of language, and ultimately that the “problem” of which language to use is a solvable one. When we say “emotion,” we really mean “language agreed to hold a certain emotional valence,” not actual human experience. All we know for certain – and we could prove this mathematically – is that the words and phrases that fall under these types of emotional language are statistically more engaging than the other 16 types. All we know is these emotions, overall, drive more clicks than others.
Interestingly enough, we can also figure out the least persuasive emotions used in email marketing. (Indeed, we could actually write the worst performing message, though we have found there isn’t much of a market for this.) Want to write close to the worst possible email marketing message you can? Well – to the extent you resemble the average – invoke a sense of urgency in the subject line, and encouragement in the body, and your results should be pretty poor.
Thanks for an interesting analysis. We came to some similar conclusions when developing Phrasee. However, our factor analysis showed that there is a much smaller number of distinct emotional dimensions that drive response rates. For example, there is significant (semantic and statistical) overlap between "urgency" and "fear of missing out".