Are You Smart Enough to Test Your Own Subject Lines?


Although the email industry has been resistant to automation, machine learning may be better at testing subject lines than a human marketer could ever be.

Mass personalization has made email a mass marketing tool that can be tailored for one-to-one communication in a way that social media and search simply cannot match. Basically, all effort in email goes to increasing one metric: revenue per open, a number that is very sensitive to other metrics such as open and click rate.

Essentially, all the work of email marketers is aimed at creating incentives and content that drive revenue per open. But are human marketers best suited to improving these metrics?

Compared to retargeting – where machine-made “image-flattening” and cookie-targeting combine to provide marketers with the ability to enable your browsed products to follow you around as you browse the Web – email’s personalization often lurks behind man-made subject lines that are A/B tested to deliver a lower percentage of failure.

According to research from SilverPop, most people don’t open the email they get, despite how much hard work and love the sender puts into it. I won’t bother to recite the statistics, but if you’re reading this you know that if you get an open rate of more than 20 percent you are at the top of your class. Although message timing and user attention are certainly a major factor, many rightfully credit poor subject lines to low open rates.

In a vain attempt to increase open rates, marketers have resorted to all sorts of crazy stunts, like inserting emoticons and Re: in their subject lines, intentional misspellings, personalization Dela, and even trying to guess when to send the message. Most fail at this. Gimmicks are not sustainable and marketers – being people – become addicted to what worked last time, get lazy, and don’t try new things. Or they don’t know what really worked last time.

The simple truth is that optimizing subject lines is not child’s play. It might not even be adult’s play. Increasingly, it has become clear that subject line optimization is best left to our Robot Overlords.

Machine learning is better at subject lines than you are.

Email has been very resistant to automation. Sure, you have RSS feeds. And then you have your triggers and transactional mail. There are, of course, technologies that live within the email to monetize and present real-time content. Subject lines didn’t get this kind of investment until very recently. So I thought I would share the reflections from the founders of some of the services that are developing this technology: Persado and Phrasee. I spoke with them in order to get their insight into how technology is aiding how we leverage subject lines.

According to Persado’s Alex Vratskides, “When the millions of unexamined choices that go into writing a piece of copy by a human are replaced by a data-based approach that machine-generates the most persuasive language for a marketing message or advertisement, the art of persuasion has been automated. Math and science replace the guesswork of copywriting.”

Not one to shy from a confrontation, Phrasee’s Parry Malm has some thoughts about the intersection of data-driven decision-making and subject lines as well. Parry has spent enough time trying to optimize campaigns to know when to hand it over to a machine: “Your subject line is the key driver to getting more opens. Yet, aside from ad hoc tests, few marketers have a structured split testing strategy. And pretty much no one uses machine learning to optimize future tests.”

The tides are changing. Famous for the only real Koan of email, “Mail More,” Dela Quist of Alchemy Worx has recently joined the chorus. When someone who has embraced frequency as brute force methods of getting more opens embraces a concept as nuanced as subject line testing, is the end of the world nigh?

If you are a marketer who is slaving away in the mill, creating beautiful compelling content that is not getting its fair share of opens, clicks, and conversions it begs the question: How could you get more engagement?

Perhaps it is time to admit defeat. Admit that you cannot possibly manually test and deploy enough subject lines – or calls to action – to improve your open rates significantly. Is it time to give into the ghost? The ghost in the machine?