March 18, 2020
Parry Malm, CEO Phrasee, offers a peek at how his company uses AI-generated writing and deep learning to auto-create winning email subject heads, Facebook ads, Google ads, push messages, Twitter posts and similar ad copy.
Malm’s rundown on Phrasee’s method is excerpted from his new book, “The Language Effect: Why AI-Powered Copywriting is a marketer’s (new) best friend.”
Here’s Malm’s pitch:
What Phrasee does is awesome enough to be used in nearly every possible form of advertising — from email marketing, to push messages, to SMS, to social ads, and all points in between.
What Phrasee does is cool. And here’s how it works:
1. You brief Phrasee like you would brief a copywriter. In fact, it’s even easier – because Phrasee doesn’t talk back!
2. You then click the Magic Button (and yes, we have a magic button in the platform, because of course we do!)
3. Phrasee generates thousands upon thousands of human-sounding variants of marketing copy, all in your brand’s voice.
How? We have built a sophisticated Natural Language Generation (NLG) program.
It’s basically a programming language to generate language.
Our team of computational linguists code a language model bespoke to each customer’s brand.
It’s pretty amazing, and I’m not just saying that because I’m incredibly biased.
4. Once we have thousands of versions, we input them into our Deep Learning Engine and the output is a ranked order of predicted performance.
The system doesn’t look at human-defined variables like length or emotion.
Why? Because these are constructs that humans actually respond to.
The human brain is much more complex, and so is our system.
Our neural net looks at thousands of parameters and spots infinitesimal patterns that are hidden to the human eye.
It doesn’t suffer from human problems of cognitive bias – instead, it lets the data do the talking.
5. The end result when you combine our NLG and deep learning systems? You get high-performing copy to use in your campaigns … all at the click of a button.
6. Oh yeah – all of that happens in real-time and takes under 10 seconds. There are no humans involved, so there’s no delay.
7. And that’s not all! Using a method called ‘transfer learning,’ we can take the learnings from one language model – say for example – email subject lines – and extrapolate it across other channels – for example, Facebook ads, Google ads, push messages or Web copy.
What that means is that you use your email marketing as a Petri dish.
And once the scientists make a discovery, spread the results across all of your marketing.
Adds Palm: “Now, of course, there’s a lot more going on behind the scenes. A huge amount more.
“Over the last five years, we’ve built, tweaked, rebuilt, retweaked and optimized the system.
“Hundreds of customer calls led to hundreds of updates.
“Thousands of user sessions led to thousands of tweaks.
“And millions of words and phrases, written by Phrasee’s AI in real-time, led to untold millions in extra revenue for our customers.
“What matters most is this: Our NLG system has been honed over the last five years to become a world-leading (and unique) piece of technology.”
(Editor’s note: Research marketing firm Gartner recognized Phrasee on its ‘cool vendor’ list in 2018.
And CB Insights, an analyst firm, named Phrasee 2017’s Most Innovative AI Company.
Phrasee also snagged the Tech Entrepreneur of the Year title at the UK Business Awards in 2016.)