If you've recently received an email from Ebay, Domino's or Gumtree, there's a good chance that the message was crafted by a robot.
All of those companies are clients of Phrasee, a London-based startup that has developed an AI-powered system that creates human-sounding, brand-compliant marketing material for email subject lines, Facebook ads and push notifications.
Phrasee's team of data scientists, computational linguists and developers create custom language models tailored to the specific guidelines and tone of voice of each of these companies. These models allow the natural language system to generate brand-compliant, human-sounding marketing copy at scale.
A deep learning engine then predicts the efficacy of different versions of the same message and picks the most likely to perform option for the individual campaign.
The results can range from better open rates to more click-throughs and ultimately more revenue. The system gave Domino's a 57 percent click uplift, Gumtree a 35 percent growth in open rates, and Virgin Holidays millions of pounds in incremental revenue.
"Say that you're Superdry, who we work with, and you've got a sale on shoes of 50 percent off, the number of ways that you can spin that into a subject line is actually in the trillions," Parry Malm, CEO and cofounder of Phrasee tells Techworld.
"There are literally trillions of ways to say the same thing. Now, a human could never come up with all those various options, and even if they could, humans at scale are not able to predict what's good and what's bad. They base all that on their own gut instincts and their own cognitive biases.
"What our software does is solves those two problems. Number one, it can write stuff that's indistinguishable from that which humans write, and secondly, it can predict at scale what's going to work."
Developing the concept
Parry worked in digital marketing for a decade prior to founding Phrasee and traces the origins of the concept for the business back to when he was sending out millions of emails for a FTSE 250 media company.
"We'd always test out different subject lines and one of them would win, and we'd always scratch our heads and say why did that win," he says.
"We tried to make Excel spreadsheets and build a model around it, but never really could. Then I left that job and I joined an email marketing technology company and ran the commercial team there, and I met literally thousands of people who were email marketing practitioners, and the question that they always asked me was 'what should I use in my email subject line?'"
He began to form an answer while sat in the Brewdog bar in Camden Town. Parry and his old friend Dr Neil Yager were knocking back a few pints en route to a metal gig.
Dr Yager has a PhD in computer science from the University of New South Wales in Australia, has published more than a dozen academic papers, a book on data mining, and holds a bunch of patents in AI. Parry was a digital marketing veteran who knew that no one else had found an effective form of subject line optimisation.
The two Canadians realised that the combination of their different expertise could produce an automated solution.
Read next: AI startups to watch
To help grow the idea into a successful business, Parry recruited Victoria Peppiatt, the founder of a creative agency called The Pink Group who he previously employed for branding and web design.
In February 2015, the trio launched Phrasee, the world's first SaaS platform that creates and optimises email subject lines. They then expanded into machine-generated push notifications and Facebook and Instagram ads.
"There aren't very many companies in the world who are actively commercialising natural language generation, and that's because it's a very hard problem," says Malm.
"Some of the technologies which we use weren't even in existence two years ago. We're three and a half years old and a lot of our deep learning stack is using methods and algorithms that just hadn't been invented when we first started Phrasee, so we're constantly updating stuff, integrating new methods and algorithms."
Phrasee finished its third fiscal year with a profitable quarter and 280 percent revenue growth, and recently raised £4 million in a Series A funding round to help them diversify their product and quickly expand overseas.
Marketers may fear that Phrasee is coming for their jobs, but Parry doesn't envision the product soon extending to the more complex problem of the body text in emails and other longer forms of marketing copy as it's difficult to isolate the independent variables.
"The great thing about a subject line or a Facebook ad is you can split test them and you can actually benchmark the success of a given piece because you can isolate the copy variable," he says.
"What you can't do with longer form copy is isolate independent variables. You have much more going on. It's not just language. You have layout, you have graphics, you have colours and also there's many, many more words involved. So actually proving the case is much more challenging."
He argues that Phrasee has in fact created new jobs, by offering linguists a new careers path as language technicians.
In short, his objective is to support marketers rather than replace them.
"My basic vision is we want to change how brands communicate with their customers online," says Parry. "It's been left to inefficient processes and human biases and gut instinct for far too long and this is going to change."