The capabilities and advances currently being explored by AI have seen an incredible rise in recent months as businesses begin to implement the latest innovations into their technology stacks. But the stigmas around AI still exist; its security, its accuracy, and of course its potential to replace marketing teams with robots. While the latter might be a bit of an exaggeration, there is arguably a lot of misinformation about what AI can achieve within a brand’s marketing strategy right now. We spoke to Parry Malm, CEO of Phrasee, to find out more about the technical abilities AI can supply to save time and increase revenue.
People like to debate whether AI or humans are better at doing tasks than the other. But the real use cases for AI at businesses today show more of a symbiotic relationship. That being said, does this mean the whole humans versus AI debate is reductive?
PM: In recent mainstream media there has been a lot of apocalyptic talk about AI and it has guys like Elon Musk and Stephen Hawking weighing in with their philosophical views on its various applications. From a digital marketing standpoint, we need to accept the fact that what we can benefit from are specific applications of AI. Whilst its consequences are far-reaching, they’re not as life altering as AI driven warfare, for example, or AI-assisted cancer treatment.
The only reason a business should invest in AI is if it’s doing one of two things; its either making you more money, or it’s saving you money. We quantify the capabilities of our AI by testing every subject line that Phrasee generates against one from a brand-side human marketer. This allows us to benchmark our results, and show people that the AI is actually doing something powerful and profitable that humans can’t do. There would be no point using AI if it couldn’t do a task better than humans. I think people have to focus on these sort of specific point solutions. Leave the philosophers to their debates, and just focus on tools which you can use to make your job easier and more effective.
How does Phrasee draw from humans to create a human-sounding voice for the email subject lines? And what kinds of skillsets are embodied by the Phrasee team to enhance this understanding?
PM: AI is actually an umbrella term that encompasses numerous different technologies, and at Phrasee we use AI in three ways. The first of these is ‘natural language generation’, and this is a system that can generate human language at scale, which is indistinguishable from real human language. We then have a ‘natural language processing’ layer that quantifies various linguistic components, such as sentiment and syntax. This is really important to ensure context sensitivity of the language itself. The third really cool thing we have is an ‘end to end deep learning system’, which is self-learning, and able to predict the performance outcome of language at a really high accuracy before it gets sent out.
The way I view it is like this: Facebook spent a billion dollars on developing an AI to recognise a cat. And that’s really useful, but a two-year-old could do that. What Phrasee can do is take a set of 20 or 30 subject lines, and tell you which one is going to generate the most opens and clicks before it gets sent out. This is something humans can’t do. That’s what makes this application of AI so valuable, because it’s actually creating a new sort of skill set that was impossible before Phrasee existed.
So as far as who actually does all this, we first have a team of computational linguists; these are people who are trained in formal linguistics, and they create these generative algorithms on a client to client basis. Then we have a data science team, which focusses on building advanced deep learning methods, and improving our modelling techniques.
How has this processed changed the way in which you view language? Emojis for example, are becoming more and more common in email subject lines, and Phrasee uses them too. What impact have you found they have?
PM: Emojis themselves are represented in roughly 5 per cent of global subject lines right now, and that rate is increasing. What we found when we first started seeing them is that there was an immediate bump and everybody jumped on board. Of course, when everyone is using the same methods, you eventually get to a tipping point of effectiveness. We’ve done a lot of research on emojis and some pretty intense tests. We have an eye tracking studio in our office, and we’ve actually tracked how our eyes perceive emojis and other linguistic components commonly found in subject lines. What we found is that emojis have an amplification effect on subject lines. So if the core subject line itself sucks in the first place, you can put an emoji on it, and it’s then going to be a subject line that sucks – with an emoji on it. But if the subject line is good in the first place, it then amplifies that. Emojis themselves aren’t necessarily causable towards subject lines performance, but they amplify whatever performance the quality of the subject line was going to move towards in the first place.
Do you think that the proliferation of AI, and the huge range in the functions and capabilities available, has given the area a bit of an image problem? What is your advice for marketers looking to overcome these and introduce AI capabilities right now?
PM: It does have a bad image, and I think a lot of the AI we are seeing at the moment is quite gimmicky. The problem is that people think AI can do everything. They think it’s a lot more advanced than it actually is.
Rather than asking how to introduce AI to their business, what people should do instead is look for specific point solutions and specific business challenges they already have. Spend 95 per cent of your time thinking of the problem, and 5 per cent of your time thinking of the solution. If you define your problem really well, you can then determine whether some form of AI is an appropriate approach, or if there is a different methodology that’s going to be equally or more effective.
Until you really know which problem you’re trying to solve, then you’re just buying kit for the sake of buying kit, and that’s kind of pointless. This is where marketers meet resistance from their CMOs, because they share this viewpoint- they want to know what the investment will be worth. But the fact is you’d build a business case for an AI system the same way you’d build a business case for anything else. The fact that something uses AI doesn’t necessarily make it better or worse than different solutions. Build the business case based upon a series of KPI-driven benchmarks and if the solutions hits it, the fact that it uses AI doesn’t really matter, it’s just really cool.