We speak with Parry Malm, CEO of Phrasee, the London-based marketing software company using Artificial Intelligence to generate marketing language.
Phrasee uses advanced language technologies to quantify, generate and optimise email subject lines, and other language types, that out-performs humans and boost email response rates. To date, Phrasee-generated language has beaten humans more than 90 per cent of the time. It was recently named among the top MarTech start-ups in the UK and has raised £1 million in seed capital to boost international growth. Global communications business Next 15 has led the investment, taking a minority stake in Phrasee. The funding marks the first major investment in the UK marketing technology market following the recent EU Referendum, demonstrating the resilience of high growth MarTech businesses.
1. Can you explain how the algorithm works and what makes it so accurate?
As customers, even as people, we’ve naturally evolved to be able to spot what we see as a machine-written email. We usually then either delete it without opening it or filter it to our spam folder. Any marketer will tell you that getting the click-through and eyeball time of a customer can be an act of alchemy.
Accuracy and natural tone are intrinsically linked. Phrasee's natural language generation engine generates human sounding, brand compliant language at scale. Then, an iterative experimentation framework is used to get actual customer response data. Optimisation routines (machine learning) then progressively and continuously improve the linguistic constructs to get our clients increased response rates online.
2. How do you see A.I changing the future of the workplace?
There's been a lot of apocalyptic talk of A.I of lately and most of it is tripe. Machines aren't going to take over the world (The Terminator was ALL LIES!) What machines will do is help humans on specific, quantifiable, and repeatable tasks - freeing us up from mundane tasks and allowing us to get on with more creative endeavours.
You won’t have to worry about The Terminator knocking on your door but if your job involves repetition, then you may be advised to keep a look out.
Here's the thing about A.I. It's not just a magic box where you put in a bunch of data and hey presto, magic. I was at a conference a couple months ago, and someone said, "Should your company do machine learning?" I bit my tongue. You can't just do machine learning, or A.I., or whatever else. As phrases they mean absolutely nothing and at worst are being bandied around as bandwagon jumping exercises.
A.I. will change the workplace in profound ways, but only where there is 1) a need for superhuman understanding and analysis; 2) adequate levels of data; and 3) actual, demonstrable ROI results from doing it. Doing A.I. for the sake of being able to say, "We do A.I." is tautological, and will likely lose a bunch of people a bunch of money.
If you're wondering how your business can benefit from A.I., define your problem first, then seek a solution. Phrasee has done this. Subject lines, for example, are a problem our A.I. tech can optimise and drive ROI for customers.
3. How far off are we from generated language beating humans not just 90 per cent of the time, but all the time?
Actually, our stat, as of yesterday, is now at 95 per cent. But the win-rate isn't the point. What actually matters is the incremental value A.I. language delivers versus humans. The opportunity cost of not using optimised language is huge - and that fact predicates our business model.
Always bear this in mind: algorithms are only as smart as the data inputs they receive. If you feed poor quality data into the most advanced algorithm in the world, the output will be A.I. optimised, but the quality won’t necessarily be any different. A great example of this was Microsoft's Tay chatbot. In terms of technology, it worked wonderfully and it responded dynamically to human input. It was just a real shame that, at least on Twitter, humans are the worst.
4. How is MarTech revolutionising the marketing industry?
One of my first bosses used to say, "90 per cent of marketing doesn't work because it doesn't get done." This is because a lot of a marketer's day to day job is repetitive and boring. What most current MarTech ultimately does is automate the boring stuff. The current market is awash with tools, but most of these tools (i.e. "marketing automation") are effectively commodified.
The future revolution in MarTech is optimising WHAT message you send out, not HOW you send it. MarTech should never act as the emperor’s new clothes – it should be used to compliment your work, not fundamentally alter it or reinvent it. To paraphrase a slightly over-used phrase the message is always the medium, and companies must fundamentally understand this because all the tech in the world will be rendered useless if what you’re saying is bullshit.
5. In a post-Brexit world what does the future look like for the UK MarTech market?
So much of the current fear and hysteria is speculative, and not based upon actual fact or evidence. Let's remember that we are still not post-Brexit, as Article 50 hasn't been triggered yet.
Regardless of what happens, I remain bullish for UK tech. The UK produces a huge amount of highly talented tech people - I mean, that's why Google Brain is based in Kings Cross and not San Francisco!
For global-facing businesses a cost base in a cheap Pound, and revenue streams in strong Dollars, Euros etc., is actually quite powerful.
6. Beyond marketing do you see any other applications for the algorithm?
I wanted to use Phrasee for Tinder, but sadly, Photoshop is the only thing that can help me there...
In all seriousness, our vision with Phrasee is to use our A.I. expertise to supercharge digital marketing. There are numerous non-marketing uses for natural language generation technology... talk to me in 5 years and who knows what we'll be doing!
This is why we, our customers, and our investors are so bullish on Phrasee's future. The applications of our existing tech stack are already widespread, and being used by huge brands across the world. But some of the new stuff we're developing now... I'm not grandiose enough to say we're going to change the world. But, I'm also not humble enough to say we aren't going to change the world.