Squeezing new uses from existing technology sparks creative revolutions. Dancing to recorded music was weird until someone who understood the crowd started to link the right songs together in the right place, and created the nightclub.
(It’s probably best not to dwell on one of the main claimants of that innovation, mind…)
Fast forward to 2020, though, and we’re in a potentially similar place. Music streaming apps are a series of near-identical windows through which lies all music. But where to start, where to go next, and how to care about a certain song or artist?
The best DJs, whether in the clubs or on the radio, have been trusted sources, but where do they sit within the streaming world? And, if they do have a place, will they be humans or artificial intelligence? Enter Radiant, a startup whose app we spotted in February this year.
Radiant is Spotify with added chatter from an AI presenter. If the idea of a virtually bantering radio DJ sends your skin crawling out of the room, wait a second: this is a DJ that claims to know you; the information you want to hear; when you want to hear it; and what music you like (and might like).
Co-founder Pat Quinn is aiming to create a complex brew of a human-like personality, song recommendations and a shared user experience, bootstrapping the first version of Radiant with a three-person team, all of whom have existing day jobs.
Its app works hard to get you onside. Open Radiant, authorise its access to your Spotify account, and Rad – the virtual host – briefly introduces him or herself (you can choose) then plays a song that it knows you love. ‘It’s a Sin’ by Pet Shop Boys in my case.
It followed with songs that I liked less (but didn’t skip – it’s interesting that in a ‘hosted radio’ setting, I felt conditioned not to skip songs – although Radiant does offer this functionality), and then Rad started… chatting.
By adding a human-like DJ to streaming music, Radiant is taking what might appear to be a giant gamble. Radio DJs often fall into the love-’em-or-loathe-’em camp – a dichotomy that the aim-it-at-the-middle tech mindset is averse to. Quinn is happy to take the gamble, in the hope that this will set Radiant apart.
The aim of Radiant is to add guidance and information, and to make the listener feel comfortable with what they’re listening to.
Quinn says that humour and quirkiness is essential. “When we started, we gave Rad a really shiny personality – and it started to annoy us after a few weeks of listening! So we started again and gave it a lot of sass. This is what people want – even more so from an AI DJ. They want C3PO not HAL!”
When Rad jokes, it’s funny – or deliberately weird – enough to make you pay attention, just like a real DJ. Quinn explains that knowing when to crack a gag and when to play it straight is a combination of matching song-level data from Spotify’s API and some real human input.
“We know which songs are bangers, or popular, or upbeat – and we have a semi-algorithmic collection of phrases which can be connected to them. And for some of the most popular songs, we have human-written lines.” Quinn has also been talking to radio DJs to understand better how they speak on air.
This investment in sass came across when Rad described ‘Sledghammer’ to me as “one of Peter Gabriel’s more tolerable songs”. Some of his banter is generic – “Now, here’s something weird and wonderful from the Go-Gos” – but this throwaway chatter feels human. Not fully noticing it may, in a funny way, make Rad more realistic, not less.
These interjections are contextual to the music you’re listening to, and where you are. The difference is that the main contextual data point is you, the listener.
“There’s a lot of temporal information that we use for the DJ – news, travel, weather – tailored for the individual,” says Quinn. Rad can then choose songs based on the weather for instance: sunny songs if the sun comes out, and others for when it’s drizzling.
Rad’s input is tweakable: how many songs play before they jump in with some song facts, how often you hear news, traffic and traffic reports, and what kind of music you want to hear.
All of this contextual info is geographically-sensitive and Radiant learns as you use it. It’s not just about news and weather though.
“At the moment we are pulling data from Last.fm and Discogs, so you hear a bit of biography,” says Quinn. But Radiant has bigger plans – imagine, he says, listening to a song and Rad saying “did you know that in 1974 they played a gig locally?”
Pairing the familiarity of radio with the functions of modern technology will make sense to listeners, Quinn hopes, because they’re already used to the tech.
“Radiant becomes the natural way to listen to music on smart speakers. Listening on an Alexa, for instance, is a very lean-in experience at the moment.”
This means more nuance to the content and creating a personalised schedule. “We want to layer in podcasts as part of the experience. You could hear music and then, for instance, Joe Rogan’s podcast every day at 10am, introduced by Rad as continuity announcer,” he suggests.
Quinn’s not sharing user numbers yet, but the demographic has two main tranches at the moment: “People who do a lot of driving, like taxi or truck drivers, who can be using the app for six hours at a time. And then the 35-50-year-old rock and metal listeners – often they are US-based. They like the lean-back listening experience of college radio – they want that “radio” feel.”
At this early stage, Radiant’s data shows 74% of the users as male, with the vast majority aged over 30. The average user listens for 43 minutes per day, with the top 10% of users listening for over three hours a day.
Radiant’s income could come from a monthly user subscription; taking fees for connecting businesses, users and artists at a local level; and/or licensing Rad’s technology at a B2B level.
One way Radiant hopes to offer more value than a simple playlist is information. Hourly news headlines on the hour could be augmented with favourite artist updates, top tweets, emails, notifications from other apps – anything. Advertising could be part of that.
“I love local radio,” says Quinn, “and one great thing we could do is recommending local gigs.’ Rad could play a band’s song, win listeners over with contextual banter, and then, he explains, “direct listeners to ticketing websites, with Radiant taking an affiliate fee”.
Quinn sees a huge potential audience for Radiant, and their goals fulfil existing needs, he says: “Spotify says they’re specifically interested in accessing people who still want the radio experience.”
Radiant is currently building a version to work with Apple Music, and sees more ways to take advantage of online communities to stretch the idea of what a radio show can be.
“We’ll add group sessions where you can listen together. This could be a show where the music is tailored to all members of a group, or where a station is tailored to one (possibly famous) user,” he says. So, it might be an extension to what artists already do when directing fans to their ‘favourites’ playlists.
Radiant’s technology might even be used by the entities – radio stations – who you might expect it to be disrupting.
“We have no interest in replacing DJs!’ says Quinn, “However, often local radio stations play show repeats at night – and Radiant could be tailored to a station’s listenership and music playlist to create a hosted show.”
If the last 20 years of music-tech have been about building platforms, then maybe the next few years will be about people creatively plugging the parts together and seeing if they can make this technology more fun, more creative, more human.
Or, in the case of Radiant, human-ish. Radiant is possibly an early signal of music tech’s maturation into Lego-brick-like collaboration: pulling data from an API here, grabbing contextual info from a dataset there, and filtering it all through an AI system to build a novel service that jumps through all the licensing hoops.
There are two big questions, the first of which I posed to Quinn: isn’t Radiant more of a Spotify feature rather than a standalone app? He’s open to the idea. “Radiant sits perfectly as a feature. I do like the idea of being a sustainable business… but I’d be very happy working with, or being owned by, a DSP,” he says.
The second big question is one for listeners to answer. How willing are we to swap real human DJs – the background burble in homes and workplaces the world over – for AI DJs? That’s one for the future.
Category: AI radio
Funding so far: Bootstrapped
Contact details: email@example.com
Read Music Ally’s previous startup files interviews here.