Machine learning in events and culture
How DeepMind Health got me thinking about ticket pricing. While the price mechanics of events is perhaps not as life and death critical as bed notes and morphine, there’s something about the mix of data, experience, targets and hunches which is similar.
Machine learning in events and culture
Back in September 2016, after wandering around Nesta’s FutureFest for most of the day, checking out VR haptic tech exhibitions, social media post-mortem solutions, and a same-sex headphone dating scenario - all par for the course really - I grabbed a seat to rest my legs at Mustafa Suleyman’s presentation.
Mustafa Suleyman is Co-Founder and Head of Applied AI at UK tech flagship DeepMind, charged with finding real world applications for their astonishing technology. It was one of the most mind-blowing, exciting and game-changing 30 minutes I had, in a year that turned out to be packed full of mind-blowing and exciting 30-minute moments.
Aside from the early video clips of algorithms learning to play Space Invaders from scratch (and getting terrifyingly good at it) and headline grabbing wins at Go (for it is their AlphaGo project that beat top Go player Lee Sedol and subsequently world number one Ke Jie in May this year), it’s DeepMind’s work in healthcare that really got me thinking.
Mustafa launched DeepMind Health in early 2016, a machine learning (the science of getting computers to act and learn without being explicitly programmed) application which builds clinician-led technology in the NHS. In short, DeepMind Health uses collected data to learn from the real outcomes of patient diagnosis and intervention, with the aim of providing continuously optimised clinical support to doctors and nurses. The more it’s used the smarter it gets. What’s more, all of a patient’s care team are brought into the loop, with centralised charts and notifications at predefined trigger points. No more bed notes. No more double doses.
DeepMind’s smarty-pants solution doesn’t replace the hands on care of nurses of course, a friendly face, a caring voice - as my specialist nurse wife with countless hours on the wards will tell you - but in an understaffed, underfunded industry where decisions go from casual to critical in a heartbeat, the shift from dumb data to smart data, the benefits of the swift and efficient sharing of information is a no brainer.
And all of this got me thinking about ticket pricing. No, really. While the price mechanics of events is perhaps not as life and death critical as bed notes and morphine, there’s something about the mix of data, experience, targets and hunches which is similar.
My approach to pricing has always been pretty straightforward, fairly standard I think, as I’m neither a true spreadsheet jockey nor a 100 per cent gut-feeling guy. I’m somewhere in the middle. In the past, my event pricing went like this: what’s the deal, what’s our capacity, what’s our sales expectation, what did we do last time, what’s the venue down the road charging, how much can we push it. Kind of a mix of data and hunches that - from my recent conversations with CEOs and box office managers all around the country while developing my company Neon’s venue-driven reselling - seems to be what most people do, even in the biggest, most sophisticated venues.
But imagine this instead. As you’re typing the name of your next headline artist or touring show into your ticketing platform, the system recognises who you’re talking about and auto-populates the rest for you. It has a bunch of data from your previous shows, so it suggests the best room to use, maybe even the optimal day of the week and, looking at your previous prices and sales volume makes a price suggestion. You’re done in five minutes and your setup and pricing is the very best it could be. This is how machine learning will save you hours and add 15 to 20 per cent to your gross revenues in the coming years, and it’s the thought that kept me up for a few nights after FutureFest and made me build a whole new ticketing system (despite people telling me I was crazy!).
My insomnia didn’t stop there though, and my gleeful prediction to ticketey-type people I meet now, is that in five years’ time the idea of inputting a price at all, picking a performance space or date and time anywhere in the on-sale process will disappear altogether. Instead, you’ll enter some key parameters - from your direct costs and the shape of the deal, to your priorities (bums on seats or revenue) and Neon’s algorithms will do the rest for you, on the fly, hour by hour. The idea of “face value” will disappear, much like it has already with flights and hotels, and what we’ll have instead is optimised pricing based on the makeup of each event, how it’s performing, and the individual habits and behaviours of every customer.
Perhaps we’ll always need the human touch of a box office manager or artistic director, culture is so nuanced after all, but I’m not convinced we should get too comfortable even with that. Once “the machine” knows your customers, their clicks, their purchases and their history, when they dropped off and when they came back, then recommending what time or day of the week will work best for your sales, what space to use, or even who to book or what to commission - based on your customers past behaviour - is only a few lines of code away.
Netflix, Spotify and Amazon are already there. From Spotify’s bespoke playlists and Amazon’s tracking of your page turns and drop off points with every e-book you read, machine learning is already making significant inroads into the culture industries.
Netflix, for example, uses machine learning to make its recommendations super effective, adding you to one of over 2,000 “taste communities” to work out what you might like and where they can nudge you next. Netflix tracks what you watch, for how long, what time of day, whether you binge or slow burn, and matches you up with other people just like you. So far so big data, but the clever bit is how its machine learning algorithms dynamically weight all of this, working out whether it’s more important that you watched two crime dramas yesterday, or that you’ve been watching Glee once a day for the last three months. In short, Netflix probably knows you better than you do.
Imprompdo, a UK-based research project developed at Cardiff University, takes this same behavioural learning and applies it to the problem of to-do lists, and importantly, when they should bug you to complete a task. Rather than the normal scenario of making a task and then setting a day or time to remind you, Imprompdo looks at your location data, time of day and other indicators to find the most effective time to send you task reminders.
So what? Well, imagine membership renewals, new programme announcements or late ticket offers reaching your customers not when you send or schedule them, but at the moment they are most likely to respond to them, in the format they are most receptive to. Each and every one tailored to each and every customer. Not the “best overall” time you can currently get on Mailchimp or similar, but individual, personalised, optimised marketing conversations. How much do you think your conversion rates would go up by? One per cent? Five per cent? 50 per cent? This, by the way, is the curious paradox of big data and machine learning, it becomes more personal, not less.
But it’s not just Neon and the likes of Netflix and Amazon who are harnessing the opportunities presented as machine learning finally comes of age. We learnt this working with a group of incredible people; cultural businesses, entrepreneurs and idea makers, firstly through the Digital R&D Fund for the Arts with our dynamic pricing research, and more recently as part of the Digital Arts & Culture Accelerator, where we first began researching and modelling how we could build a ticketing and retail system fit for 21st century cultural businesses, including how to harness secondary selling. CultureCounts’ benchmarking platform, Abandon Normal Devices’ VR aggregator and even LiveStyled’s customer experience mobile apps all represents various pathways in this revolution.
Machine learning may be late arriving in the live events space, but now it’s coming on strong, through business applications at first, but its impact on creative content and the way the sector makes decisions will inevitably follow. Much like the rise of social media, there will be profound, democratising changes, and shifts in power and control which will be simultaneously exhilarating and unnerving, but the fact that we are on the tipping point to a new way of making and serving cultural products and experiences is undeniable.