How programmatic is evolving as machine learning kicks it up a gear
Mark Torrance, chief technology officer at Rocket Fuel, talks about a role that combines his interests in big data, machine learning, and user interface design.
This article originally appeared in The Intelligence Issue, our April/May issue of Marketing magazine. It was contributed by our issue partner, Rocket Fuel, to let readers know they can access a copy of ‘The Definitive Guide to Creative Optimisation‘ »
At MIT, Mark Torrance rubbed shoulders with Tim Berners-Lee, creator of the World Wide Web. With that early exposure, Torrance founded Stockmaster.com, the first consumer finance website and one of the first few hundred websites.
Later, at web-development company Vinq, he worked on projects that involved artificial intelligence (AI) and human-centred design. One of the biggest projects there was Radar Mail, an email service with AI mixed in. It was a sister project to one that would spin off from SRI and become Siri, Apple’s virtual assistant.
Working on projects like that when that kind of technology was nascent, says Torrance, was a lot of fun, but he left it behind to join adtech company Rocket Fuel where he could marry the interests he held in big data and artificial intelligence with user interface design, to try to improve marketers’ experiences with automation.
Over the past seven years at Rocket Fuel, Torrance has helped it grow from about 20 staff to more than 900. He’s responsible for the machine learning aspects of the platform as well as usability. His focus is on how the software connects with human operators in a way that engages and makes them feel part of the process, while still letting the software do as much as possible.
Less hands-on these days, he spends time speaking with clients and running Rocket Fuel’s Labs team, which develops innovative new products and applications. Marketing caught up with Torrance on a recent visit to Australia to talk about the evolution of programmatic as machine learning kicks it up a gear.
Marketing: You founded a company in the early days of the web in the stock trading industry, and now work in automated advertising trading – given the former has been around longer, can you compare the two and get any hints of where advertising will go?
Mark Torrance: I think you can. We still haven’t reached the levels of real-time decision-making time pressure in advertising that you have in Wall Street. The timing on Wall Street is at the level of microseconds not milliseconds, so it’s 1000x faster. Both systems are ones where a smarter algorithm can win and big data can help you win. It’s, frankly, surprising to me there aren’t more companies like Rocket Fuel truly trying to apply big-data programmatic trading approaches to what to buy online. I think that most of programmatic right now is still humans operating the levers. Programmatic is just a means for the buying to take place. But there are a few companies, including Rocket Fuel, that have at least bits of AI in them that try to make that work.
M: How do you explain how that works to a layperson?
MT: For me, I would say we talk about it as ‘moment scoring’. We try to decide what would be the value of showing an ad to a particular person at a moment in
time, and not treating that person as part of a segment where you treat them the same throughout the day on their different devices. We’re going to treat you differently depending on whether you’re at the café or you’re at home or you’re deep inside a website or you’ve just arrived there and you might be more easily distracted away. If we can understand more details about your behaviour over time, then we can make a better decision about how likely you would be to respond to each ad.
That decision has to be done by computers. It has to be done in real time. There’s no way for a person to do it. All the person can do is pre-decide what the segments should be and then lump millions of people together into segments and treat all those people the same. Even with today’s technology it’s possible to get five times better or 10 times better with machine learning in programmatic than with humans on the same level of campaign.
M: How big of a focus is user interface?
MT: We have within our set of product priorities for the coming year ‘addictively easy to use’ right up there as a priority around user interface. The idea there is
partly consumerisation of IT and partly simplification of it. And just using good design principles to make sure that the stuff that’s most important to people is most prominent and you’re not distracting them with a bunch of unnecessary information. I think that for a long time many adtech companies designed their software with the most sophisticated analytical user in mind. Because those people worked for them and it was easy to go downstairs to ask them, ‘What do you think of this?’
And they’ll say it needs more columns, it needs more rows, it needs more detail.
At some point you have to let that stuff go. Most customers won’t have time to go to that level of detail. We do need versions of the interface that provide that level of detail so that our engineers can check that things are working correctly. Those interfaces don’t need to be as user friendly but they’re not the same interface as the ones you should give customers.
M: It’s an easy sell to the user, but how do you sell it to the purchasing officers and other decision makers?
MT: That’s a challenge, too, yes. You need to have some parts of it that seem sexy and powerful like good dashboards and data visualisations. Those are the things that you’ll show in the demo. That’s the sizzle that sells the steak. Then there needs to be some really solid steak there too that gives the day-to-day operator the level of detail that they want, and the tools and controls that they want to be able to interact with the system.
M: With AI I imagine it must be tempting to just think about everywhere you could change things, not even just within marketing or just within business.
MT: It is. You know, most of the energy that we put into our approach to AI is focused on marketing problems and on understanding the marketing-relevant behaviours of people and building the big data systems about those people. So that is our general purpose AI that is applied to marketing.
We’re going to stay there for the foreseeable future but I can tell you that our founder George John, who
was a close friend of mine at Stanford, absolutely had
the ambition to take Rocket Fuel in a more ambitious direction around the use of AI for flying cars and for smart refrigerators and the internet of things and education. Let’s transform education.
That’s part of the reason he chose a name for the company that is not specific to marketing. One that could potentially move into other areas. What we found in running the business is that we’ve only just scratched the surface of what we can do with marketing. I think that we have a long way to go before we will find the appetite or the need to expand outside of it.
M: And what are you turning your attention to now on the marketing side?
MT: Creative. The content of the ads themselves. And the psychology of how they influence people. I think that a lot of marketing tech and adtech has become caught up in the targeting part. There’s so much to do there that we’ve lost sight of what we are trying to say to people to influence them. There are great opportunities to do a better job there too.
I would like to see a much better understanding of what creative works and personalise that. Which things work for which people or in which context? I don’t think that is the job of an independent company, which tries to use its own data, to decide. I think that getting the right answer is part and parcel of doing a good job on the targeting side too, so I view that as an appropriate part of the domain for Rocket Fuel to branch into.
‘The Definitive Guide to Creative Optimisation’ is a comprehensive guide for building optimal digital creatives based on analysis of hundreds of thousands of digital ads served programatically.
Image copyright: yarruta / 123RF Stock Photo