Display advertising has come a long way since the dark days of pop-ups and poorly served banner ads. Big data has not only arrived but made itself at home, as advertising has become increasingly focused on a granular picture. Tools like Google’s AdSense have already enabled small- and medium-sized enterprises (SMEs) to target display ads to the right audience. Real-time bidding (RTB) has evolved to offer greater efficiencies and control over a company’s media buy. But that doesn’t mean that there aren’t even greater developments in the offing.
At one time display advertising was the uncontested domain of the ad networks. “They would work together with lots of web publishers and aggregate that media,” explains Raoul Witherall, CEO of IO Technologies, producers of learning-based real-time media buying platform IO. These ad networks helped provide some much-needed buying scale and could offer a consistent and less fiddly way to serve ads to the consumer.
However, initially, buying in bulk through a network also made granular targeting more difficult and took place in anything but real time. Witherall continues: “It’d be booked months in advance, in the same way that print publishing or TV is done.”
And perhaps, had our browsing habits not changed, this would have been less of a concern. But the introduction of Web 2.0 and much more sophisticated platforms meant getting the right ad in front of the right demographic wasn’t quite as straightforward. “Suddenly there was a super-abundance of media and from an advertiser’s point of view it becomes much harder,” says Witherall. “How do you find the person who wants to see your ad when there’s 50 times more media available than when it was on a simple website?”
This fuelled a drive for greater efficiency in the buying process and – rather than buying media and hoping your brand was a match for every one of a site’s users – the focus began to shift toward buying impressions, based on the information available on the individual user. “In about 2010, the world changed with the introduction of the first real scaleable ad exchanges,” Witherall comments. “These allow you to buy ad impressions and – here’s the cool bit – you buy them one at a time and you do that by participating in an auction that happens as a website loads.”
This was the birth of real-time bidding. When a user clicks a link and a page begins to load, auctions are initiated. Based upon the demographic and data available, advertisers’ technologies will assess how valuable the opportunity is and bid a dynamic price. The highest bidder will secure the space and the ad is automatically served to the user. Whilst it sounds like a fairly drawn out process, the entire exchange will take place in under 0.5 seconds – the user is left entirely unaware of the frantic bidding war they have sparked when loading that page.
Real-time bidding is unarguably a smart process, allowing advertisers to monitor and tweak the spaces they are buying as ads are served. However, this doesn’t mean the process is entirely without flaws.
“The technology brought to decision making, to evaluating which ad to bid on, is still quite crude,” explains Witherall. Especially in this big data age, analytics can take some deciphering and are subject to interpretation; moreover the freshness of purchasing intention, the risk of brand damage associated with certain spaces, the provenance of data, all muddy the water to an extent that mere humans might not be especially well placed to make the most informed decision. Witherall continues: “That led the way toward what’s going on now, which is people are looking at machine learning and artificial intelligence to make the absolute optimal predicted decisions on which to buy in real time.”
And it seems inevitable this is the next stage in the evolution of online advertising. As we’re dealing with an increasingly labyrinthine data picture, it has become essential to begin relying on our tools to interpret and learn from the market. Currently hundreds of millions of ad opportunities are whipping past every day, each being resolved within a tenth of a second and relying on assessment and observation to know whether it’s the right choice to make. “We’re doing that up to tens of thousands of times every second,” Witherall says.
Which is why bidding driven by machine learning is becoming increasingly important. Figures referenced by Witherall place the monthly display advertising spend in the UK at around £100m, the vast majority of which is planned and steered by human decisions. But the sheer quantity of data to consider means the time where we are required to surrender the reins on our media buy may be drawing ever closer. As Witherall explains, “There’s going to come a point where no human planning buying team at any ad agency or any human controlled technology will be able to keep up with machine learning.”
There are also subtler benefits to accepting a hand-up from our silicon siblings, particularly for SMEs who may be looking to cut any extraneous spend. Rather than requiring enterprises to hire agencies to manage their data-driven advertising, Witherall firmly believes tools driven by machine learning will help to put control back into the company’s hands. “What’s going to happen is technology is going to disintermediate any interim step,” he says. “All the information flows through to you; there’s no intermediary who needs to figure it all out for you.”
Sometimes the world of real-time bidding can appear off-putting to newcomers, particularly as there is little support for the amateur who wants to handle ad decisions in-house. Fortunately, it seems technology that can learn from the data at its disposal is perfectly placed to help interpret the available information and suggest powerful purchasing decisions. “Our assertion is this: in the future, all digital advertising will be powered by learning,” concludes Witherall. “Data is the proper domain of machines.”