skip to navigation skip to content

Brain food


How networks are turning to mathematicians to maximise ad dollar returns

Media planners be warned: the next great battle for eyeballs will be conducted via algorithm.


When your customers are paying tens of thousands of pounds for a 30-second advertising slot, careful planning of where – and when – a spot appears is crucial. After all, nobody wants their ad to appear right before or after the competition, do they?

Dr Houyuan Jiang, Reader in Management Science at Cambridge Judge Business School, says he first came across the thorny issue of ad planning a few years back when a broadcaster contacted universities asking for help with maximising returns on advertising slots. Dr Jiang and his colleagues have since applied their mathematical minds to the problem and come up with an algorithm (based on no fewer than three fiendishly complex formulas). Their paper, “New formulations for the conflict resolution problem in the scheduling of television commercials”, aims to make sure competing ads never appear during the same time slot.

“The problem is all about how television stations allocate advertisements in the ad breaks,” explains Dr Jiang. “Suppose they receive a lot of requests for ads – for cars, movies and soft drinks for example. If one ad appears against a different brand it defeats its purpose. They have to find a way to avoid it, but there is also a huge number of possible time slots. The potential ways to position the ads could add up to millions. You can’t do this with a pen and pencil.”

At Britain’s Channel 4 television station, Head of Airtime Management and Ad Operations Tanya O’Sullivan agrees. “At Channel 4 we have our own algorithm, years in the making, which overlays maximising return,” she says. “And it’s a jigsaw puzzle. Channel 4 Sales plans ads across 28 channels – there are five and a half million spots per year – and the revenue from Channel 4 linear [that’s television as opposed to video on demand] is £1.1 billion a year. So it’s a powerful decision-making tool. We are always looking to improve our algorithm, so something more efficient will always be of interest.”

Dr Jiang, who has previously worked on algorithms to schedule trains or, currently, to reconfigure emergency and outpatient departments, sensed a challenge. “Looking at the existing literature,” he says, “it seemed there was no research advanced enough to increase the efficiency of the process – only heuristic methods which were too approximative.”

“I have a mathematical background,” he continues, “I like applying mathematical models to solve real world problems. This is business analytics – the process of exploring small or big data to generate management insight and improve results.”

In this case his number-crunching produced three formulas which have proved more accurate than heuristic methods in 76 per cent of the cases tested by Dr Jiang and his colleagues. “Our models generate optimal solutions which minimise the conflict between commercials,” he says. “However, we appreciate that broadcasters would need to add in factors such as the revenue attached to a particular ad and desired exposure to a particular audience. Our formulas don’t do that.”

In fact, O’Sullivan explains that not only does the algorithm need to maximise revenue for the television station, it also needs to manage two distinct types of ad conflict: “One is the product category, e.g. cars, which must not appear together, which schedulers call ‘clashing’,” she explains. “But it must also take into account restrictions from the UK Code of Broadcast Advertising (BCAP), preventing children from seeing ads relating to alcohol or gambling, for example. These are the regulatory restrictions.

“And we physically check, because no algorithm can account for juxtaposition. For example, during a documentary about shark attacks, one broadcaster made the mistake of showing an ad for the shark-horror film Open Water!” It is usually humans who make mistakes, she says, by not logging an ad correctly, or missing an advisory restriction. “So we computerise the process as much as possible.”

Furthermore, the algorithm has to manage audience – it has to aim particular ads at particular people. “In linear television there are 22 audiences for broadcasters, based on age, gender and class,” O’Sullivan says. “We use a panel of 5,100 households who represent the UK population.” This is the ‘one to many’ principle of advertising.

But in video on demand, where TV programmes are streamed to your phone or computer, the viewer is an audience of one – known as ‘one to one’ advertising, where the broadcaster knows more about you, so ads can be a lot more targeted. This is perhaps where the challenge lies for the ad planner: being able to dynamically insert tailored advertising during live streaming. Sky Go, a service of Britain’s satellite broadcaster Sky, for example, has 60 potential channels.

Dr Jiang agrees that this would be a “very good project.” He muses: “You would need an intelligent algorithm to target people. I can’t promise anything, but if Channel 4 or Sky wanted to collaborate with us on this we can give it a go!”