Work in the new age of automation – an agency researcher’s view

How can media agencies take advantage of artificial intelligence to deliver planning insight? Geoff Copps explores.

A few years ago I recall being commissioned to analyse the most popular content on the Gourmet Dining section of a well-known website. At first glance, the web analytics software I was using had clearly identified the most trafficked items. The top entry was somewhat surprising: a brief review of an Italian-themed restaurant in Battersea.

On most sensible objective measures the item performed best: views, unique user browsers, bounce rate, etc. But why?

With a second glance at the data, the mystery was easily solved: the restaurant in the review was called Bunga Bunga, and the webpage’s ‘popularity’ had grown in the wake of allegations concerning the Italian Prime Minister. With this knowledge I revised my conclusions appropriately.

Nowadays barely a day goes by without a headline on the twin topics of AI and automation. Everyone’s at it, from the mainstream press (‘Rise of the robots’) to our own trade publications (‘In 10 years’ time your agency will be an algorithm’).

Amid the speculation and grand theorising, I still find this simple example of restaurant data interrogation instructive. For here I believe we can see, in microcosm, the complimentary skills of machine and human.

‘The first glance’ is made possible by the machine – its vast processing and discovery power, trawling through millions of data points in under a second to yield up an exhaustive list.

The ‘second glance’, meanwhile, remains best handled by a human: able to interpret the item in its full cultural context, to bring it to a subjective, intuitive understanding of human motivations and behaviour…and (most human trick of all, this) to ‘get the joke’.

[advert position=”left”]

It isn’t that a computer can’t be trained to do part two nor a human to do part one; it’s just that neither part can yet be done as sensitively, comprehensively or instantaneously by the other. (As the ‘second glance’ example shows, the claim of greater speed and efficiency isn’t necessarily limited to machines…) The real point is that genuine, timely insight comes from a collaboration of the two.

Over the last 12 months, in the Mediabrands Marketing Sciences team we’ve intensified our experimentation with automating processes. We’ve been working with ever greater volumes of client data, open-source feeds, APIs, third-party and proprietary software. We’ve been writing code to unlock insight from datasets in fast, flexible and efficient ways.

Our efforts have been encouraged by support from the wider business: earlier in the year IPG Mediabrands held its first Hackathon, a three-day event that brought together our community of data experts and developers from across the globe.

Machine intelligence remains particularly useful to us for extracting value from large data sets, such as customer databases and transactional records. Academics have long cited ‘discovery’ as the area within which such processes are set to have the greatest impact on white-collar workers, and it is easy to see why: machines work at scale, with precision, without human error, and are adept at pattern spotting (though perhaps slightly less so at anomaly identification, as my restaurant example shows).

It is these considerable advantages that we are leveraging. Over the last year we have built bespoke ranking tools, customer segmentations, attribution models, and automated processes to aid search and social discovery.

However, all this automation work needs to be seen in its proper current context. For when working with data to generate insight, the most fruitful and productive results continue to come through a combination of human and machine application. It’s all about getting the right balance to improve outcomes for our clients. In support of this thesis, I would like to give one final example.

At Mediabrands, we are proud of our market-leading proprietary planning tool. It sits on every planner’s desktop and forms a core part of our agencies’ planning output. The jewel in the crown of this tool is its optimisation functionality, which allows planners to calculate the best way of spending their clients’ budget across media and over the duration of a campaign.

The tool calculates optimal outcomes using a powerful genetic optimisation method. This method is based on evolutionary algorithms, proceeding by a form of logic akin to ‘natural selection’: calculating a range of solutions to your problem, selecting the best then recalculating on that basis, then doing the same over and over again.

This is machine learning in action, the power of advanced automation at the very heart of the planning process. In the UK we have recently built upon this solution to make the tool truly cross-media, factoring in clients’ owned assets as well as paid spend. We continue to integrate new data sources such as Facebook reach and frequency inputs.

So great, you’re thinking: a cross-platform media plan created entirely by a machine with no human involvement. Right? Not quite.

The tool cannot do all the work on its own. It proceeds by powerful logic of efficiency and effectiveness, but it needs the planning team to set the strategy – to feed it with an audience, objectives, insight, plus any number of relevant and ever-shifting client, market and campaign parameters. And of course it cannot handle the big ideas or, for example, make intuitive leaps.

A category-defining sponsorship of an event? A content-driven approach that sets the agenda and creates fame? A counter-intuitive focus on niche audiences whose cultural influence ensures product sales are stimulated in the most sustainable and brand-enhancing way? What machine can yet advise on this? Again the concept of fruitful human-machine collaboration comes to the fore.

When speaking of media research and planning insight, then, I always prefer to talk of automation as a complementary endeavour. Some of my colleagues elsewhere in the business – the ‘true’ data scientists who look ahead to the ‘Technological Singularity’, a future in which computers are capable of redesigning themselves and become effectively self-governing – may disagree.

In the meantime, in Mediabrands Marketing Sciences we continue to explore the dual benefits of human and machine intelligence. We typically use automated processes more and more to cover the kind of heavy-duty, time-consuming data tasks humans often fail at – and as a result, we are free to concentrate on the most creative tasks.

So it is with an understanding of the growing potential for computing intelligence to transform our working practices, and with sensitivity to the distinction between human cognition and machine learning, that we look forward to a bright, innovative and insight-full future.

Geoff Copps is head of research at IPG Mediabrands UK.