At OMD EMEA, we're actively pursuing a data strategy that marries the best of human creativity, ingenuity and problem solving, with the scalability and speed of making a decision offered by machine-driven, AI-based predictive analytics.
Many discuss the introduction of AI and machine-learning processes in advertising and communications, simply as a convergence of two industry archetypes; the “techy geek” and “creative tsar”, conjuring images of awkward small talk, language barriers and the type of uncomfortable interaction you’d find at a networking event catering to two different alien species.
This doesn’t capture the reality of what's happening in the industry.
In essence, the process of developing AI in the real world means significant input from real people. To make AI a reality, you need skilled people with diverse expertise, to set up the conditions necessary to support machine-driven intelligence. It’s important to recognise this reality because increasingly, those people will bridge the gap. We have developed several solutions for clients, which leverage large connected data sets, and advanced modelling techniques, to drive predictive personalisation at scale across channels. Our understanding of the interplay between man and machine is based on this experience.
Working at the vanguard of predictive analytics often requires a fundamental restructure of resource within the agency. We require a well-integrated, diverse team of subject matter experts, ranging across; data engineering; data sciences; ad operations & tag management; dynamic creative optimisation; marketing automation; and privacy & compliance. The sophistication of the tasks set out in building the foundations of, and executing, an AI-driven strategy means that these people are best served working in close proximity to one another, utilising agile project management techniques, and maintaining close relationships with suppliers. This team looks and behaves very differently to the traditional agency planning and buying team.
Developing AI models that can predict the next best piece of content to show a user, or the best channel to reach them in, based on historical behaviour, is not straightforward. Predictive modelling requires substantial preparation and maintenance and poses abstract, multi-faceted challenges on a daily basis. Challenges we have observed when faced with the practical realities of developing a “data-driven nirvana”, mostly revolve around the collection, normalisation, matching, and structuring of data sets. This is before machine learning processes even begin.
Throughout the process of developing and maintaining machine learning models, the human involvement is substantial. Insight generation, interpretation, model manipulation, partner relationship management, and solving for interoperability between tech and data partners has driven our demand for highly-skilled data and tech specialists up.
As with a lot of previous past innovation in human civilisation, we are not simply seeking to use technology to solve old problems in a more efficient manner, replacing monotonous and laborious tasks done by people with machinery and automation. We are also adapting these new capabilities to attempt to answer new questions or to address problems that had previously been outside our ability to influence. This requires significant strategic planning and an in-depth understanding of the data and technology ecosystem.
In many cases, it is not development of AI models that creates the bulk of the effort but knowing when and how to apply them.