AI and the media agency

by Jed

This is long, so apologies in advance, but I think it’s important.

The advertising and media world has once again been set alight by new technology. This time, it’s artificial intelligence (AI). Campaign’s Gideon Spanier wrote last week that “Media agencies should be in fear of automation”, and Blackwood Seven (a widely tipped “AI SaaS media planning and buying service” – more on this later) recently launched a UK office. Both articles talk of the threats that AI poses to existing agencies and large networks (which, for full disclosure, I am part of). I’m not sure there’s as much fire as there is smoke.

First, lets address some of the terminology. Automation and AI aren’t necessarily the same things. Automation can include a basic form of AI, which is decision-making based on a set of rules or logic, but automation is a much broader church. It could be something as simple as automatically printing off a report on a Monday morning.

AI is broadly defined as a computer based system that performs tasks that would usually need human intelligence. This could mean understanding what a picture is, what someone has said, or making a decision based on a set of rules or logic.

The issue with using the two terms interchangeably is that when people see the term ‘AI’ they usually envisage an actual robot, like HAL from 2001, or Sam from Her. So immediately, we’re thrown into the future and everything is terrifying and scary. The reality is, well, quite different.

The development of AI is plotted across three major milestones; narrow, general, super-intelligence.

  • Narrow AI is where an AI engine is given a set of rules or logic and completes a set of tasks limited to those rules.
  • General AI is what most modern science fiction focuses on. This is where an AI engine can apply logic and reason to any subject – like humans.
  • Super-intelligent AI is where the AI engine learns to teach itself. At which point we’re all doomed (well, that’s what Elon thinks).

Narrow AI is where we are right now. We can design an AI to complete simple, repetitive tasks. The breadth of the tasks it can do is determined by the corpus of information it can work from. For this reason, customer service bots are great. We can train them to understand how people ask questions in different ways, and we can feed them FAQs.

However, the AI can only respond based on its narrow design. This is why Deep Blue could beat a chess grandmaster, but couldn’t make a cup of tea after the match.

For media agencies, let’s break it into two; first automation, and then AI.

For anyone who has spent more than a few weeks in a media agency, you’ll know that there are myriad manual tasks that can be automated. Reporting, analytics dashboards, timesheets…(!) The most obvious use of automation within agencies has been programmatic. This (arguably) sits somewhere between automation and a basic narrow AI. Even this basic form of AI has opened up more opportunities for creativity. Now we have strategists, buying specialists, and creative teams working together to tell sophisticated stories using this type of logic. We’ve barely scratched the surface with what’s possible.

Artificial intelligence is the next level up. Blackwood Seven launched in the UK earlier this month to much fanfare. It describes itself as a platform that helps “advertisers to predict the impact of, execute, and measure media more effectively than humans, using a combination of machine learning and artificial intelligence”. The industry press have lapped this up, presumably because the press release and interviews all mention AI a lot.

If I’m not mistaken, Blackwood Seven seems to offer what most media agencies do – except with a SaaS platform and a different pricing model. Predicting the impact of media is just modeling, and it can be done to different levels. For a narrow AI, it would need to be taught (by a human) how to analyse data, so I struggle to see how this would be better than human analysis.

Blackwood Seven claim to improve results by 25-50%, which falls in line with much of programmatic advertising. The data sources that are fed into the platform are the same that every agency has access to, and the same that they ingest into their platforms (accept we call our ‘platforms’ planners). From what I can see, the agency has just found another way to badge what every good media agency already does.

Our market is too complicated, and has too many suppliers to use a single, AI-driven platform to drive media planning and buying. For a start, there are no exchanges that trade TV, radio, non-digital OOH, or print – these are all direct buys.

When working with large volumes of data, someone still needs to identify the key insights from that data. An AI would still need to have been ‘taught’ the rules and logic of planning and strategy by a human. This is complicated stuff. Earlier this week DeepMind (Alphabet’s AI unit) announced that it had managed to solve a critical problem with AI. It had figured out how to give its AI engine a memory, so that it can make a decision, and then use the result from that decision to make another decision. Sophisticated AI media planning and strategy might be a way off yet.

So, if AI isn’t currently running any agencies, how could it in the future? I think there’s a huge opportunity for AI to augment how we currently work.

AI will mean we can run complex models of human behaviour. We’ll be able to test different hypotheses, rather than declaring a single strategic direction as the only direction to take. We’ll be able to use AI to help us identify killer differentiating factors between different cohorts of people. It should also help us spot problems, anticipating issues in advance, and avoiding unexpected results.

I want to work with an AI that can help simplify the data, so that I can do what I do best – working with clients to fix problems, build brands, and sell products and services.

Much of the conversation around automation and use of AI in media and advertising seems to be about optimization. Ensuring that media performance is as sharp as possible. Driving the highest possible return, for the lowest possible cost. Media planning and buying can be driven by AI, and the results can be exceptional (as shown with programmatic buying). But until it’s advanced significantly, AI won’t be able to demonstrate a nuanced understanding of how humans behave.

The best advertising taps into a human insight – it understands culture. Culture is an organic system very few people can understand, let alone teach to an AI. More so, it won’t be able to predict irrational behavior (for more on this, read Nassim Taleb’s Black Swan). So the power of AI-driven media planning and buying will be in the creativity and insight of a human-driven strategist. Human insight, and strategic thinking is what brings and idea to life, and lifts advertising above pure performance. If you sharpen your pencil too much, you end up with nothing but a stub.

Ultimately, I’m excited about integrating AI into our world. There are a whole host of people who like to talk about how automation and AI will replace 47% of human jobs (although if you take the time to read the report, rather than the headline, it’s not that this will happen, but that 47% of jobs contain tasks that could be automated), but I share the view of Barack Obama, who believes that AI will augment our jobs, rather than replace them (his talk with Joi Ito of MIT is eye-opening, Obama is both well briefed and incredibly insightful).

My personal dream? I don’t dream big, I just dream of a timesheet bot…

This post originally appeared on AgencyVoices on LinkedIn.