The dangers of relying on Artificial Intelligence

Three experts from M.i. Media each look at media tech through their own lens and explain what they see…

AI in the media world has progressed dramatically in the last 10 years: algorithms, automated dashboards, off-the-shelf attribution… This progress has been brilliant in so many ways – not least because all that time saved by not needing to do the data crunching frees up the space for thinking about strategy and reducing human error. But is it all positive, or are there also dangers associated with relying on what the black box says?

TV attribution 
– Sophie O’Brien, Business Director 

“So everyone knows that TV is golden. Reach: tick. Effectiveness: tick. Trackability: tick. The measurement aspect has been revolutionary for many advertisers, with TV attribution models offering an effective way to understand the direct impact of TV investment on short-term response metrics. It’s a sophisticated way to inform where efficiencies lie and where investment should be prioritised. 

“But is it as simple as data in, data out? Unfortunately not. What is absolutely critical in this process, is that these tools are being used correctly and that the best quality information is being extracted. If you’re targeting sales, why only track visits? If the business impact appears greater than what the computer says, why not challenge the computer? We live in a world where we can always measure more, so why settle? If we settle, there’s a danger that we make the wrong decisions. Or if we don’t trust the data, we regress back into a world without it. And that’s not much of a revolution is it?”

Digital acquisition 
– George Hobday, Biddable Director

“Since the advent of the internet, it has been easy to become overwhelmed by the volume of data available. With the amount of information we have to deal with, finding insight within it can be hard for human planners alone. To keep up we have become more and more reliant on black boxes – automated, machine learning, algorithms – to find the insights we need. When they work as we’d hope, machines can act faster and across more levers than a human being ever can. 

“But how do we know they are working the way we want them to when we can’t see inside them? Are the insights these machines acting on actually right? The data ingested by the machines is very often what it can read, not what it ought to have. As a result, there can be a disconnect between what data the boxes optimise to and actual business KPIs. 

“We have to be conscious of this and push to ensure our brands give AI better, more advertiser-centric first-party data wherever possible. With the changes to cookies, this is increasingly important for other reasons too. With the advent of data privacy laws and tech updates from the likes of Apple, some technologies are losing access to the data they need to be effective at all. Facebook is a key victim of this post iOS14.5. We are advised by the likes of Google and Facebook to consolidate into single campaigns rather than splitting our targeting so the algorithm has enough data to optimise with. If we do, we will have even fewer levers and less granularity than before on exactly what is working… without the confidence the algorithm actually has the data it needs. 

“So whilst we embrace many of the new approaches available capitalising on AI, we will always try to have a balance to ensure there are manual levers to pull when something isn’t working, and that the AI is being fed quality, first-party data to ensure the best chance of success.”

Media agencies
 – Clive Howse, CEO

“Is the current fixation with AI making us all a bit too accepting? Are we too easily dazzled by the possibilities presented? Have we stopped looking for the limitations of black box solutions? When they allow us to automate so much, is it even in an agency’s interests to question their effectiveness?

“Media agencies use tech to help process the huge amounts of data on a daily basis. But it’s important never to become too reliant on that tech. We’ve been processing campaign data for more than 20 years, so we tend to spot very quickly anything that looks a bit odd. We’re also infuriatingly inquisitive. So if we do see something that surprises us, we immediately want to understand what is really happening. Is it real or just a glitch in the system?

“Working with multiple tech providers means it’s possible to test competitive systems against each other and find specific issues that, without further scrutiny, might lead to some fairly questionable insights into media effectiveness. So the dangers are very real. You will spot these issues if you remain hands-on when it comes to analysing data. Placing your faith in black box technology alone may well be a cheaper option but the impact on advertising performance could well be costing you far more than you realise."