We need to get real about ‘art vs science’ in digital media

You can’t order an algorithm to manage your campaign, any more than you can order a toddler to bake a cake, writes The Kite Factory’s search director

Machines are doing almost everything for us nowadays but there are some areas in marketing where humans undeniably have ‘the edge’: this tends to be anything requiring empathy.  A machine can tell us which ad is performing better, but it may take a human brain to confirm why individuals aren’t interacting with a particular creative.

From a consumer perspective, it’s an algorithm that will decide what you’re shown next on YouTube, or which Facebook ad you get shown, but a human who has created the content itself.

There’s a pretty simple way of identifying opportunities where machine learning and algorithms could (or should) play a part: generally speaking, if it’s art, it’s human; if it’s science, it can be machine led. Although there are notable projects currently taking place within the industry to build AI that can write film scripts etc, we’re not quite there yet.

Traditional marketing channels had very few ‘science’ opportunities – the main limitation with something like a TV campaign is budget. A familiar refrain might have been: “If we had more budget, we could run more spots”.

In terms of digital campaigns, you’re now more likely to hear, “We’re not spending the full budget as our ad delivery is limited… but we’re working on improving the click-through-rate to improve quality score and impression share”. Slightly different ballgame, right?

Due to the heavy science skew, the high operational input required, and the sheer volume of data involved, digital relies heavily on technology, and so is ripe for automation. Anything repetitive, or anything that can follow an “If x, then y” format can be programmed to self-activate as and when required.

Baking with numbers (or not)

However, with great power comes great responsibility.

I have seen overspends of up to £250k over a matter of weeks because a human misspelled a campaign name, so the automation skimmed over that campaign when looking for certain naming conventions.

I have seen automated bidding run a campaign into the ground; as the remarketing audience became smaller, the bidding platform was still trying to gain volume, so set cost-per-clicks to over £70 in an attempt to secure traffic.

Imagine having a toddler helping you bake a cake. If you were to tell the toddler how to make a cake, would they create a Mary Berry-style wonder independently? No. They would get flour everywhere and probably try to eat a raw egg.

So, instead, we give simple, smaller instructions: “put one spoon of flour in”, “add the water”, “stir it all up” are directions a toddler can understand, and the process will still result in a cake, just with a little more operational rigour.

This is how we should think of automations, algorithms and machine-learning when it comes to digital marketing. You can’t order an algorithm to “manage my campaign” any more than you can order a toddler to “bake a cake”. However, you can tell the algorithm how to change your bids, how to target audiences, how to display content, and any other direct instructions that will improve your performance.

Automation doesn’t mean you’ve solved all of your workload challenges, but it does mean that we humans can free up more time for art such as strategy, creative, building relationships and identifying opportunities.

And yes, we still need to check in to make sure our automation is running as planned (you wouldn’t leave a toddler to bake a cake alone, would you?) but this kind of technology negates the need for us to be tied to our desks every second of the day, manually tweaking bids by a matter of pennies (which was a key aspect of my first digital role, believe it or not).

Proof is in the performance pudding

You may wonder why we’re in such a rush – why can’t we just make those tweaks less often? Well, marketers aren’t the only ones furnished with algorithms and programming. Platforms such as Facebook and Google are built on proprietary algorithms which, for example, display relevant content to each user, or vary the ads you may be served.

These algorithms thrive on science.

Engagement rates, landing page quality, relevance to the user – all scientific, quantifiable elements. The Facebook system doesn’t have the capability to identify stunning creative – the proof is all in the performance pudding. If a campaign doesn’t achieve strong results, so begins a slippery slope of trying to get back in the good books of the algorithm – I don’t know if you’ve ever tried negotiating with a faceless piece of code but you don’t get much back.

There’s no negotiation to be had, no “oh but please”, no “I’m sorry, our account manager was on holiday”; there is no chance for appeal. The algorithm monitors how you campaign is performing and will respond accordingly in the best interests of the consumers it serves; either decreasing your costs and increasing visibility to congratulate you for being a best practice whizz, or exiling your ads to eternal marketing damnation, rarely to see the light of day again, through decreased delivery and increased costs.

For modern day marketers it’s key to identify which elements of your activity are ‘art’, which elements are ‘science’ and then resource these areas accordingly, either with a bubbly artistic human, or someone with technical capabilities who can implement automations to keep up with the ever-demanding marketing algorithms.

AI is here to stay, and it’s component parts (algorithms, machine learning) are only expected to become more embedded within products, processes, and our everyday lives beyond.

I’m sorry art fans; when it comes to digital, science has become at least 50% of the battle.