Algorithms on The Edge: Why Marketers Cannot Afford to ‘Set and Forget’
An unexpected outcome of Covid-19 on the advertising industry is the effect it has had on platform automated optimisations. Facebook and Google have fine-tuned their algorithms based on years of historical consumer data. But now consumer behaviour has fundamentally been disrupted, it has rendered these algorithms almost redundant.
There is no longer a discernible pattern to consumer spending across nearly all verticals, so how is it possible for advertisers to establish and capitalise on trends in these uncertain times?
Suddenly, six months ago, everything changed. A small, localised coronavirus spread worldwide, with the World Health Organization declaring Covid-19 a pandemic in March this year.
Every business and every person has been affected, including the marketing industry, with budgets plummeting -50.7% in the second quarter of 2020, arguably at a time when connecting with your customer was more vital than ever. Yet, there is another unexpected outcome from Covid-19 – the effect it had on platform automated optimisations.
Facebook and Google have fine-tuned their algorithms based on years of historical consumer data. But now that consumer behaviour had fundamentally been disrupted, so too were the algorithms.
Why Automation Alone is a Strategy Bound for Failure
Yet artificial intelligence cannot plan for the unknown, such as a worldwide pandemic or a localised unexpected event that it has not been trained to expect or spot (so-called edge cases), some of which will be momentary and others, such as Covid-19, longer term. In time edge cases become standardised as the machines learn to spot the signals.
Even with the best technology, humans must be a core part of the equation to help spot these anomalies and take over in the short term at least. Leaving everything to automation is a strategy doomed to failure because of these wild and unpredictable instances that are impossible to predict. We lose the human factor.
It’s time to reappraise what a successful approach to algorithms looks like in 2020, if you haven’t already, and consider the key components of installing and operating AI solutions.
There are no shortcuts to investing in AI and ML. It must be a business commitment that is fully backed and resourced. When it comes to choosing or changing an algorithm it is important to select from a shortlist of at least five AI/machine learning solutions and to be highly familiar with at least two or of those. This means that if the unexpected occurs, you have back-up algorithms that you can switch to improve performance.
Don’t make the mistake of believing that you can install the technology, press ‘go’ and leave it alone. You will need to have a dedicated team who know how to use the systems you install and who know how to iterate it and know when to take over manually. That team might be purely in-house, agency or - increasingly - a partnership of the two, particularly when you need a breadth of expertise. For example, there is an increasing set of service partners that specialise in Google Cloud for Marketing.
Our Seamless Search platform integrates paid and organic SEO: we have to train our algorithms and know how they operate and behave inside out. There are also an increasing number of algorithms that we are being encouraged to plug into by Google, Amazon and Facebook. These are algorithms that have been written and trained by someone else, so you can’t necessarily adapt them, you can only adapt how you interact with them.
Algorithms on the Edge
The above advice should be heeded even during business-as-usual, but it is in times of flux or crisis that the blend between machine learning and human nous really comes into its own. It could be the difference between success and failure, particularly if long term brand equity and consumer trust is at risk. Here’s what else to consider when advertising ‘on the edge’.
Edge cases - such as at the start of this pandemic, will almost certainly require human intervention, as the algorithms will likely not have been presented with these situations before. However, edge cases can soon become the new normal, as we are seeing six months into the pandemic, with data accumulated feeding into the algorithms that are being constantly iterated.
So, should a similar pandemic arise, or second and third waves, your AI should have been trained to recognise it. It will become more predictable. But there will be other cases, such as, perhaps a second Trump presidency or tragedy, such as the Beirut port blast, that will throw the algorithms out of kilter. This is particularly important when it comes to messaging.
It is possible, including within Google itself, to automate messaging or Google Search ads. Yet, when there is so much sensitivity, and heightened anxiety around health and economic circumstances, it would be foolhardy to rely on a machine to set the tonality of the creative and the messaging. Automation can still be used to scale that messaging, but it is important to ensure that you have a solid quality assurance process in place to sign off on messaging before it leaves the building. The alternative could not only be off topic but at its extreme severely dent trust in the brand, eroding its equity in the longer term.
An advert posting the wrong opening times for a store might be an annoyance, but messaging showing celebratory fireworks in Beirut in the aftermath of its portside tragedy would be considered poor taste, however innocent the execution.
Amazon has some of the best AI systems in advertising, but was probably the most interrupted by the pandemic. Overnight people stopped searching for books or belts and instead for face masks, sanitiser, toilet roll, trampolines and jigsaw puzzles. This will have immediately changed the way products and brands were presented to customers and many advertisers will have lost out in that process.
Google Search auction algorithms are based on a reasonably consistent or constant set of behaviours by competitors bidding on the same keyword. Suddenly, if most of the advertisers press pause, there will be bargains to be had. Conversely, new keywords will have appeared out of nowhere, with a new (and possibly large) set of competitors coming forth. Who would have thought that PPE or toilet rolls would be among the phrases of 2020?
Above All, Know Your Customer
Your Plan B really should be your Plan A. For now, it is clear that marketers and their agency partners must not rely on the ‘tried and tested’ or set and forget, but instead strategise for the days, weeks, months and years ahead.
Without a discernible pattern to follow, we must rip up the rule-book and establish how to connect and reconnect with consumers in ways that help them. That means everything - from paid activity to organic SEO, customer experience and all offline communications must start with the customer.
Those brands who know and understand their customers will be best placed to ride the peaks and troughs, because if automation fails, or situations change suddenly and dramatically, the manual overrides are easier.
Call it a backup plan, but really it’s marketing 101. After all, marketing isn’t just about machine learning, algorithms or artificial intelligence, it’s about deeply understanding your customers, connecting with them in an authentic way, and building and maintaining lasting relationships with them in ways that benefit both business and individual. Remember that, and the rest is just pipework.