Why you shouldn’t leave Conversion Rate Optimisation to AI alone

By: Robin Nichols


AI has been a blessing to marketers, freeing up more time for them to analyse and strategise.

But that’s not to say you should leave your CRO fully in the hands of AI.

This article will explain why online retail conversation rate optimisation still need the human touch.

Online retailers live and die by conversions. Anything that helps marketers and CX experts tweak a CTA, newsletter content, purchase funnel or product page design to increase micro or macro-conversions is seen as a blessing.

To that end, we’re seeing MarTech tilt more and more towards platforms powered by algorithms and artificial intelligence, and away from more ‘manual’ solutions, as a recent infographic published in Adweek illustrates:

Marketing infographic

Adweek: 2017

Activities associated with conversion rate optimisation are no exception, from hype over ‘1:1 personalisation’ to predictive segmentation, smart analytics and automated testing and experimentation. It seems that the market can’t get enough of AI, seeing it as a panacea for all of their optimisation, personalisation and conversion struggles.

AI use graph

Marketing Charts: 2018

There are indeed many reasons to embrace AI-powered technology: increased scalability, freeing up marketers from drudge work so they can focus on strategy, faster and more accurate data analysis, and so on. However, there are also significant caveats.

I’m not arguing that marketers or CX specialists should abandon AI altogether, nor that AI isn’t better in some use cases. I’m arguing that the industry as a whole shouldn’t throw themselves headlong and unquestioningly at all things algorithmic. In other words, that conversion rate optimisation (CRO) experts should ‘stay human.’ Here’s why:

1. Public ambivalence regarding algorithms’ overall impact on society;

2. Specifically, AI-powered personalisation has significant limits, and

3. Machines just don’t always get it (enter psychology)

1. Ambivalent feelings about AI

As far as public opinion goes, the jury is still out regarding the increasing prominence of AI in every aspect of our lives, whether it be in marketing personalisation or in government, healthcare, media or anywhere else.

In 2017, Pew Research Center asked 1,302 technology experts, scholars, corporate practitioners and government leaders: Will the net overall effect of algorithms be positive for individuals and society or negative for individuals and society?

The responses were varied: 38% predicted that the positive impacts of algorithms will outweigh negatives, while 37% said negatives will outweigh positives and 25% said the overall impact of algorithms will be about 50-50. Though some benefits were clearly identified, a wider range of significant concerns and challenges were also mentioned:

Themes of the algorithm era

Pew Research Center: 2017

Anxiety that ‘humanity and human judgement’ will be superseded by machine reasoning seems particularly salient. Everyone’s experienced the awkwardness of interacting with an algorithm that just doesn’t get it, whether it’s a frustrating conversation with a chatbot, ridiculous personalised product recommendations or irrelevant retargeting ads that follow you around the internet.

Isn’t there a risk we’ll ‘algorithmify’ our digital interactions to the point of putting us off them? Not to mention the more sinister concerns, like increased unemployment, societal divisiveness and bias.

Steven Finlay, author of, Artificial Intelligence and Machine Learning for Business, explained why human expertise is still central to an AI marketing platform. He outlined how a marketing AI application for a drinks company might typically be used to build a predictive model using personal data curated from social networks to determine how likely someone is to buy a particular brand of whisky.

However, the algorithm might very well identify that children and ex alcoholics as likely whisky buyers - and if left to its own devices, would have triggered ads to these groups that would be ethically unacceptable.

Steven concluded that, “These two rules [regarding automated ads selling alcohol to children or ex alcoholics] are a great example of why human expertise is required to support automated machine learning based systems, especially where systemsare being used to make risky or controversial decisions about people.(Artificial Intelligence and Machine Learning for Business, p.13 Kindle edition).

This is only one example among many that algorithms don’t always know best. If left to develop unchecked, the societal consequences might not all be positive.

2. Caveats to AI-Powered Personalisation

One of the fastest growing areas of conversion rate optimisation - and the one in which AI is beginning to cover the most ground - is that of personalisation. Product recommendations, personalised news feeds, website personalisation or curated content...all of it increasingly deployed by algorithms, in the service of increased conversions. But with personalisation in particular, there are significant limits to a purely AI-powered approach.

Black Box

Very few - if any - AI-powered personalisation platforms are transparent as to how their algorithms actually work. If you’re getting the results you want, great. But what happens the day your KPIs start to slump, and you can’t take a peek behind the curtain? And how can you make sure you’re in line with your industry’s compliance regulations? 

For example, banks are prohibited from considering race, religion, national origin, gender, and other criteria, in determining whether or not to provide their services. This is a problem if their marketing team, say, wanted to use Facebook’s advertising feature ‘Lookalike Audiences.’

The feature uses algorithms to identify demographic attributes from a customer list to use in targeted marketing campaigns. Since they’re essentially a black box — you can’t identify whether the algorithms are in fact using any of the above criteria — compliance teams can’t be sure they’re not running afoul of the relevant legislation.

AI-powered personalisation platforms will certainly cause these same kinds of problems — especially considering that the General Data Protection Regulation, which further restrains the use of the very kind of data that powers personalisation campaigns, is coming into effect May 25, 2018.

Uncanny Valley

Masahiro Mori famously wrote in 1970 about the ‘uncanny valley’ - the feeling of eeriness and even revulsion elicited by humanoid technology almost realistic enough to fool us into thinking they’re alive - but not quite.

There are many examples of uncanny personalisation, like Clive Thomson’s recent description in Wired of how his Google app creepily sent him a curated photo scrapbook of his vacation, totally unbidden. He went on to explain an entire study conducted about this kind of high tech customisation. The results? “The more personalised the services got, the more people liked them—until they got too personalised. Then they seemed freaky.”

If we continue to use artificial intelligence to push personalisation to its maximum potential, are we just hurdling straight towards the uncanny valley?

Echo Chamber

As Kartik Hosanagar opined — again in Wired, “Many of us seem to feel trapped in a filter bubble created by the personalisation algorithms owned by Facebook, Twitter, and Google.”

What he meant was that personalisation algorithms, by their very nature, are designed to identify users’ interests and ‘personalise’ the content they serve up by giving them more of the same, whether this means news stories, product recommendations or TV series. 

While this might make a user’s experience more enjoyable, it also closes them off to many experience of new or different ideas, and certainly doesn’t do anything to combat the natural tendency towards confirmation bias.  Will consumers soon get fed up of feeling ‘trapped’ in these echo chambers?

Connected, But Alone?

Is there something a bit sad about an algorithm doing its best to mimic human interaction? Like when a website headline reads, ‘Hello Robin!’, or when a chatbot sends you your plane ticket and wishes you a nice day?  In the long run, will these kind of ersatz human interactions take on a toll on our capacity to socialise?

MIT professor Sherry Turkle has written extensively about how modern technology has warped society’s sense of togetherness. In her words, “we’re designing technologies that will give us the illusion of companionship without the demands of friendship.” In the long run, will extensive, AI-powered personalisation technologies only contribute to the problem? 

3. Machines Don’t Always Get It (Enter Psychology)

There’s a saying we use: Your ROI is only as strong as your ideas. And we find that the best ideas come from people, not machines.

It’s true, algorithms are best suited to sorting through mountains of data, spotting patterns or making predictions. But coming up with optimisation ideas based on a sophisticated understanding of human behaviour and attitudes, to match these machine-driven insights with the right message and the right trigger? This is a typically human endeavour.

One perfect example is how CRO experts exploit cognitive biases to increase their conversion rates. Quoting from our very own CRO Pocket Glossary, ‘cognitive bias’ is:

“The systematic tendency, used by the brain as an information processing shortcut, to base judgment, memory, decision-making, etc., on one’s personal frame of reference instead of on rational logic.”

Cognitive biases like the Von Restorff Effect, the Bandwagon Effect, the Peak-end Rule effect and others can be used to great effect to understand user behaviour, and therefore influence decision-making (i.e. conversion). While a human CRO expert can study up and learn over time how to best exploit these very human tendencies, it’s more difficult to explain this to a machine.


There are benefits to AI-powered marketing technology, there’s no doubt about it. But this doesn’t mean we have to throw ourselves in head first. Technologists have a responsibility to understand and reflect on the impact their endeavours are having on society at large, and especially digital consumers. Algorithmic tech does have limits, and isn’t the best placed to solve each and every CRO challenge. For that, you need a human expert at the wheel, to drive, oversee and keep checks on the increasingly prevalent AI-powered tech.


Join thousands of other Online Retail professionals

Get unique insights straight to your inbox for free, and improve your understanding of online retail. Subscribe to Online Retail Weekly now.

Webinar Scroll Banner
Join thousands of other Online Retail professionals

Get unique insights straight to your inbox for free, and improve your understanding of online retail. Subscribe to Online Retail Weekly now.

Webinar Scroll Banner