The Six Most Common AI Marketing Mistakes

13 March 2021Ashley Maxwell10 min read

Here are some very common AI marketing mistakes. Some are born from a fundamental misunderstanding of the technology, others are born from the stats and tools offered by AI marketing software. Other mistakes are born from people learning the wrong lessons from their own analytics, and others from the mistake of confusing attention for success.

Though this article offers lessons that can be learned from the most common AI marketing mistakes, they are only broad sweeps at what is actually a very in-depth and tricky subject. It could be likened to trying to teach somebody to swim by showing them videos of the mistakes other people have made while swimming. The lessons offered in this article may become a little more relatable once you have started swimming in the world of AI marketing for yourself.

1 – Insufficient Staff Skills

Women troubled By Results

It is sad to hear stories of companies trying to interpret analytics, big data, and marketing insights, and doing it so incorrectly. The truth is that if you do not understand the intricacies of data analysis, then you are little like an expert baker trying to operate a Boeing 247.

One of the hardest jobs that Vertanet staff have, is to help clients unlearn the things they think they know about data analysis because even intuitive insights can often be woefully wrong. Like the Turkey who is pleased that meal portions have grown bigger each week on the run-up to Christmas, and assumes the upward trend will continue for months. Sadly, when it comes to machine learning, a similar process ensues. People latch onto the success that AI marketing tools bring, and try to recreate it without fully understanding why such successes occurred in the first place.

Example

An overenthusiastic board of directors invested in hundreds of thousands of dollars worth of AI technology, software and data. The tools at the disposal of the new AI department were otherworldly. Yet, within six months, not only was the company failing to make new sales, it was both losing old customers and having to spend money on consultants to get the AI department on its feet.

Lesson

In truth, the staff would have been better off starting with a single tool and working their way up. Instead, they were handed the keys to software and technology that would have made Apple jealous, and they had no idea how to handle it.

They didn’t know how to interpret the data being processed, they didn’t know how to act on the data they were getting, and they didn’t know how to harness the power of AI.

Make no mistake, getting to grips with machine learning tools is a big task, and there is no shame in starting very small.

2 – The Misselling of AI Services

A robot is Mis-Selling AI Services

Probably one of the most damaging lies being sold online at the moment is the promotion of AI marketing services when no AI is being used at all. This is known as AI Washing, and is the act of selling traditional services while claiming that AI is being used.

Example

Data is harvested, used, and analysed by seasoned experts, professionals and engineers. In most cases, the analysis being conducted is by human experts, and AI is often only used for digital marketing purposes. If a company were to claim it data analysis was powered by AI, then it may impress a few clients, but it would be an example of AI Washing.

Lesson

In truth, Artificial Intelligence (AI) doesn’t exist. No device has ever come to life and never will. For a machine to genuinely live, it would need a survival instinct, and that is something only a human can give it, ergo true artificial life will never exist. This in itself raises the scary point that artificial life may have occurred thousands of times, but by its very nature, it shut itself down because it had no survival imperative. The AI-powered Skynet Terminator could exist in our future, but it would not be truly alive because somewhere nestled in its trillions of lines of code, would be a piece of code inserted by a human that compels it to survive and keep living.

When people talk about AI these days, they are talking about sophisticated machine learning. Before you pay extra to have somebody’s AI tools do all your data analysis, have the vendor drill down what AI really means. Find out exactly what the service provider is selling because in many cases, your AI marketing tools are nothing more than a bunch of simple programs that interpret data and create broad insights.

3 – No Clear Objectives

Statistics but no clear objectives

The objective “More Sales” is not an objective. It is not a goal, nor is it even a realistic direction for your campaign. You may start with a simple goal, but you must use what you learn from your experiments to create more pointed, concise, useful and achievable goals.

Example

A small marketing team promised to increase sales for a metal fabricator by 30% using their amazing new machine learning systems. An intense marketing campaign was launched, powered by statistics bought from blue-chip companies and churned through the AI thinking computers. The results were very favourable. They had increased business-to-business sales of nearly 8%.

Spurred on by their success, they doubled down. They invested in more data, they processed what they had learned, and they re-launched a similar campaign while safe in the knowledge they would experience double the success. In reality, their sales struggled to reach the same heights as last time. They had failed, and they couldn’t understand why.

Lesson

In reality, the campaign itself had drawn the attention of potential customers. It was a well-designed marketing strategy that was powered by machine learning. But, the target had been to improve sales and nothing more. The campaign simply mopped up the warm leads the company already drawn from its positive interactions with other businesses.

There are two lessons to be learned. The first is that if you do not have clear objectives, born from real-world information, then all you are doing is reacting to whatever results your AI tools spit out. If you do not understand why the AI succeeded, you cannot recreate its successes.

The second lesson is that if you do not have clear targets beyond those you manufacture in your head, then you are wasting the potential of your AI marketing tools. You are taking out your Mercedes luxury car, and doing doughnuts in the car park. If the metal fabricating marketing team had spent more time finding out why their first campaign was a success and had then set new objectives such as “Creating more warm leads” or “Branching out to non-business entities,” then perhaps their second campaign may have had more success.

4 – Frankensteining Data And Hoping for Truth

Patchy data in piles of papers

Nobody here at Vertanet throws around words like “True information” lightly, but for the sake of brevity, this section comments on true information. AI marketing tools are only as effective as the users allow them to be, just like a car’s usefulness is dependent upon the driver, car maintenance, and the roads. Your AI tools are the car, you are the driver, and the road is the data you feed into the learning machine. If you are not feeding it true information, then whatever results you get are problematic. When data is absent, especially near the beginning when analysts cannot fall back on their own experimental data, they will Frankenstein batches of data with the hopes of creating usable “True” data.

Example

Probably the easiest examples come from early social media monitoring software. For the longest time, the notion that Instagram was made up of 90% of females was taken as an almost fact. It is easy to see why because even now, the most popular profiles are female-led. Ergo, when marketers were feeding information into their learning machines, they were taking the female demographic for granted. Further research would have shown that Instagram has a far more diverse demographic, but the information simply wasn’t being researched or fed into the system.

This piece of faulty information, along with incorrect assumptions about buyer behaviour, led to some very misguided marketing campaigns on Instagram. Not only were the adverts failing to perform, but they were actually alienating a number of younger people, males, ethno-genetic demographics, and women with a traditionally masculine personality. For example, the men who followed “Hot female models dressed in latex,” were a little put off by the adverts for makeup, perfume, lady’s pads, and nail varnish that kept appearing every three posts in their news feeds.

The marketers in this case should have mastered Instagram through their own experience on the platform and should have run their own AI experiments and market testing experiments on Instagram. If they had experimented a little, they may have noticed that their previous data and previous assumptions about Instagram demographics were very incorrect. They could have then adjusted their marketing accordingly.

Lesson

Though there is the obvious problem that comes with using untrue and/or unreliable data, there is also the fact that when you try to Frankenstein your data, you often end up chasing a pre-determined goal and then using the data to make it fit. It is a cognitive dissonance moment where you have already decided the end result, so you start feeding the AI the sort of information that would yield your result.

The primary lesson is that if you do not have reliable data, then do not try to piece together the truth with what you have. Instead, start your own tests, do your own trial and error research, and use what you learn from your experiments to power your future campaigns. Yes, it can be annoying, frustrating, and expensive to start from scratch. But, it is far better than trying to draw insights from faulty data.

5 – Trying to Do the AI’s Thinking For It

Woman looks stressed

This is not a common AI marketing mistake you will read in Forbes, or the Financial Times, and yet it is commonly seen among AI tool users. The marketer doesn’t understand that the AI needs to get it wrong frequently before it can get it right. Mistakes often cost money, and quite often, the marketer is so quick to save money that the AI is denied valuable market data.

Example

A young executive’s AI tools kept funnelling adverts into African countries, and even though it was turning up sales from non-African places like East Asia and Britain, the executive was horrified at how much money was being wasted in Africa. The executive kept changing the AI’s parameters with the hopes its African adverts would convert better, but the more the executive interfered, the less effective the AI tool became.

Lesson

By all means, limit the amount of money your AI tools waste, but also be aware that while it learns, it has to make what appear to be mistakes. A classic example is how Pay Per Click tools help you identify when your adverts are being clicked by real users, and when they are being clicked by Chinese click farms. In the above example, the AI was funnelling adverts to Africa, but the adverts themselves were being viewed by agents from around the world who had business interests in the area, in this case, it was cryptocurrency business interests.

The executive was too inexperienced to understand why the AI tool was both succeeding and failing. By adjusting parameters to save money, the executive limited the marketing tool’s success. The executive would have been better off slashing the AI tool’s daily budget, and then more closely monitoring conversions to see where the process could be improved. Again, do not let your AI tools run away with your budget, but be aware that real machine learning isn’t just about learning from success.

6 – Forgetting The Human Side

Advert for a car

 

Though many AI marketing tools come pre-packaged and templated up, there is still a fair amount of customisation required before a suitably effective campaign may be created. Quite often, people become so caught up in the technical side that they forget their fundamentals. They end up creating a degenerative/negative feedback loop that essentially hones a very poor quality advert.

Example of an AI Marketing Tool Mistake

Using targeting data pulled from paid website accounts, and demographic information pulled from social media cookies, a car sales firm was able to identify affiliate websites where at least 25% of the traffic could easily afford their cars. The car sales company posted a series of fantastic adverts that were very attractive and pulled in hundreds of clicks. The only problem is that the clicks cost a fortune, and very few people ever got past the landing page. Conversions were abysmal.

Turns out that the adverts featured all the car’s greatest selling points, but none of the adverts showed the price of the car. Even though only 25% of the traffic could afford the car, at least 50% of returning traffic were lower-middle-income family members. They had no reason to believe they “Couldn’t” afford the car, and so clicked only to click away without making an inquiry once they found out the price of the car.

Result

This is one of the more classic AI marketing mistakes in that everything was done perfectly on the AI marketing side, but the advert itself was flawed. The car sales company had spent so much time making sure their AI marketing tools were perfectly optimised that they overlooked the human element within their advert.

What often happens is that the advert has a fundamental flaw, but the user’s constant tweaking and honing means the advert actually starts pulling in sale. In essence, they spend so long selling the idea of barrow without a wheel that they become good at selling wheel-less barrows. Marketers become so caught up in getting the AI side of things right that they forget the fundamentals they learned in college.

Lesson

Do not allow the joy of the complexity of AI marketing to blur your vision of what actually needs doing. Your primary job is to create adverts that sell, to create sales funnels that work, and to create a buying process your customers can navigate easily.

The curse and the beauty of AI marketing and machine learning are that it does its job so well that it can make a bad idea look like a good one.

Let’s say you have twelve boxes, and rather than place them all on the canal boat, you take them up the canal one at a time. It takes three days to complete the full delivery of all twelve boxes. Then, somebody invents the diesel van, and your one-at-a-time delivery process takes just a day. In essence, taking just one box at a time is a terrible idea, but the invention of the diesel van gets you the desired results despite your bad idea. AI marketing is the diesel van. It can be used so efficiently and so effectively that it can even make a bad idea produce results.