I’m an absolute geek when it comes to artificial intelligence. I truly believe that it will revolutionize marketing, and the rest of our daily lives in the long run.

The marketing AI revolution will enable us to accurately target our audiences one-on-one, giving them the right content at the right time in the right medium. It will identify our target market before our target market even knows they need a product like ours. And, yes, it will probably create all of this content for each person based on their individual traits, backgrounds, and biases.

But right now, it kinda sucks, especially if you don’t know what to look out for.

If you’ve been flirting with the idea of implementing any form of marketing AI into your martech stack, here are some things you should look out for.

Over promising and under delivering

I had high hopes for AI in 2016 when I co-founded Ceralytics. The goal of our company was to have AI help determine what content was resonating with audiences at the awareness, engagement, and conversion stages of the buyer’s journey. We built models, ran tests and did some fancy AI-powered stuff.

But it didn’t work.

We had used IBM Watson—the same IBM Watson that conquered Jeopardy!—for our natural language processing (NLP) but had got caught up in the hype of the brand. On the surface, Watson seemed like a magical creation that could do everything perfectly.

But when we saw the results, we realized how far from perfect we were.

Even mighty Watson brought back results that fell well short of our standards. When we sent our data to our testing group, they’d shrug and tell us, “I get why it came to this conclusion, but I don’t agree with it at all.”

Our failure led us to reconfigure our entire NLP stack, which allowed us to turn out average results into great results. These early failures hammered home that we shouldn’t blindly trust in the marketing promises of AI system providers.

While Watson still plays a big part in our company, we recognize that it has its limitations, and we’ve put in other systems and algorithms to make up for its shortcomings.

Looking back, I realize that AI wasn’t the solution to our problems, it was just a part of it. And we wouldn’t have realized that without first testing it and then being willing to admit that AI wasn’t the entire answer to the problem we were trying to solve.

Blindly relying on technology because a vendor over-promised is probably the biggest way to get burned in marketing AI right now. You really need to know why AI is the best answer to the problem you have. It shouldn’t just be a shiny new toy.

Garbage in, garbage out

Marketing AI requires a ton of data. Machine learning algorithms—the technology that makes up the bulk of marketing AI solutions—learn from the data you have, looking for recurring patterns that could get lost in a huge spreadsheet. If the data you feed your AI is bad or incomplete, your AI won’t be able to pick out genuine patterns.

If your goal is to mine data to understand what kinds of content converts users from browsers to buyers, you need to have solid data through-and-through.

And it’s not just a matter of exporting all of your marketing data, dumping it into a special machine, and letting the machine work it all out. If your data isn’t in order, you’re not going to get orderly data back out. To quote Christopher Penn of Trust Insights, “80 percent of time in AI is cleaning data.”

In short, unclean data will produce incorrect results. Make sure you’re taking your time cleaning your data inputs to verify everything is in order so your outputs are accurate and useful.

Sometimes old fashioned statistics do the trick

Last month I headed up a project that helps predict online patterns. It seemed like a perfect job for marketing AI. However, the data we needed to analyze was very hard to retrieve, so we only had small fragments of data to start—much too small to run in structured machine learning.

So we broke out Excel and started doing things the old fashioned way.

After looking through the data for some obvious patterns—there were very few—we started visually plotting various columns of data. Through this methodology, we found several correlations between items.

A few days of trial and error later, we ended up with a statistical model that lined up extremely well with the sample data. After bringing in more sample data as a proof, we were satisfied that we had the winning result and we had achieved it without AI.

However, data analysis isn’t for everyone. If you mention regression analysis at a project meeting, you’ll often see people’s eyes glaze over. But in many cases, traditional data analyses can generate great results at a fraction of the cost of AI solutions. So explore traditional data analysis before buying a cooler AI product.

Don’t underestimate humans

In the spring of 2018, Telsa’s Elon Musk was in a bind. Musk had promised that Tesla would be producing 5,000 cars per week by the end of the summer. But Tesla’s focus on full automation had backfired as its machines kept breaking down, which held up the production lines and prevented  the company from meeting its production goals.

Tesla’s factory was producing just 2,200 cars per week—less than half of what Musk has promised.

So Musk thought outside the box. Well, outside the building. He built a new production facility outside of the factory in a tent—yes, a tent—using any spare parts and components they had to hand. In this tent, Musk ordered skilled laborers, not machines, to take over production.

The tent was the polar opposite of the automated factory assembly line—but it had a dramatic effect. By the end of the summer, Tesla was hitting its 5,000-car weekly goal and it was thanks to people, not machines.

The same is true with marketing.

Marketing AI can take you a long way, but it’s up to humans to do the work of making the content, promoting it well, and connecting with an audiences. Without a human component, the best AI in the world just isn’t going to cut it against a skilled, data-driven marketing team. At least, not yet.

It’s not magic, it’s statistics

As a co-founder of a company that relies on machine learning, I’m probably not doing us any favors by pushing back against the use of AI. But I’m seeing a lot of people jumping into the deep end without any idea of what it will take on their part.

Marketing AI isn’t magic. It’s statistics, exceptionally good statistics. And because of that, we need to make sure that it’s the best solution to the marketing problems we have before going all-in on the technology.

What about you? What experiences have you had with marketing AI technologies so far?

The Author

Brandon Andersen is the Chief Strategist at Ceralytics, a content intelligence company that helps organizations understand what content and topics are resonating with their buyers. Bringing together over a decade of marketing and product development experience, Brandon oversees product development, marketing and sales.

This article is about
Related Topics:

Comments

COMMENT GUIDELINES

We invite you to share your perspective in a constructive way. To comment, please sign in or register. Our moderating team will review all comments and may edit them for clarity. Our team also may delete comments that are off-topic or disrespectful. All postings become the property of
Content Science Review.

Events, Resources, + More

Workshop: Are You Ready for AI?

Is your organization really ready for AI at scale? Let the Content Science team guide your leaders through assessing 4 areas of readiness.

Course: Prompting Text Generative AI

Learn how to bring out the full potential of text generative AI to create impactful content from this on-demand course.

Webinar: Benchmarks for Content Effectiveness

It's not about more content. It's about more effective content. Gain tips based on Content Science's unique research + experience.

The Ultimate Guide to End-to-End Content

Discover why + how an end-to-end approach is critical in the age of AI with this comprehensive white paper.