Opportunities for AI in Content Marketing Easily Explained

Until recently, the closest I’ve come to understanding artificial intelligence is knowing that it powered tools in my martech stack (e.g., marketing automation, predictive lead scoring, etc.).

Beyond that, I found the concept hard to grasp until Chris Penn’s presentation at Content Marketing World, How to Use AI to Boost Your Content Marketing Impact.

Chris, co-founder and chief innovator at Trust Insights, covered several real-world applications of AI. His examples helped transform abstract concepts into tangible use cases.

Chris implemented these examples himself via hands-on coding in the R programming language, using a deep understanding of mathematics, data science, and machine learning. But most marketers don’t have data science and computer programming skills. Later in this article, I share Chris’ advice about how marketers can apply these AI concepts.

Here are several of Chris’ experiments.

Driver analysis: What results in profitable action?

When you have a bunch of data but you’re not sure what matters to the outcome you want, driver analysis is an effective tool, Chris says.

Machine learning software excels in this case. You feed in all the data and it tells you what matters in it. Chris explains that the analysis concludes with something like, “Hey, this combination of variables seems to have the strongest mathematical relationship to the objective you want.”

Chris performed driver analysis on the popular PR and marketing blog, Spin Sucks, where the primary business objective is lead generation.

“(It) determined that organic search was the third most powerful driver. The team focused a lot of time and energy on it, and they should, but email was the No. 1 driver,” Chris says.

By understanding better what drives leads, the Spin Sucks team could decide to shift more of their time to email marketing because it was the most effective source.

Whether your objective is page views, social shares, leads, or revenue, a ranked list of drivers can help you plan resources, priorities, and budgets more effectively.

Implementation detail: Chris used the R programing language to implement Markov chain attribution. For a detailed look at one such implementation, read this post by data scientist Sergey Bryl, which will give you a good sense of how much mathematics and data science is involved.

Text mining: Reveal topics, keywords, and hidden problems

Text mining is an application of AI that ingests content (e.g., text) to classify, categorize, and make sense of it.

Chris notes that text mining uses vectorization, which transforms words into numbers. It looks at the mathematical relationship among those numbers and determines how similar those words are. It is a form of deep learning.

Reverse engineer Google to reveal key topics and terms

The Google algorithm, which uses a heavy amount of AI itself, is an example of a deep-learning system. “Google’s search algorithm is so complex now that no one knows how it works, including Google,” Chris said. “They have very little interpretability of their model.”

You can use text mining to reverse engineer the Google algorithm for your targeted topics. “We can deploy our own machine learning models to say, ‘OK, for a search term like content marketing, what words do the top 10 or 20 pages all have in common?’”

Here’s a sample output from reverse engineering Google:

The resulting lists hint at what words or categories to cover when developing new content around your reverse-engineered keyword. Having this set of common words gives you a higher chance of success with organic search than simply saying, “Let’s write a really good article about content marketing.”

Implementation detail: Chris implemented text mining and topic modeling via the R programming language, extracting related topics from a corpus of text (e.g., the contents of articles found in the search engine results pages).

Extract hidden insights via text mining

In 2014, Darden Restaurants, the parent company of Olive Garden, replaced its board. The new group implemented changes, including enforcing its existing but mostly ignored breadstick policy (serving one per person plus one extra).

As Chris explains, employees then spent their time enforcing the policy by counting the number of breadsticks in the basket based on the number of people at the table.

Chris used text mining on 2,500 publicly available reviews written by the company’s employees on Glassdoor. Here’s a glimpse of the results:

Text mining surfaced breadsticks as a problem. If Olive Garden was looking to repair low employee morale and a poor customer experience, a manual review of its Glassdoor reviews, where the usual restaurant worker complaints like low pay and long hours abound, may have led them down the wrong path.

Text mining revealed the breadstick problem. (After intense public pushback, Olive Garden returned to its previous breadstick approach.)

Text mining of unstructured data can be applied in many useful marketing contexts: customer reviews, poor/high performing blog posts, transcripts of customer success phone calls, etc. Extracting that hidden gem of insight can point you to courses of action with a high ROI.

Implementation detail: Similar to the reverse engineering Google example, Chris implemented text mining via the R programming language.

Time-series forecasting: Analyze competitors’ brand searches

Let’s combine math, statistics, and AI to create a Magic 8-Ball.

“Wouldn’t it be great to know what’s going to happen,” Chris asked. “It would be so much easier to plan, to set budgets, to staff, to have an editorial calendar.”

Chris did an exercise of predictive time-series forecasting for Cleveland hotel search data. He looked at more than 12 months of branded searches — where searchers named specific hotels (e.g., Hilton Cleveland, Holiday Inn Cleveland, Hyatt Cleveland, Marriott Cleveland).

The results predicted when search volumes go up and down for each hotel:

“If you (worked at) the Cleveland Marriott here, you now know that right around the end of September you have more search interests than your competitors. You could be running campaigns against them to take even more market share away from them,” Chris said.

Any brand could benefit from predictive time-series forecasting – analyzing brand searches for your company vs. your competitors. You can search for when your brand underperforms, for example, and use that data to bid on your competitors’ brand names with a relevant content asset or promotional offer.

“Imagine search topics, conversations, social media. You can forecast more than search volume,” Chris said. “You can forecast lead generation from your marketing automation software. You can forecast revenue from your CRM or your ERP. Anything that is regular data over time you can forecast forward.”

Implementation detail: Chris used R to process five years of Google search data, then implemented a statistical method called autoregressive integrated moving average (ARIMA).

How content marketers can try these AI uses

I know what some of you must be thinking about now:

• “Wait, really?”
• “The data science and probability are over my head.”
• “I’m too busy and can’t possibly learn to do this myself.”

These reactions are understandable. The good news is that you have options. And you don’t need to learn the deep nuts and bolts covered earlier.

Chris offered three recommendations for marketers thinking about approaching AI.

Do it yourself. This approach fits for the small percentage of marketers who have a genuine interest in data science and machine learning. You should be interested in going deep with math, statistics, and probability – and comfortable writing code.

If you decide to go this route, Chris suggests checking out Google’s Machine Learning Crash Course, available free online, which takes you through 40-plus exercises, 25 lessons, real-world case studies, and lectures from Google researchers.

Chris notes that IBM Watson Studio has an intuitive, drag-and-drop user interface. While Watson does enable programmers to write code on its platform, the UI can be useful for marketers who are not inclined to write code.

For those interested in coding, Chris recommends learning the R and Python languages, which form the basis for a lot of AI tools and libraries. Be prepared to spend six to 12 months to learn the programming language and another six to 12 months to learn the data science.

If you’re just getting started with coding, the Dummies franchise has books that may be useful: R for Dummies and Python for Dummies.

Tap your staff data scientist. The second option applies to larger organizations that employ data scientists (e.g., Google, Facebook, and Uber). “Staff with data science skills are quantitatively inclined and know how to use the technology properly, so they can be of great help,” Chris said.

Think back to the use cases I mentioned. For text mining or time-series forecasting, in-house data scientists will understand your objectives and goals, build the right models, then implement the necessary codes.

Outsource. This option works for organizations that don’t have AI and data science talent in-house. The answer is to outsource to the experts: people or agencies with the necessary AI know-how and experience.

Chris puts it this way: “Agencies and consultants can help you use the methodologies. You can do small projects on a one-off basis. If the need is ongoing or more frequent, they can help you build software that runs when you need it to.”

Next steps

No matter which of the three options makes sense for you, there’s one thing I urge all marketers to do: Learn about AI and understand the role it plays in marketing technology.

While you don’t need to understand Markov chain attribution or how to program in R, you need to know enough to determine where and how AI can help your marketing. Basic AI knowledge will also help you better evaluate vendor solutions and claims.

Think about the kind of knowledge you need to buy a computer. You don’t need to be a chip designer, but you need to know the difference between a 32-bit and 64-bit processor and whether a 1.5 GHz processor is better than a 2.7 GHz processor. With AI, when a vendor says, “Our predictive analytics solution uses the latest AI techniques,” you need to know how to question the claim and how to distinguish fluff from reality.

Since AI is a topic often covered in business, marketing, and technology publications, I’m soaking up as much as I can. Next, I’ll probably enroll in some free, online courses in machine learning.

What about you? What’s your interest level in AI for marketing and how are you staying informed and educated?

Here’s an excerpt from Chris’ talk:

Further your tech skills in 2019 by attending ContentTECH Summit in April. Register today using code BLOG100 to save $100. 

Cover image by Joseph Kalinowski/Content Marketing Institute