How machine learning can perfect your pitching

Here’s how to use technology—much of it free and user-friendly—to elevate your media relations and improve your journalist outreach.

AI for PR efforts

You may be dazzled, spooked or annoyed by the ability of online retailers to predict which products you’re interested in purchasing—but what if you could play the same game?

Whether it’s Amazon recommending soap or Netflix suggesting a movie, more companies are making personalized predictions using variations on a machine learning tactic called market basket analysis. This technique uses an algorithm that sorts through behavioral data to determine how frequently certain actions (purchases, views, etc.) are associated with other actions. The algorithm provides the statistical likelihood that if one action takes place, another desired action is likely to follow.

Sophisticated algorithms aren’t limited to shopping carts or movie recommendations, however. The same principles can be applied to all kinds of behaviors and disciplines—including PR. PR agencies can use algorithms, AI and machine learning to better target journalists by predicting which ones are likely to write about specific topics based on what they’ve previously written. This might sound expensive or beyond your expertise, but it’s neither. You can harness the power of machine learning with free software and data that you probably already have.

Here’s an example of using market basket analysis to identify reporters likely to write about “probiotics” based on their previous coverage:

Step 1: Collect the data. We built a dataset in Excel with all the articles that ran over a 30-day period in 2018 on several infant health topics:

  • Antibiotics
  • Constipation
  • Diaper rash
  • Eczema
  • Asthma
  • Allergies
  • Childhood obesity

Meltwater, Cision or any other media monitoring platform can provide similar data.

Step 2: Clean the data. The original dataset contained more than 5,000 articles. However, many of these were syndicates, press releases and company blogs.

To ensure results accurately represented patterns in journalist coverage, we eliminated more than half of the articles, leaving about 2,500 original articles written by actual journalists.

We also did quite a bit of “normalization,” which is a fancy way of saying we made sure spelling was consistent.

Step 3: Load (free) software. To analyze the data, we used R Studio, a free statistical software that’s commonly used by data scientists to build machine learning models. Python Spyder is similar free tool.

Step 4: Build the model. We built a market basket analysis model based on the Apriori algorithm to, as SearchBusinessAnalytics puts it, “uncover relationships between seemingly unrelated data.”

Market Basket Analysis might tell you, for example, that if you bought a “Star Wars” DVD and also a “Harry Potter” movie, you’re two times more likely to purchase a “Lord of the Rings” film. In this case, we analyzed how a journalist’s coverage of certain topics might affect the likelihood that he or should would cover a topic we want them to cover.

Step 5: Analyze the results. Our model yielded the following info:

“Antibiotic” + “Constipation” > “Probiotic”/ Lift = 3.7

“Constipation” + “Eczema” > “Probiotic”/ Lift = 1.06

“Allergies” + “Eczema” > “Probiotic”/ Lift = 1.05

This might look scary, but the main measurement we’re looking for is “Lift” on the right. This is another way of expressing how much the desired behavior—writing an article about probiotics—is increased by combinations of other topics. A “Lift” larger than 1.0 means you’re likely to see a behavior leading to another behavior—and the larger the Lift, the greater that likelihood.

Antibiotics and constipation: A winning combination!

In this case, we found that reporters who had written about “constipation” and “eczema”—as well as “allergies” and “eczema”—had respective Lifts of 1.06 and 1.05. This means that if they had written about these terms, they were slightly more likely to write about “probiotics.” However, for a reporter who’d written about “antibiotics” and “constipation” within the last month, the Lift of 3.7 implies a stronger relationship. In other words, a reporter writing about these topics is 3.7 times more likely to write about “probiotics.”

If you use standard media monitoring tools, you’re basically sitting on the “Big Data” gold mine we hear so much about. It’s easy to identify reporters who have covered certain topics, and when machine learning is applied to that data, you can eliminate (some) guesswork from media targeting.

Sometimes, the best predictor of future behavior is past behavior. This basic principle reveals why every PR pro should take advantage of technology that can hone your pitches and make your media targeting more successful.

Michael Burke is an account director at San Francisco-based PR firm MSR Communications.

COMMENT Daily Headlines

Sign up to receive the latest articles from directly in your inbox.