Marketing AI Institute | Blog

Exploratory Data Analysis: The Missing Step for AI [Video]

Written by Cathy McPhillips | Dec 5, 2022 1:37:04 PM

You’ve piloted a small test case using artificial intelligence. Excellent! Now you’ll need to show the success of your first test case. Long-term buy-in is dependent on the success of the programs that have been completed, so measurement and analytics are essential to set up before the launch of any program.

During his MAICON session, Christopher Penn explained how few things guarantee success with AI, but one thing guarantees failure: skipping exploratory data analysis. 

Penn, TrustInsights.ai’s Chief Data Scientist, showed MAICON 2022 attendees how to increase the odds of success for any AI project, whether with a vendor or in-house.

He explained: 

  • What exploratory data analysis is and isn't.
  • Why AI demands proper exploratory data analysis, and why so many companies skip this vital step.
  • How to conduct proper exploratory data analysis, from data integrity to feature selection to principal component analysis.
  • When to put the brakes on an AI project because your data isn't ready.

The good news? Chris is able to explain exploratory data analysis to any level practitioner. You don’t need to be a data scientist to read this or to watch this short clip. 

Watch Christopher Penn’s MAICON 2022 Session

 

Watch this full session plus all MAICON main stage keynotes, sessions, and panels with the MAICON 2022 On-Demand Bundle

Exploratory data analysis is a critical component of successful AI implementation. It’s a missing step that many marketers overlook or consider too late in the process.

Picture this scenario: You’re working on a blog post or an editorial calendar and ask yourself, “How can I create a great performing article?” This is a clear user story, and has a well-meaning intent. We want to create great, meaningful, and helpful content. From this question, we can begin to take into account the considerations that might go into us creating amazing content, but we also need to carefully keep these questions in mind:

  • Do we care about bias? 
  • Do we care about fairness?
  • Do we care about an outcome that could be misinterpreted? 
  • How could this go wrong?

NOW is the time to answer those questions, because depending on your AI project, many things could go wrong. 

In the video clip shared above, Chris says data is to AI as ingredients are to cooking. Meaning, “if you have bad ingredients...I don't care how expensive the blender is or how good a chef you are, if the ingredients are spoiled, you're not making edible food.”

Likewise, bad data in = bad models out. AI is not a magic box, and unfortunately, an astonishing number of companies skip the part of the process where good data is needed to surface good information and output. 

The onus is on us as marketers, but some of this responsibility goes back to the technology solutions and partners. If you hire someone to implement your machine learning, this process, exploratory data analysis (EDA), is critical on all fronts. 

IBM says, “Exploratory data analysis is used by data scientists to analyze and investigate data sets and summarize their main characteristics, often employing data visualization methods."

Penn points out the most important part: "The main purpose of exploratory data analysis is to look at data before making any assumptions." 

He continues, “When you talk to a vendor and you say, ‘Hey, I want better attribution analysis,’ or ‘I want better subject lines,’ or ‘I want better content.’ you are making assumptions. You have made assumptions about the problem before you've looked at the data. So you look at your blog and say, Well, gosh, our blog, we're not getting enough traffic from Google. We need AI-generated content. How do you know? You haven't looked at the data yet.”

The exploratory data analysis process has eight questions you need to ask, or steps you need to take: 

  1. What are you trying to do? 
  2. How did you get the data?
  3. What are the characteristics of the data?
  4. Examine the data.
  5. Make sure the data fits the problem. 
  6. Prepare the data.
  7. Engineer the data.
  8. Deploy the data.

Continuing with the food analogies, Chris said, “Well, a lot of it's like, okay, you can like bring a personal chef into your house, but you still need to look in the fridge to make sure that you've got the ingredients.” 

Before you start too far down a journey with your marketing and AI partners, take a step back to start off on the right foot. Trust Insights and other partners can assist you if you’re ready to get started but don’t know where to begin. Chris also wrote a series on 5 Ways Your AI Projects Fail which was republished on our blog.

Become a next-gen marketer by checking out the resources at the Marketing AI Institute. Read our blog posts, take our Intro to AI for Marketers class, attend webinars, join our community, download reports, guides, and templates (all free), read Marketing Artificial Intelligence, look into AI Academy for Marketers and Piloting AI Bundle, and our annual MAICON—Marketing AI Conference.

Get access to all MAICON main stage keynotes, sessions, and panels with the MAICON 2022 On-Demand Bundle