Creativity and AI: Coexisting in content creation

I am a relatively non-technical person working in AI (Artificial Intelligence), a traditionally technical space. AI is commonly associated with complexity, sci-fi, and large amounts of data. And while I am a fan of science fiction, as a Creative Director, my work requires me to live in Figma and create endless Google docs. A far cry from complex datasets or machine learning algorithms.

As The Economist wrote back in 2017, "data is the new oil," which is used to fuel better marketing strategies. Since then, marketing has continued to gravitate towards using data and AI to build personas, predict campaign performance, and reach audiences in infinitely better ways than before.

Let me tell you how I embraced machine learning and made it central to my creative work at Obviously.ai.

Identifying the real challenge 

In 2019, I was leading our creative content initiatives when we launched our no-code machine learning platform. We were facing a few challenges. First, we needed to find out exactly who our existing audience was very quickly. As a startup, we’re in a race against time, making a quick start to understanding our user personas critical. 

Secondly, we needed to identify and create content that would resonate with our audience. It was the classic launch conundrum. To accomplish this, we created a survey and sent it to our users. This survey helped us collect a vast amount of user persona data in a matter of weeks and apply it to our content strategy. Here’s how we did it.

Getting creative with data

When we were launching Obviously.ai, I wanted to get an idea of which user persona was most likely to be satisfied with our platform. The end goal was to use the survey data, coupled with no-code machine learning, to discover the relationship between the demographics and their interaction and satisfaction with the platform. 

Precisely, I wanted to know:

• How frequently our customers used our platform 

• How satisfied customers were with our platform

• Which features customers liked most

I knew this would help me:

• Target content to the most interested user demographic

• Create strategic content tailored to the specific user personas

• Provide insight for product features to promote on the blog 

I designed a survey with this goal in mind and shipped it out to our user’s inboxes through Hubspot. User persona surveys are great for seeing how satisfaction relates to the type of plan a user is on, their title, company size, and what they use your product for. 

I made the questions really concise because we knew we’d run predictions with the data. For example, most of the questions were simple drop-downs, where users could select their company type, payment plan, satisfaction, and more from a few options.

A best practice for making great predictions with tabular data is to make the values as short as possible (1-3 words) and mostly quantitative. Since I wasn’t making a sentiment analysis or time series prediction, it did not make sense to use an open response. I wanted to minimize the text to make my data visualizations as easy to understand as possible.

Working with the predictions 

My dataset looked something like this. It had an identifier column, user attributes, satisfaction, plan type, and the kind of features they used.

After downloading the spreadsheet into a CSV file and uploading it to Obviously AI, I was able to see what kind of user was most satisfied with the tool. In particular, the data identified which user persona is most satisfied, including the above categories (company type, job title, machine learning skills, whether or not they use our data store, etc.).

These categories are important for us, as they help us understand who to target. Here’s how it looks:

While letting the user rank their satisfaction 1-5, I found those who had little to some coding knowledge were most likely to be satisfied with Obviously AI. 

This identified our most satisfied core users. It also concluded we still had some work to do to appeal to more technical users.

Here’s another example:

Again, those with little to mid-range machine learning skills seemed to be the happiest. From my experience, here’s a tip: The more columns you have—or questions you ask in your typeform—the more information you have and the better predictions you can make. Though it’s also important to make your form short enough to still have a great completion rate.

Let’s also take a look at satisfaction versus a specific feature within our platform: The Data Store.

Our Data Store is a feature inside our platform that allows users to make predictions with clean public data to play around with and get inspiration from. What’s interesting here is that those who used the Data Store were far more likely to be happier with the product than those who didn’t. This helped us realize we needed to advertise the Data Store and make it a bigger part of our onboarding experience and marketing so users understood its value.

Based on the personas we found, I could predict the satisfaction of different kinds of users. These predictions helped us see which users are more likely to be serious customers, so that we can prioritize content for them. This was just what we needed to create content that would resonate with our audience.

Data-packed results worth writing about

With the Typeform responses, I was able to get a much clearer picture of how our customers used our product. And how satisfied they were.

In our feedback survey, we found out Obviously AI had the most value for:

• Non-technical teams who wanted to focus on being domain experts

• Business analysts who usually use SQL for analytics

• Small to medium sized businesses that wanted to jumpstart their AI system

• SaaS, marketing, and creative agencies

I could now let the data drive my creative content creation. This prevented wasted time and resources and led to our audience resonating more with our mission to democratize AI.

For example, we created blog posts geared toward our less technically-inclined customers in order to introduce them to simpler AI concepts. Topics like: how to use Google Analytics and make the most of the data. Or, a step-by-step guide on how to get started with data-driven systems.

We’re one year into this beta phase and so far, have gathered over 3,000 blog subscribers and are receiving approximately 400 signups a month. It’s a huge improvement from the 15 signups we averaged before I sent out the survey.

The survey response data truly helped us achieve a hockey stick curve and allowed Obviously AI’s creative content to find our early adopters. As a non-technical person trying to use data to boost my creativity, I was happy to finally be able to take a data-driven approach to my work.