Logo

The Data Daily

Automated Data Storytelling Is Not the Future of Analytics — Juice Analytics

Automated Data Storytelling Is Not the Future of Analytics — Juice Analytics

Data Storytelling
Automated data storytelling is the future of analytics.
That’s the argument put forth by James Richardson during a conference hosted by automated data storytelling vendor Narrative Science (as reported here ).
I've spoken to Mr. Richardson on a couple of occasions and deeply appreciate his understanding and enthusiasm for data storytelling. He's been a champion for data storytelling at Gartner for years.
","url":"https://media.giphy.com/media/pIMlKqgdZgvo4/source.gif","resolvedBy":"embedly","floatDir":null,"providerName":"Giphy"}" data-block-type="22" id="block-yui_3_17_2_1_1625748869918_6127">
It is his modifier ‘automated’ that worked me into a Stephen Few -style lather. His argument is as follows:
Business intelligence has struggled for adoption (yep)
Data storytelling represents a path to greater adoption (I sure hope so)
Few people in an organization have the capabilities to create data stories (true, for now)
Therefore we should get the machines to do it so we can create a lot more data stories, leading to more people engaging with data (…and we’re off the tracks)
I think I've been fair to the argument, as it was similarly summarized by Narrative Science's own Dr. Nate Nichols here . This line of thinking breaks down on multiple fronts.
What are you calling data stories?
First, we need to look at what Mr. Richardson and Dr. Nichols are referring to when they talk about "automated data stories." They are pointing to automatically generated text that can look like this:
This is the same programmatic approach that has started to generate short newspaper stories, particularly for financial and sports summaries ( The Rise of the Robot Reporter ).
There is nothing wrong with a text summary of data that pulls trends and outliers from the data (though you could argue it would be better done with a visual summary to show change). But is that a data story? To me, it is closer to a dashboard expressed using words, or a USA Today factoid.
We should set our sights higher if we are going to use the term ‘stories’. We should aspire to the quality of Hans Rosling or the New York Times design team (examples: 20 of the best data storytelling examples ). A bunch of computer-generated sentences summarizing data is not in the same ballpark as a thoughtfully constructed data story.
This is not to say that these machine-generated summaries don't have a place. I need a weather alert and appreciate that it is expressed in a sentence.
Richardson’s describes data storytelling as
simply the translation of data into common language in order to inform the decision-making process...a narrative about the numbers.
That's a reductive description of data storytelling that I’m not ready to accept. I prefer:
Using the techniques of traditional storytelling — narrative flow, context, structure — to guide audiences through data and provide flexibility to find insights relevant to them.
Let’s not write people off just yet
The automated data storytelling argument also rests on a belief that there aren’t enough skilled data storytellers to go around. That’s true. But it ignores a few factors that offer a more hopeful view:
Mr. Richardson focuses on the population of data scientists and data analysts as potential data storytellers. That ignores a massive group of people who wouldn’t consider themselves data analysts, but who analyze data and communicate results all the time. These are consultants, product managers, business analysts, sales managers, marketers — information workers of all stripes. In my view, these people are even more well situated to become our future data storytellers, if enabled with the right tools , training, and data access.
There is a growing movement toward Data Literacy in organizations. People like Valerie Logan are leading the charge in teaching people how to build skills and change cultures, to “build a shared language around the use of data.”
So, even if the population of data storytellers (or data translators ) is small today, it doesn’t have to stay that way. My preference is to help people create better data stories before giving up on them in favor of machines.
Scaling the volume of data stories isn’t the answer
If data stories are good, let’s create them in volume.
As Dr. Nicoles says:
It’s hard for people to write good data stories, and it’s impossible for them to do it at scale. AI is necessary.
We need "scaling" of machine-generated content in the way we need automated tweets.
I don’t buy that more volume is the answer. We’ve already been down this road with data reporting and dashboards. I can’t talk to a large organization without them bemoaning the volume of reports that are being created. It is not uncommon to hear about a big organization having thousands of reports and dashboards, the vast majority of which no one is using.
If we start calling them “data stories” and automatically generated them with AI, have we really solved the problem — or created more information for people to have to sift through.
Machines lack the empathy that is at the core of storytelling
There are a few starting points that I consider critical for designing data stories :
Understanding and describing the context;
Knowing your audience;
Clear defining the purpose and message of your story;
Guiding the narrative from the initial problem/conflict to resolution and action.
No offense, Computers, these are not your strengths.
Context is often not in the data…it is in the real world. For example, what is your boss stressing about this week?
You need to know what actions your audience can take and what insights will help them take those actions. Not in the data.
Combining data with these types of human understanding is what leads to powerful data stories. Ignore these skills and you are repackaging the most rote reporting, now expressed in words instead of tables and charts.
Dr. Nichols even acknowledges the challenges of good data storytelling:
It’s a lengthy process, and it requires analytical skills, domain knowledge, and the ability to communicate effectively. Because of this, great data storytellers are hard to find.
Somehow we are going to take a human activity that requires a unique skill set combining empathy, creativity, and analytical understanding...and replace that with an algorithm? Automation doesn't work that way -- it replaces the simple, repetitive tasks first until it finds where humans are more skilled than machines.
We need more humanity in analytics, not less
Good data stories aren't just about communicating a bunch of numbers — they are about empathetic audience engagement. Encouraging change based on data takes more than the right visual or a well-crafted sentence.
More is not better. Not when we are all over-tapped with information. We need better storytelling rather than more storytelling.
For all the power of AI and ML, we are also in a golden age of expression. Podcasts have given people a platform to tell their stories. TV is in a golden age of storytelling. It is easier than ever for creative people to express themselves. We can do the same with data storytelling. We can make the presentation of data a medium that is accessible to more people and delivered in more compelling ways.
Avoiding the same old mistakes
I don’t fault Narrative Science for advocating for their product, or Gartner for defining a future vision that embraces technology. In fact, I strongly believe there are opportunities for AI to augment data storytelling activities by identifying outliers and learning the storytelling patterns that work.
But after 15 years in the analytics space, I don’t want to see us continue to make the same mistakes again. Analytics’s challenges with adoption have less to do with more advanced technology and more to do with undervaluing people.
We need to consider both the people who need to absorb data to make decisions and those who combine their knowledge with data to find insights. The concept of “automated data stories” feels dismissive to both.
I may find myself relegated to a community of "slow analytics" practitioners who (over-)value human input, thoughtful design, and creatively expressing insights. But I think the people who receive the data will appreciate that it came with some human thought.