A number is better than a slide. A slide is better than a deck. And a deck is better than a paper.
After over 10 years of working with data, first as a biostatistician, and then as a data scientist in politics, advertising, tech, emojis, and now finance, this is the crux of what I’ve learned.
And it might seem counter-intuitive, wild almost, compared to the constant fetishization of “artificial intelligence” and “machine learning” and “big data” in the media, in tech industry PR, and unfortunately, in academic curricula as well.
But the reality is that the goal of most of data science, and certainly the part of data science that excites me the most – data insights and data storytelling – is to drive action. And the best way to persuade human beings who are bombarded with massive amounts of information all the time is to keep it simple.
Tony Robbins’ teaching about the three most important decisions we make every moment of our lives is surprisingly relevant at the intersection of data science and storytelling as well.
- What should I focus on?
- What is the meaning I’m going to give this?
- What am I going to do now?
I accidentally discovered this only recently while reflecting back on early moments in my career. One of the first projects I worked on at Facebook was to understand how ad agencies were buying Facebook ads.
Ad agencies are giant, complex, multinational organizations, and serve as the intermediaries between most advertisers (i.e. firms that engage in advertising, such as Coca-Cola or Toyota) and publishers (e.g. Google or Facebook or the New York Times). A large percent of all of advertising spend goes through ad agencies.
Facebook at the time had a robust sales organization and a growing measurement team, both of which were mostly focused on end advertisers (e.g. Coca-Cola or Toyota) and helping them run more effective ads. When I joined, the agency relationship team was smaller in comparison and the company was just beginning to think more strategically about how to engage with agencies.
My first project was to understand where in our vast data stores our agency relationship mapping data was located (so we knew, for example, that Agency A managed Coca-Cola’s spend on Facebook in the US, but Agency B managed Coca-Cola’s spend in France, while Agency C managed it in India).
Then, when I found the mapping data, I started thinking about all the ways we could break out how an agency was advertising on Facebook. What were its top clients, how much was it spending on mobile vs. desktop, what types of bidding strategies was it using, what types of audiences was it targeting, how were its ads performing, which audiences were its ads performing best with, what types of creative was it using, so on, and so forth.
And then I spent weeks searching through our databases, finding where each data source was stored and verifying its accuracy, comparing it against other known valid data sources, and often pinging the table owner and asking them which of four similarly labeled columns to use, often to be told that the whole table was outdated and there was an alternate table I should use. (I keep telling students that so much of data science is about humans so much more than it is about computers!)
I then wrote a data pipeline to process all of this data, stored in trillions of rows, using hundreds of lines of SQL code, and then pipe a smaller dataset into R to automatically generate a 30-slide presentation for each of our agency clients. We would use this report in our Quarterly Business Reviews (QBRs) with their senior most leadership with the goal of impressing them with our “360 degree view” into how they were buying ads on Facebook – and convincing them to embark on a learning agenda with us so they could run better ads.
When I had the entire process finalized and automated, I showed the 30-slide presentation to my boss. I was so excited because I had taken the project from beginning to end and I was about to see what he thought of it. I walked him through the data discovery, the data cleaning, and the data processing. He kept nodding along. I showed him beautiful heatmaps for which I had spent multiple days choosing the perfect color palette. He kept nodding along. And then, on slide 26, he noticed one number on the screen.
Domino’s Pizza was spending 30% of their monthly Facebook spend running ads on FBX (Facebook Exchange). FBX, which was finally phased out in 2016, was outdated even in 2014 and was a legacy way to buy ads programmatically on desktop. It was *not* the best practice and he was shocked that Domino’s was spending so much money (millions of dollars) on an outdated ads technology that was not as effective as newer technologies that were replacing it. He was also excited because this was a tremendous insight he could walk into the next agency QBR with. We didn’t even finish the rest of the slides – that’s how excited he was.
I presented the same deck to the agency development team – and they had the same reaction. “Domino’s is spending 30% on FBX?!?!” For months, that number became my calling card at Facebook. People knew me as the data scientist who had discovered that number – and with it, improved Facebook’s thought leadership credentials with that agency, which now recognized that we could effectively discover insights hiding in plain sight and use them to help their business.
Now, the irony is that I rolled my eyes a bit while this was all happening. I personally didn’t care that much about whether Domino’s was using FBX or not. I thought it was silly that everyone was super excited about one number I had found, and didn’t appreciate the *process* I had used to find that number. Data science for me, was, and is, an art more than anything else. It’s one of the truest expressions of my creativity. Finding meaning where most others see only trillions of rows of messy data? SIGN ME UP.
And here were all my colleagues who didn’t appreciate at an intellectual level what I had built, how many sleepless hours I had spent assembling the perfect dataset and visualizing it in the perfect manner.
But now, with the benefit of many more years of experience and maturity, I realize that being able to reduce a complicated technological data-driven discovery to one number is a feature, not a bug. Although my colleagues may not have known they were following Tony Robbins’ roadmap, they were following it to the letter.
What should you focus on? Domino’s is spending 30% of their ad spend on FBX, an outdated and legacy technology, when they could be spending it much more effectively.
What does this mean? Facebook knows a lot more about your advertising, across all your clients, thank you do and can give you a proper 360 view.
What should you do now? You should put your trust in Facebook and partner with us to discover even more insights (and spend more money on our ads 🙈)
And as my career progressed, the same pattern would repeat itself, again and again and again. I would produce some incredible, beautiful work of data science, putting in weeks of sweat, blood, and tears, and at a presentation, someone would find one number that stood out to them more than anything else, and that number would become the defining aspect of that work.
People won’t remember 30 slides. People will definitely not remember 30 pages. People will remember one number.
Your task as a data scientist and as a data storyteller is to find that one number, that represents the data truthfully and accurately, and that is also most relevant to and will have the most impact on the product or the business (or the cause.)
The following year at Facebook, I undertook a similar project, this time looking at how people were engaging with video ads, which had just been rolled out. Just like before, I meticulously collected all the source data, built a pipeline to process it properly and visualize the results. Again, I presented all these intricate dashboards and visualizations to various stakeholders. Everyone noticed one number right away. Most of the drop off in video ad viewership happened by 3 seconds. (In the early days, 92% of viewers had stopped watching by second 3.)
Our creative shop, our in house creative studio, immediately grabbed that number and built an entire campaign around “The 3 second audition,” teaching advertisers and creative agencies that the first 3 seconds were the most important parts of a video ad on Facebook and that they should do everything they could to grab people’s attention in that time span.
I doubt anyone today remembers any of my original code or my data pipelines or scripts. But many people still remember me as the data scientist who did the research that first documented “The 3 second audition,” which became pervasive across the advertising and creative industries in subsequent years.
Now, you may be wondering something. If in both of these examples, the critical number – 30% for Dominos Pizza, and 3 seconds for video ads – was so immediately obvious to everyone I presented my research to, why wasn’t I the first to notice it?
That’s a great question. I probably wasn’t mature enough as a data scientist. I didn’t fully appreciate the importance of making people care and how that was the only way to impact the business. I wasn’t familiar enough with the domains I was researching – bidding technologies on Facebook or video ads. I lost sight of the big picture and got lost in the weeds.
In hindsight, I see that the number is what tells people what to focus on. The title of my presentations should always have been “92% of video ad viewers drop off by 3 seconds” not “A framework for measuring video ad engagement on a modern social network” (jk that was never a title of one of my presentations, but you get the point). The next slide should have been “This means that it’s very hard to grab people’s attention when they’re scrolling in their feeds.” And the last slide should have been “Now, let’s go to advertisers and teach them how to make better video ads.” Everything else should have been in an appendix.
Earlier in my career, because I didn’t fully appreciate the big picture, I wasn’t always able to participate in the full life cycle of my work. I would find a bunch of numbers, and other people would tell me which ones were important, and they would decide what we would do with this information.
As I’ve matured as a data scientist, I’ve realized that data storytelling and using it to drive impact is the beating heart of data science. It’s what makes it relevant. It’s what makes people care.
After I left Facebook, I went deep into data journalism, thinking as a journalist and a storyteller first, and only then as a data scientist. And I used everything I had learned at Facebook to do that effectively. I led with the numbers. I told them what it meant. And I told them what they should do next.
We need to do a better job of emphasizing the importance of real world impact in data science education. Real world impact is hard – because it requires creativity, and because most of the skills involved are not technical in nature, but are human skills. Intuition. Hustle. Persistence. Domain Knowledge. (I first spoke about this at Penn State in 2016.)
I’ve had thousands of resumes from recent graduates come across my desk, all seeking to one-up each other in the latest and fanciest technologies, all having worked only on well-defined projects where someone else did the hardest work for them. Someone else defined the problem, someone else collected the data, and after they built a model, someone else decided what it meant, and what to do with it. There are data science jobs that only focus on the technical element, most of them very well-paying.
But those are not the types of jobs that reflect the true potential and promise of data science – to advance our understanding of the world around us, to help us tell stories, to help us use data science to tell people what to focus on, what it means, and what to do next.
Even though I only began to think of myself as a data storyteller a few years ago, I had been drawn by the concept since the very beginning – a fact I only discovered recently.
Here is what I found I had written in my statement of purpose for my application to graduate school in biostatistics at the University of Michigan, in 2008:
“Soon, data sets—large, unseemly, raw collections of numbers—began to tell me their stories. Whether these stories concerned demographics, human physiology, or the economy, they shed light on our understanding of the world. I saw firsthand how the language of statistics worked, and the immense power it held.”
Stories and statistics together have immense power. Keeping Tony Robbins’ three questions in mind helps us harness that power to its full potential.