Author Spotlight

Making Good Decisions: An Art, a Science, or a Bit of Both?

In the Author Spotlight series, TDS Editors chat with members of our community about their career path in Data Science, their writing, and their sources of inspiration. Today, we’re thrilled to share our conversation with Cassie Kozyrkov.

As Chief Decision Scientist at Google Cloud, Cassie advises leadership teams on decision process, AI strategy, and building data-driven organizations. She is the innovator behind bringing the practice of Decision Intelligence to Google, personally training over 20,000 Googlers. Prior to joining Google, Cassie worked as a data scientist and consultant. She holds degrees in mathematical statistics, economics, psychology, and neuroscience. You can explore her popular machine learning video course here.

First things first: what does a Chief Decision Scientist do – and how does this role differ from, say, that of a Chief Information Officer or Head of Data Science?

Good decision-making is the act of turning information into better action. What we’re seeing with the rise of data science is that the information available in our modern world is increasingly complex and being captured, stored, and processed in electronic form. Data, in other words, is becoming a game-changing opportunity for improved decision-making. I’d go so far as to say that the 20th-century model of educating decision experts in qualitative decision skills without building their data science skills is unlikely to stand the test of time.

While it seems straightforward to pitch data science as a modern must-have for effective decision-making, there’s a need that’s just as vital in the opposite direction: skilled decision-making is a must-have for effective data science. The more complex the information, the more technical prowess is required for deciding how to decide. To rephrase my opening sentence: good data-driven decision-making is the act of turning data into better action. Who decides what "better" should mean? Which actions are on the table, and why? Which information would be relevant to the decision—and why are we weighing this particular set of decisions and not another? What might be worth doing in the first place and what information would it take to change our minds? These questions are prerequisites for successful data science, yet they’re the province of a separate role entirely. Too often, this role is subtly missing, resulting in what statisticians would call a Type III error: using all the right math and technology to solve the wrong problem.

Of the three reactions an organization might have to this problem, two are, in my opinion, very brittle strategies. One approach – which is, sadly, the traditional way – is to ignore the problem entirely and hope that someone on the team stumbles upon their inner decision expert. Unfortunately, expertise takes training, experience, and intentional practice. There’s a very real possibility of mismanaging your investment in data science as a result and damaging your business or the careers of the data professionals you’ve hired. Another approach is to require every stakeholder with decision authority to simultaneously be an expert decision-maker with years of training. This is a tall order, making it another brittle organizational strategy.

The third option is collaborating with a decision scientist. This involves taking a sober inventory of the decision skills represented on the team and, if more expertise is needed, bringing in the decision science function to work with the decision-maker, supplementing them instead of replacing them. Decision scientists help ensure that data science is effective and applied to the most valuable problems, and they’re also able to assist and advise on decision-making that doesn’t involve data. As individual contributors, they improve the quality of their teammates’ decisions and data projects. As leaders, their role is to pull up their organization’s level of decision excellence.

Before Google, you worked for many years as a consultant. How did the transition to working at a tech giant change the way you think about data science as a discipline?

When I worked as a data science consultant, most of my clients had minimal in-house data science expertise. I can’t think of a team I helped back then with more than five data professionals on it, so any theories I might have formed about how data science organizations might function with different arrangements of specialized personnel were just that: theories. If I’d had to build a large organization from scratch after that experience, I’d have been making it up entirely as I stumbled along, the way that many of my peers were forced to in the early days of data science. And even then, I would only have seen one strategy’s outcome with no counterfactuals.

I could wax lyrical for hours about all the ways in which joining Google was an incredible learning experience for me, so I’ll just focus on this one: the sheer variety of data projects and teams that I had the opportunity to observe, and eventually advise, was transformative. It gave me a much better grasp of what works and what doesn’t, and it taught me to zoom out and see the bigger picture for data science as a discipline. But, most importantly, it gave me a new perspective on what’s possible for our field as we learn to collaborate more effectively across roles and domains.

Many entry-level data professionals wonder whether they should launch their career as freelancers, join a small startup, or aim for a FAANG-size company. What kinds of questions should they ask themselves to reach the right decision (for them)?

I’d suggest that they start by asking themselves why they might want to be data professionals and what they’re hoping to achieve in their careers to ensure that they’re heading into work they’ll enjoy and find fulfilling.

I would also encourage them to think about whether they enjoy working as specialists or generalists, and which of the data professions appeals most to their personality. Speaking of TDS and choosing a data science specialization, I’d point entry-level folks to this article I wrote for you a few years ago. Each of the specializations faces different challenges at different scales of organization. For example, working as a pure statistician in a smaller organization typically puts a higher burden of self-advocacy on you than doing so in a larger company.

I’d also encourage them to consider a few other things when weighing the pros and cons of each type of organization:

Infrastructure. **** The larger the company, the more likely it is that making your tools work is someone else’s job (so you’ll spend less time doing battle with infrastructure, and more time working with actual data). If you’re a freelancer, you’ll be dealing with your clients’ infrastructure (or lack thereof), so if you opt to work with many different small-to-medium clients, _you_r battles with infrastructure could grow into the stuff of sagas.
Quality of decision-maker. Your ability to have impact as a data science professional is vulnerable to the quality of the decision-maker(s) you serve. The great thing about larger organizations is variety: if you’re allowed to move between projects and switch teams, you’ll have more opportunities to gravitate towards decision-makers who create the right conditions for you to do your best work. Freelancers, likewise, learn to look for clients who bring the right skills to the table for an effective collaboration. A startup setting is more risky; life is great if you’ve signed on to be the data expert for a brilliant decision-maker, but if there’s no one to work with except your hiring manager and it turns out you misjudged this person’s ability to make good use of the insights and models you create for them, it’ll be tricky to make a graceful pivot without quitting your job. On the other hand, when there are fewer qualified people around, it’s often easier to move into a role where you’re the one in charge of the decision. If that appeals to you, the startup setting might be a good fit for you despite the risk.
Learning opportunities.You’re most likely to have help on the job in a large organization, so you might grow your skills more effectively there, whereas you’ll have to figure everything out on your own in a startup. A word of warning about freelancing: while full-time employers are often happy to invest in you learning on the job because they’re expecting you to stick around, as a freelancer you might find your clients insisting that you do your learning on your own time.
Access to data. As a freelancer, you’ll have the opportunity to filter your clients and reject gigs where there’s no data and/or no data engineering (or you’ll voluntarily take on some paid data-design work, which is important and valuable), whereas if you’ve signed on to be a full-time employee in an organization that is having trouble acquiring data, you’re in danger of having no means of creating impact while still being asked, "What have you done for us lately?" That’s a tough career spot to be in, so make sure you check that there’s data for you to apply your data science to before you begin. If you’re joining a startup as an early employee, there’s a very real possibility that instead of being asked to sit on your hands while your colleagues scrounge for data, you’ll be asked to take care of every step in a data project’s lifecycle, including data design, collection, data engineering, and even negotiating with vendors. Some people find the challenge energizing and others run for the hills, so it’s up to you to determine what’s best for you.

You’ve recently written several articles that dispel common myths about the work of data analysts. Where do you think all these misconceptions come from – and why is it important for you to dispel them?

As the title of my first article on analytics suggests, I do believe that analysts are data science’s most misunderstood heroes. Their work is extremely valuable, yet they’re often told to learn statistics or machine learning if they want to progress in their careers. This is a tragic waste of potential – it’s painful for the individual analyst and it leads to worse outcomes for organizations.

I suspect that part of the reason that analysts are undervalued is that analytics is like the writing profession: the basics are easy to get started with and it’s an art, so there are few barriers to entry and anyone can call themselves a "writer." There’s no guarantee of quality in the title. However, being barely-literate hardly makes you Toni Morrison or Gabriel García Márquez – the best writers are light-years away from beginners and they change the world.

That’s how analytics is, too. The variance in the profession is massive. On the other hand, barriers to entry for, say, statistics, are higher, so the minimum-level folks are more impressive than minimum-level analysts, but there’s also a narrower range of virtuosity. Analysts at the top of their game are some of the most impressive technical professionals out there, but unfortunately too many employers lump all analysts into the same bucket and, by under-appreciating them, miss out on the heroes best placed to help them identify threats and opportunities while they’ve still got time to chart a good course through them.

Beyond this recent series, you’ve been a prolific public speaker and author for years, including here at TDS. Does that part of your work relate to—or feed back into—your official role?

Many of my articles are born out of compelling questions I’ve been asked by students, audience members, coworkers, and friends. Sometimes those questions rattle around in my skull until I hit upon just the right analogy, and then I find myself bursting out of bed like a Jack-in-the-Box at midnight to write it down before I forget it. Those notes grew over time and the act of writing became more enjoyable the more I did it, so I feel like many of my articles wrote themselves.

Plus, it does help that every time someone tells me they learned something useful from me, my step gets a bit more bouncy. And, between you and me, many of those questions keep popping up over and over, so I’m pretty sure I save time in the long run by being able to send a link with my answer instead of launching into a monologue every time.

To learn more about Cassie’s work and stay up-to-date with her newest articles and videos, follow her here on Medium, on Twitter, and on YouTube. To explore some of her numerous excellent contributions on TDS, here’s a small sample of highlights: