Mental Model: The Talent Flywheel

Todd Schiller
4 min readFeb 13, 2020

--

In the engineering world, a flywheel is a device with a rotor that accumulates, stores, and smooths energy flow. A flywheel has a high moment of inertia, so once it’s in motion, it’s almost as if it were a perpetual motion machine.

A steam engine flywheel

In the business world, a flywheel is a cycle where a company gains an edge, leading to better outcomes, which in turn compounds that edge. In the field of Systems Theory, such a cycle is known as a virtuous cycle or reinforcing loop.

The flywheel business concept was popularized in Jim Collin’s 2001 book Good to Great. Over the past few years, it’s seen a resurgence of interest as a way to explain Artificial Intelligence and Machine Learning’s potential impact on market dynamics.

Artificial Intelligence Flywheels

An AI flywheel exists when a company gains an AI advantage, which allows it to attract more customers, which in turn allows it to improve its AI, and the cycle continues… The compounding improvement can produce an impenetrable economic moat.

For example, take Netflix. Netflix has droves of data scientists building models of what its customers want watch. It uses these models to acquire and recommend content which increases viewership. The additional viewership data in turn enables Netflix to improve their models. And the cycle continues¹…

Netflix’s Artificial Intelligence Flywheel

Talent Flywheels (aka Human Intelligence Flywheels)

The flywheel metaphor doesn’t just apply to Artificial Intelligence though — Human Intelligence flywheels have existed since the birth of the corporation in the 18th century.

By hiring better than your competitors, you get better performance, which leads to a better results and stronger employer brand, which makes it easier to out-hire your competitors, and the cycle continues…

Interestingly, while Netflix was building its AI flywheel, it was also building a talent flywheel. In 2009, they bucked recruiting norms and published their now infamous culture deck. The deck outlined a performance-oriented culture, evoking the metaphor of a company as a pro sports team rather than a family. The deck, along with the ensuing public discourse, established a strong recruiting brand. Based on this brand, a stream of performance-oriented candidates self-selected into their talent acquisition pipeline. The influx of better hires helps reinforce the culture of excellence and fueled the company’s success. And the cycle continues…

Netflix’s Talent Flywheel

Jim Collins highlights a similar talent flywheel in Turning the Flywheel. However, it’s not at a tech unicorn with inordinate salaries. Instead, this flywheel was built at an elementary school in rural Kansas. He details how the school was able to achieve outsized results by: 1) selecting the most passionate teachers, 2) developing programs to help each and every student succeed, 3) establishing a brand as a great place to learn how to teach.

Building a Flywheel

So how do you build your talent flywheel? In Good to Great, Jim Collins’ team researched companies that made and sustained the transition from good to great. They found that each had a flywheel that they established through a relentless and focused application of pressure. From the outside, there may appear to be catalytic events (e.g., Netflix publishing its culture deck). In actuality, the overall effect comes from the accumulation of daily behavior.

In future posts, we’ll be taking deep dives into each phase of the talent lifecycle (assessment, on-boarding, etc.), the relationships between them, and practical steps you can take to build your talent flywheel. Remember to follow us on Medium so you don’t miss out!

Interested in building the talent assessment component of your Talent Flywheel? Try out Interview GPS, which make it easy for any team to run a best-in-class structured hiring process

Footnotes

¹ Some people might argue that despite having a viewership data flywheel, Netflix doesn’t have a strong moat because 1) there’s quickly diminishing returns to viewership data, and 2) not much data is needed for “good enough” content acquisition/recommendation

--

--

Todd Schiller

Co-founder @ PixieBrix. University of Washington CompSci PhD. Formerly AI at Bridgewater Associates. Machine ✘ Business Intelligence