During the Vietnam War, Secretary of Defense Robert McNamara and the Pentagon required detailed field reports on every piece of data and outcome metric imaginable. Just one evaluation program produced around 90,000 pages of data per month.
The data that was collected and analyzed all pointed to one conclusive and unassailable fact: America was winning the war. This created the apocryphal story that when the data was crunched, it showed that the United States already won the war a few years prior.
Be data-informed, not data-driven
This is an extreme example of the most common mistakes companies make when trying to be “data-driven.” By misinterpreting signal and letting human bias influence your conclusions, you create an inaccurate worldview and add risk to any decisions made from that view.
Being data-driven can lead your organization astray because data itself is not information. It’s only through interaction with humans, who are fallible, that data becomes usable information. Organizations that strive to be data-driven run the risk of making decisions based upon flawed or biased information. Smart organizations seek to be data-informed, not data-driven.
Bias in data-driven hiring
A prime example of this is the recent rise of keyword analysis for hiring. There are multiple platforms and companies with the promise that, by examining large data sets of specific keywords or skills, they can deliver you the perfect candidate for the job.
In a data-driven world, this sounds perfect. You enter criteria and the machine makes the match. However, anyone who has ever made hiring decisions knows, what’s on a resume doesn’t always correlate with on-the-job performance. Maybe this person had inside information and knew what skills to present. Maybe the criteria you selected doesn’t match up with the outcomes you want or need. Or maybe, like the recent case with Amazon, the data you’re using to make decisions is inherently biased. Here, the data-driven approach only amplifies the biases of the individuals creating the models.
Given the impact of hiring decisions, both on the individual and the organization, we must adopt a better approach that anticipates the potential for bias. This approach keeps humans at the center of key decisions, instead of blindly letting data drive flawed outcomes. We call this being data-informed.
Focus on bias-free outcomes
Being data-informed recognizes the potential to introduce bias and the limitations of data-sets to answer complex questions. It provides guides and heuristics to minimize these impacts and keeps humans responsible and in control of the resulting decisions.
One way to remove bias in such decisions is to focus on outcomes that are bias free. With the resume example, we recognize that keywords aren’t bias free because they are subjective interpretations or proxies for performance.
For example, the phrase “detail oriented” can mean different things to people in different contexts. However, in a software development organization, being “detailed oriented” should result in fewer software bugs introduced into a production environment.
Instead of looking for the phrase “detailed oriented” in a software developer’s resume, we should look to objective characteristics or traits that result in fewer software bugs being released into production. The important step is to relate that outcome data back into an initial hiring assessment, instead of subjective proxies that introduce bias or noise. This results in better, less-biased, more data-informed hiring practices.
It’s important to recognize that data isn’t good or bad, evil or trust worthy. Data is neutral. You only introduce the opportunity for bias, mischief, misrepresentation, misunderstanding or malfeasance when humans interact with data. Being data-informed helps organization take responsibility for their data and curate decisions that stay focused on results. Without this level of data-informed responsibility, you risk being told you’re winning, when you already lost a few years ago.
– Tom Iler, chief product officer