Well, Scale This: Increasing Diversity in the Tech Industry

There is more that we can be doing to increase diversity in tech teams in the US and in Europe. I took the last couple of weeks to think about pragmatic ways in which we can build healthy and diverse organizations. Ideas that are based on empathy toward another human being and not on fulfilling political quotas. Initiatives that can bring people together and have a long-lasting effect. I am convinced that all of them are within the reach of our corporate reality but require focus and attention to achieve; and perhaps more importantly, a core belief that this is the right thing to do.

The key areas of improvement include sourcing candidates, hiring practices, retaining and promoting minority and female staff.

We have a diversity problem in IT

Buzzfeed

Can diversity be achieved without Photoshop? Source: Buzzfeed

As it stands today, the IT workforce is predominantly male, white, and heterosexual. Women make up about 25% of the tech teams and often report feeling stalled at work and being presented with few creative roles or top management opportunities. It is reported that BAMEs (blacks, Asians, and minority ethnicities) constitute 40 percent of London’s population, however none sit in the C-level suites of major tech companies. The attrition rates of those groups are higher than their white colleagues: according to Google’s 2019 report on their workforce, Black men in tech roles were 36 percent more likely to leave their roles than the average Google employee. Similar findings about female attrition levels are shared in the NCWIT report on Women in Tech.

Underrepresentation is not a hiring “pipeline” problem, and it is unfair to dismiss it as such. According to American Community Survey data, “among young computer science and engineering graduates with bachelor’s or advanced degrees, 57 percent are white, 26 percent are Asian, 8 percent are Hispanic and 6 percent are black”, yet those numbers are not reflected in the actual hiring statistics. The employers favour white and Asian-American males over other groups. Vox’s Recode article Why women in tech are being Photoshopped in instead of hired pulls together some eye opening analyses on the subject, including this quote by Deborah Singer, chief marketing officer at Girls Who Code:

“We keep hearing that it’s a pipeline problem, and it’s clear to anyone paying attention that’s just not true […] Girls Who Code has 30,000 college-aged alumni. Stanford’s computer science grads are at least 30 percent female, but try to find a tech company with 30 percent women in their technical roles.”

The lack of diversity in tech is a systemic and cultural issue. I propose that we approach as if it was any other business objective.

Creating a Diverse Pipeline of Applicants

If you get predominantly white male applicants to your team, here are some things you can try to change the status quo:

#1 Candidate Sourcing. Review your pipeline and widen your outreach

It’s vital to review the hiring software you use and the way you source your potential candidates.

To deal with the volume of job posts and facilitate matching employers with candidates, Linkedin, Monster, Indeed, and other dominant job boards strongly rely on algorithms. These determine your posts’ audience, based on their undisclosed logic. In 2018, Amazon has shut down their in-house recruiting tool after it has displayed egregious discrimination for female candidates. As reported by Reuters, "the technology favoured candidates who described themselves using verbs more commonly found on male engineers’ resumes, such as >>executed<< and >>captured<<”. Beside the documented male-bias, some engineers went as far as calling some of the system’s recommendations random. Amazon’s story provides a rare glimpse into the hurdles associated with automating recruitment. Neither Linkedin nor any other market-leading job board has been transparent about their candidate scoring logic. The Silicon Valley giants pledge to be non-discriminatory. Cathy O'Neill, a Data Science consultant and author of Weapons of Math Destruction, questions this stance by suggesting that LinkedIn should be “forced to prove that what they’re doing isn’t exacerbating inequality” in her interview with Business Insider. She points out that even when race or gender is excluded from the data feeding the algorithms, it is virtually impossible to get rid of proxies that standalone or coupled together can disclose a candidate’s personal background.

Another powerful pipeline influencer are ATS (Applicant Tracking Systems). Captera’s report has found that “75% of recruiters and talent managers use some form of recruiting or applicant tracking software.” The technology is used to source candidates from company-internal and external databases (job boards, aggregated data from the brokers), filter the applicants to match the job’s criteria, and track their progress throughout the hiring cycle. James Hu, the CEO of Jobscan, notes various shortcomings of ATS in his conversation with The Register: restrictive filtering (looking for “analytics” won’t return “analysis”), hard-coded search values, problems with parsing various file formats, document formatting, and special characters.

Awareness of those issues is a good start to find a remedy for the unbalanced candidate pipeline. Some pragmatic ways to battle the black-box hegemony:

  • Advertise on various platforms, don’t focus on one major job board to avoid biased post exposure and pre-filtered results.

  • Use non-conventional job boards that prioritize transparency to widen your reach.

  • Engage with local tech communities via meetups.

  • Get in touch with organizations promoting women and minorities in tech and advertise through their networks. Check out Girls Who Code, Geek Girls Carrots, organizations that support Black coders, for a start. Do your research and apply care and curiosity to learn, and more of those groups will surface.

  • If hiring graduates, look beyond top universities that rely on high tuitions. Some great hires can come from non-privileged groups that cannot afford tuition at the top high-ed institutions.

  • If using ATS software, evaluate its search and filtering capabilities.

  • Do not over-rely on referrals. While happy employees make the best recruiters, the inadvertent side effect of using the social network of a non-diverse staff is homogeneity.

#2 Make your language inclusive

Wording of a job offer can make all the difference. Compelling evidence exists to suggest that gendered wording in job ads is common and can lead to gender inequality. Words associated with feminine include “support”, “understand”, “interpersonal” whereas male-domain rely on “leader”, “competitive”, and “dominant”. Using gender-neutral language can increase the talent pool: available research suggests that women will find a position less appealing if they infer it to be male-dominated, and vice versa. Curiously, the deciding factor is not the perceived skill, but the projected sense of belonging.

Inclusive language does not translate to including a diversity statement at the bottom of the job description, either. Simply advertising your company as an equal employer does little to increase organizational diversity. Worse, it can badly backfire. A study published in 2013 in Journal of Personality and Social Psychology reported that diversity statements can contribute to creating an illusion of fairness. “An illusion of fairness occurs when high-status group members believe that the mere presence of diversity programs makes organizations fairer for underrepresented group members, even in the face of evidence that low-status groups are unfairly disadvantaged.” This is in line with what is reported as licensing bias: a paradox in which prior positive choices (e.g. anti-racist proclamations) of an individual may entitle them to self-indulgent choices going forward, on the basis that they now hold an irrevocable license that proves the opposite, in line with their self-image.

Harvard Business Review points to another unintended consequence of pro-diversity statements: encouraging “job applicants to let their guard down and disclose more racial information.” A grim product of our times and typical technique that minority groups use is “whitening” their resumes, i.e. removing any race-related information. The HBR’s study has concluded that “the discrimination against unwhitened resumes was no smaller for purportedly pro-diversity employers than for employers that didn’t mention diversity in their job ad”. Diversity statements, without a solid backing from the company’s culture, produce false hopes for the applicants.

We also know, from the Hewlett-Packard internal study famously quoted in Sheryl Sandberg’s Lean In, that men have a tendency to overvalue their skills, while women tend to put their abilities under more scrutiny. The report has famously found that “women working at HP applied for a promotion only when they believed they met 100 percent of the qualifications listed for the job. Men were happy to apply when they thought they could meet 60 percent of the job requirements.” Two people, with the same observable skills, will have a different perceived experience of the same job offer.

How could this be remediated?

  • Do not rely exclusively on the diversity statement. Verify if you can really deliver an unbiased process before advertising it.

  • Review your job postings from the linguistic perspective. Substitute words rooted in the masculine culture: competitive, ninja, power, expert with more neutral-sounding that prioritize collaboration and learning. I once saw a job ad I really liked and immediately felt inclined to apply for. The company proclaimed that they were not looking for an ideal candidate. They understood that every applicant will have a different background and experience. They did not expect the hire to get the ball rolling from day 1. Instead, they anticipated a learning curve that they were committed to support. Be like this company.

  • Make sure your recruiters/HR staff have a training in psychology and cognitive science to understand cultural and gender-based differences in behavior and perception.

#3 Encourage remote or flexible work

Flexible work benefits everyone. Consider this: if your HQ is based in central-London, who is most likely to apply? People that can afford to live in a reasonable distance to the office or do not mind spending hours getting to work. Unintentionally, you might be discouraging applicants that cannot afford the same financial or time resources. As women tend to take more responsibility in child rearing, they cannot afford long commute. Minority groups often come from poorer backgrounds and might find relocation to central London out of reach. Switch the policy to remote-work-friendly and you will unlock talent pool that is unrestrained by geography, and therefore free of socioeconomic entry barriers. What is more, remote workforce is inclusive towards people with physical disabilities or mental health issues that are otherwise highly skilled.

Telstra, an Australian telco corporation, has been renowned for their flexible work policy and their commitment that jobs “flex” unless proven otherwise. The policy has considerably raised the number of female applicants, and about 1/3 of all that applied has singled out the flex policy as the defining factor.

Claudia Goldin, a professor of economy and in fact the first woman to achieve tenure in Harvard’s economics department, singles out work flexibility as the critical step towards achieving labor-market equality. In her Grand Gender Convergence, she writes that the shift “must involve […] how jobs are structured and renumerated to enhance temporal flexibility. The gender gap in pay would be considerably reduced and might vanish altogether if firms did not have an incentive to disproportionately reward individuals who labored long hours and worked particular hours.” Women, especially with kids, tend to seek out temporal flexibility: jobs with flexible schedules or remote-work options. Yet, the market tends to penalize these job arrangements. Instead, long-hours office jobs are the standard, even if shorter workday is proven to be more effective, and long hours are said to sabotage productivity. For Goldin, closing the wage gap can only succeed if employers change their pay structures.

Action points:

  • Revise your policy on working from home and remote work. Audit which roles are inadvertently tied to the office, and which can be successfully performed from home. Challenge the status quo with giving assignments to remote-based contractors. Strict insistence on a traditional work arrangement limits your applicants pool.

Hiring Candidates

#1 Accountability over quotas

Assigning a person or a team accountable for the proportion of minority hires is a key strategy that companies can adopt to increase organizational diversity. This has been one of the critical findings of a 2006 study, Best Practices or Best Guesses? Assessing the Efficacy of Corporate Affirmative Action and Diversity Policies (I am referring to the other insights later in the article). Responsibility over building diverse workforce can be given to HR, recruiters, or diversity managers. A word of caution: accountability does not equal setting minority or gender quotas.

Enforcing parities, without a thought-through process can have a counter-effect on organization’s health. Having a minority representative join an otherwise homogenous group is known as a tokenism. Tokenism is commonly understood as an insincere effort that aims to create an impression of equality and cover up imbalance rather than push systemic change. However, even the best intentions of the hiring panel might result in conjuring the effect. If you add a single black woman to a male-dominated team, group psychology will work against everybody involved. The token minority will likely have a hard time fitting in, and will struggle with social isolation. The person might even repress some of their characteristics (character, beliefs, looks) in order to fit in. In parallel, the team can build a perception that the hiring process was unfair. Kecia M. Thomas files this under “concern over unfair treatment amidst pro-diversity values”, in her book Diversity Resistance in organizations. Intel’s CEO Brian Krzanich has bluntly called such resistance a “white men under siege” effect while referring to Intel’s diversity initiatives that were met with backlash within the company, including threats towards the senior management. MIT Encyclopedia of Cognitive Sciences lists another side effect: “people also develop expectations about others substantially on the basis of their group membership and the associated stereotypes […] Because stereotypes shape interpretations, influence how information is recalled, and guide expectations and inferences in systemic ways, they tend to be self-perpetuating.” The unlucky token minority will be judged based by their group stereotype rather than their individual qualities.

Iris Bohnet, a behavioral economist, proposes ingenious ways to avoid the token minority pitfalls in her book What Works: Gender Equality by Design. Two ideas stood out to me in particular: the “critical mass” effect and postponing but not forgoing application of quotas in the hiring process. Bohnet’s first technique is an antidote to the group attribution bias. To trick our brains out of unfair judgment, teams should always include multiple representatives of any particular group. This will allow each of them to be viewed as an individual, and not through the lens of a stereotype. Bohnet argues that a “critical mass” is achieved when minority participants constitute no less than 1/3 of a team. What this translates to is that in a microscopic scenario of 2 teams and 10 people, it’s better to assemble one team of 5 men, and another of 3 men and 2 women, rather than allow a 1 women to 4 men ratio per group. Bohnet recommends rethinking how quotas are leveraged in hiring. Recruitment should be merit-based, she says; once promising candidates are selected, quotas can be applied to favour backgrounds underrepresented in the company. The technique both protects the new joiner and stabilizes the organization: the employee is hired based on merit, not their ethnicity, gender, or nationality, and any accusation of forcing “political correctness” on the company can be rebuffed.

How to apply quotas in a smart way:

  • Assign accountability over diverse hiring to a person or a team in the organization.

  • Aim towards achieving a “critical mass”: avoid hiring tokens and assigning them to otherwise homogenous groups. This will alienate new hires, and eventually make them leave the company. Create diverse teams by including more than a single representative of any group.

  • Introduce minority and gender quotas in the later stages of hiring: base the preliminary selection on merit.

  • Hire in bulk. Diversifying a group is always easier to achieve than making a decision about an individual hire.

#2 Improve the interview process

We are all inherently biased. The only way to eliminate bias, is to dispose of any social or visual cues that might induce it.

The most famous example of such “reductionist” hiring are blind auditions. Orchestras all around the world notably balanced the gender ratio by applying the technique. I first heard about the phenomena in Malcolm Gladwell’s Blink, a book on the adaptive unconscious. The original study was authored by Claudia Goldin and Cecilia Rouse, and published by American Economic Review in September 2000. In an attempt to overcome gender-bias in hiring, many orchestras across the globe have revised their hiring practices. They introduced a practice of “blind” auditions, the purpose of which is to obscure the identity of the performing candidate. Two prompts are typically used: a screen to block even the slightest hint of the person’s silhouette and a sound-absorbing carpet so that male and female footsteps cannot be differentiated. Introduction of the blind auditions has had an extraordinary effect on the musical world. Goldin and Rouse have demonstrated that the percentage of women across top philharmonies has risen from less than 10% in 1970, to 35-50% just a decade later.

Other industries have been taking notice. Space Telescope Science Institute, responsible for allocating time on NASA’s Hubble telescope has pioneered a double-anonymous review process to “level the playing field for women and other marginalized groups in science.” A study on code contributions on Github, known as pull requests, has shown that women’s requests are just as more likely to be accepted than men’s, only if - plot twist - their identity is not known to the public. The authors of the study ended it with hope that “[their] results will help the community to acknowledge that biases are widespread, to reevaluate the claim that open source is a pure meritocracy, and to recognize that bias makes a practical impact on the practice of software development.”

While many selection decision aids exist that can be applied to the hiring process, there is a notable resistance toward using them. In his summary on the topic, Stubborn Reliance on Intuition and Subjectivity in Employee Selection, Scott Highhouse reports that people tend to find traditional, gut-feeling-based, unstructured interviews more effective than any assessment frameworks available, contrary to the overwhelming evidence that the opposite is true. Additionally, there is a prevailing belief that it is possible to predict, with high levels of confidence, a person’s performance on the job by a combination of intuition and experience. As shown by a collective study on 85 years of research in personnel selection, such near-perfect precision is a myth. The best techniques go as far as predicting the performance and job-related learning with 50% confidence. Some of the more notable procedures that score well are General Mental Ability tests, work sample tests, job knowledge tests, job tryouts, peer ratings, and structured employment interviews. The lowest predictors are age, graphology, interests, years of education, years of job experience, and reference checks.

There is enough compelling evidence that meritocracy hiring in tech is a myth, and that biases are prevalent. Here are some ideas of how to battle the status quo. Applying even a single of those techniques will have an impact. Blind auditions have demonstrated that using a screen only in the preliminary round can have a powerful impact; this step makes women advance to the finals 50% more likely.

  • If using unstructured interviews, stop. Unstructured interviews are likely to over-rely on the interviewer’s intuition, which is a risky combination of biases, personal background, and subjective preferences. Take advantage of available decision selection aids proven to increase the likeliness of high job performance. Consider giving your candidates tasks that mimic a typical job assignment: for example, ask for writing a sample program, proposing a solution to a problem encountered at the customer site, designing system architecture for upcoming migration, making a list of all things that can be troubleshooted in an error scenario. Introduce GMA tests, like Mettl, to measure the ability to comprehend and interpret information.

  • Introduce blind hiring throughout the process (at least in the preliminary stages) to remove inherent bias common while vetting candidates. Use blind hiring software from Applied, or Fair Hire.

  • Use a structured framework. Develop a scoring matrix that can be used by different interviewers, so that the interview notes are comparable. Make holistic evaluation of the job requirements: for programmers coding skills are a must, but so is the ability of working in a team, and adaptability to customer requirements. Most likely, you do not want a 10X Engineer, or what Camille Fournier calls a “brilliant asshole”: “That person you think can’t be replaced because he’s just so productive and so smart, but who isn’t a team player and makes everyone around him unhappy.”.

  • Discard panel interviews that are easily influenced by the most dominant member of the group in favour of 1-to-1 sessions. Use quorums for decision making, do not let a single person determine the outcome of the whole process.

  • Consult an expert on hiring practices, like BAME, to develop better hiring standards for your company. Get certified by Edge, and advertise this to attract talent.

Retaining Staff

Just as important as the hiring process, is the practice of retaining staff.

#1 Create an inclusive culture

Inclusive culture is potentially the single most important factor to building a diverse workforce. It’s typical for organizations to offer various perks to retain employees. These customarily include free coffee and tea (or beer in Silicon Valley’s bro-tech companies), table football or ping pong stations, wine-tasting outings, company-sponsored Christmas dinners. Perks do not substitute culture, neither do they comfort an employee that feels isolated in their work environment. In her HBR article, Karen Brown illustrates this as: “picture, for example, a Muslim who prays in his car because he doesn’t want to advertise his religion, a mother who doesn’t put up pictures of her children so that coworkers won’t question her commitment to the job, or a gay executive who is unsure whether he can bring his partner to company functions.” Celebrating Christmas with colleagues, or participating in after-work beer outing, while both uniting in their outset, can be straining for many employees. Note how often these initiatives project a very white-Christian-centric idea of entertainment and community. People that do not conform to a dictated standard of fun can feel alienated, and tend to hide their identity in fear of upsetting the majority. Non-drinkers and Muslim employees, if they decide to join a wine-tasting session, it is not because they particularly enjoy drinking water out of wine glasses, but because they risk alienation if left out of the event.

The most striking example of how hostile a homogenous culture can become is Susan Fowler’s account of her year at Uber. When she started at the company, the organization she joined comprised in 25% of women. Within 12 months that number shrunk to 3%. Uber’s culture has been publicly denounced as aggressive and extremely toxic: with HR function reduced to not more than shuffling papers, sexual harassment, homophobic behaviors, and promoting problematic managers who were subject of numerous complaints were daily occurrences.

Inclusivity cannot be trained: it has to become embedded in the organization’s structure. Studies show that diversity training and mentorship programs are ineffective in increasing organizational diversity, as they tend to focus on changing individuals. “Practices that target managerial bias through feedback (diversity evaluations) and education (diversity training) show virtually no effect in the aggregate.” The only technique group that proliferates and long-term retains diversity are structures establishing responsibility. These are affirmative action plans, diversity committees, and diversity staff positions. A dedicated function in the organization can view the problem holistically, rethink hiring and promotion structures, propose solutions at different levels of human resource management, monitor the effectiveness of those solutions, and most importantly, has decisive authority. It’s as simple as setting goals and assigning responsibility for moving toward these goals.

Building an inclusive company culture can be achieved by:

  • Appointing special staff members or committees to build diversity structures in the organization. Accountability is more effective than training individuals on how to overcome their biases.

  • Cultural awareness around perks and company events. Allow all employees to (anonymously) submit their ideas for policies and initiatives they would like to see in the company. Organize corporate outings in environments that are inclusive to all employees: steak house might be a bad idea. There is nothing bad about holding football pools around the office as long as alternative initiatives are present throughout the year.

  • Monitor minority retainment in the organization. Best hiring procedures might fall flat if the company culture is not ready to accommodate for diverse participants’ wellbeing. Monitoring employee retainment is one of the most telling indicators about the health of company culture.

Summary

This article has been inspired by my observations working as an Eastern-European female in the tech industry. I have often heard the “pipeline argument” and insistence on meritocracy. Then I saw my referrals or my own applications being turned down on the basis of a single hiring manager’s opinion. Those experiences made me question the validity of some common hiring practices in IT, and start to sense mythology built around them. A gut feeling that there is an invisible bias at play was not enough, so I started educating myself. This post is the effect of this work: a collection of typical pitfalls in hiring matched with actions that can be applied to improve our workplaces and make them more diverse. Bullet points against bias. I hope that this research will be useful to those, who like me, sense that working structures are out of balance in tech, and want to find practical counteractions to change them. We can improve our industry starting today.

I am still educating myself on the subject and I am open to feedback on how this can be further elaborated.