Breaking the Bias: Let’s start with data, analytics, AI and the role of HR

Data-driven bias

The world is becoming more data-driven, though so far, data hasn’t exactly been friendly to women, as highlighted by some of the issues in the book “Invisible Women” by Caroline Criado Perez. As HR becomes more data-driven, using more data, analytics, and AI, we need greater awareness of how bias creeps into everyday life and impacts decisions made by HR.

Working with data, analytics, and AI in HR isn’t easy. Data, analytics, and data are being used in countless use cases, with more applications being developed every day. Examples of bias span the spectrum from Amazon scrapping its biased hiring algorithm to Textio helping improve gender language in job descriptions. We can’t change the world, but for HR, there is a lot we can do in our organisations to help improve the lived experience of our employees.

The challenges of getting data right

We’d like to explore the challenges of data, analytics, and AI and their impact on people. HR is using more data, analytics and technology to drive evidence-based decision-making. How will the inherent biases built into our world impact women and the gender pay gap?

In “Invisible Women”, Perez walks us through the history of the hegemonic masculine culture that we find ourselves in today. The world exists in a male default mode with exceptions made by adding modifiers: women’s football, female doctor, etc. A gendered world built on generations of women (and others) being barred from taking part in many of the critical elements of society like property ownership, access to education, the right to vote, and even protection from domestic violence impacts the data we have today.

Data, analytics, and artificial intelligence (including machine learning) are being deployed to help HR in various ways. Research in 2018 found over 75 use cases of AI in HR across all domains. This number grows daily with new applications being developed by vendors and in-house by organisations. Concerns for bias in HR, particularly with recruitment, succession planning, and promotions, existed before using data, analytics, and AI. As we use data to automate processes and decision making, we have to be aware that the data on which we are basing our work processes may be flawed by biased data. How can HR use advanced data techniques to improve outcomes and not amplify past mistakes?

Artificial Intelligence to Break the Bias

If we look at the use of AI in HR, we can see two ends of the spectrum with AI. On one side, we have issues like Amazon’s biased hiring algorithm and a plethora of examples of ageist, racist, and other biased systems. These are being discovered not just in HR, but as Cathy O’Neil highlights in her book “Weapons of Maths Destruction” (and the Netflix film “Coded Bias”), these troublesome algorithms are being deployed in all areas of life. They affect students’ grades, who sees which job advertisements, and many other examples that we know about and ones not yet discovered in this under-regulated industry.

On the other hand, there are products like Textio, a product that helps recruiters develop gender-neutral job descriptions. The product started with improving the language in job descriptions and has grown into a unique analytical tool.

In testing the product across several industries and levels of job descriptions, it quickly becomes apparent how much gender language is related to different jobs. Executive job descriptions weighed in as heavily masculine on average, and on the flip side, administrative roles weighed in as heavily feminine. The results emphasize Perez’s argument in her book about the underlying issues women face with the inherently masculine language used for senior leadership roles.

However, changing the language of a job description is just part of the battle. Once in the talent pipeline, women face other challenges. HR can use analytics to highlight the stages that women struggle with by identifying dropout rates at different stages of the recruitment pipeline and understanding why these dropouts are happening.

These examples highlight the issues women face getting in the door; however, the problems continue even when working in leadership roles.

Challenge #1: KPIs

One of the challenges research has found is that women are still not trusted in leadership roles. And, with the wonderful world of Zoom that grew due to the pandemic, we have recordings of every sexism that women face at work, as many of us related to Jackie Weaver’s frustrating experience in running a local government meeting while being belittled. These challenges play out in the world we live in, are captured inadvertently in our data, and can easily be repeated by algorithms we deploy when using historical data and KPIs.

When talking with researcher Jack Daly, whose PhD work examines the hegemonic male default and its impact on the gender pay gap in British based firms, about the effects of data, analytical, and AI on women in the workplace, he wasn’t surprised. He said it isn’t only an issue with using past data but also using measures and KPIs designed at a different time and from a different perspective.

Daly says when looking at hiring or promotion KPI, for example, it’s not always about the results but about how people go about getting the results. For example, what’s the impact of focusing on billable hours instead of developing technical skills? What behaviours are related to achieving the different measures? For HR, there is an important lesson to learn here; if we want to see the changes we talk about, we need to think about the measures we are using and what behaviours they are driving. Do more billable hours mean more late nights in the office, or does client engagement mean more time spent in after-hours social situations? What are the behaviours that KPIs are driving at your organisation?

Organisations use KPI and policies that were developed at a different time and are part of the reason for the challenges today. KPIs used for promotions, for example, could drive behaviours that were beneficial to the firm 10, 20, or 40 years ago. Instead, we need to think about what skills and behaviours are needed today or ten years from now. For example, working independently was highly valued a few years ago, but effective collaboration skills are much more valuable today and in the future. Not only are the in-demand skills changing, but often there were many assumptions and a lack of research when the original KPIs were developed.

Challenge #2: Employee Listening

The next challenge Daly points out is that many organisations don’t really know what their employees need or have insights into their daily lived experiences at work. Many organisations don’t really know what diverse employees want. Though they may collect and analyse data from surveys and other qualitative data, a lot of assumptions are made by most firms. Or worse, organisations rely on generic research into the topic, trying to apply generalised findings to help with the specific challenges at their organisation.

Daly points out that it’s important to ask your employees the right questions and think about what they need to be successful. This is particularly important for leadership development in diverse populations, solutions need to be customised to your employees’ specific needs.

Focusing on the positive nature of Perez’s overall argument, she argues that there is an opportunity to change the world around us. And, as HR and HR analytical professionals, we have a massive opportunity to make major changes in our own organisation, improving our business outcomes and the lived experience of our employees.

How HR and HR analytics leaders can break the bias

Returning to a question we raised earlier in this article, how can HR use advanced data techniques to improve outcomes and not amplify past mistakes? Here are three places to start to break the bias in the data we are using to make decisions.

1) Policies and KPIs:

The first thing that HR and analytics leaders should do is review their policies and KPIs for activities such as hiring, promoting, succession planning, and salaries.

Daly tells us that we cannot expect a different outcome of workplace diversity in a structure and system that created the inequalities in the first place.

• Policies should be reviewed for their impact on different employee groups, thinking about the behaviours that they drive
• KPIs, particularly those used for hiring and promoting, should be reviewed and balanced with skills and behaviour-based KPIs

2) Diagnostic Analytics:

Use diagnostic analytics to investigate the lived experience of people in your organisation. Many organisations make assumptions about what employees need.

By using diagnostic analytical techniques, the lived experience of employees can be understood, and programmes can be developed to support their individual needs.

• Use interviews, surveys, focus groups, and other qualitative techniques to collect the lived stories and voices of employees
• Develop policies and programmes that meet their needs or are flexible enough to meet individual needs; this is particularly true of women working to gain leadership roles or are in leadership roles

3) Real Resources:

Not just talking the talk but putting time and resources into making things happen! Research has found that senior leadership often talks about diversity or improvements but is disconnected from what’s really going on. Organisations can make meaningful changes by first listening and then dedicating real resources to improve policies and programmes.

Megan Butler is a researcher and ethics advocate in the application of AI in HR and the future of work. As Rizing’s Future of Work Strategist, she is actively focused on enabling our clients to define the next generation of digital HR and HR analytics best practices using evidence and experience.