Unconscious Bias & Tokenism Explained
Unconscious bias, affinity bias, and tokenism are three of the most-cited concepts in workplace inclusion, and three of the most misused. Unconscious bias is a broad description of how brains take shortcuts. Affinity bias is one specific and very well-documented version of that shortcut, and it does a lot of damage in hiring. Tokenism is what happens when an organisation tries to address bias with visibility instead of structure, and ends up making the problem worse. Getting the definitions right matters, because the interventions that work for one are different from the interventions that work for the others.
This guide is written for HR leads, DEI owners, and hiring managers who want a clear working definition of each term, real workplace examples of what they look like, an honest read on which interventions have evidence behind them, and where the hiring process specifically is the highest-leverage place to reduce their effect.
Unconscious bias: the definition
Unconscious bias, sometimes called implicit bias, is the set of automatic mental associations and shortcuts a person applies to others without conscious awareness. The concept comes out of social psychology, most famously through the Implicit Association Test developed at Harvard in the late 1990s, which showed measurable associations between social categories and evaluative attitudes in populations across dozens of countries.
Two things about the definition matter. First, "unconscious" does not mean "innocent." A bias is still a bias whether or not the person holding it can name it, and the effects on the people on the receiving end are the same. Second, the finding that everyone has unconscious biases is not a licence to shrug. The correct response is to design decisions and processes that do not rely on individual judgement being unbiased, because it will not be.
Common categories that have been studied extensively include bias by race, gender, age, disability status, name origin, accent, weight, and education pedigree. Each shows up in different parts of the employee lifecycle. Name-based bias in resume screening. Accent-based bias in interviews. Pedigree bias in promotion.
Affinity bias, defined
Affinity bias is the specific unconscious tendency to favour people who share a background, experience, or trait with the person doing the evaluating. It is the boss who mentors the new hire from her old graduate programme more actively than the equally junior colleague she does not have that connection with. It is the hiring manager who rates the "very engineering-culture" candidate higher without being able to say what "engineering culture" concretely means.
Affinity bias is one of the best-documented biases in hiring specifically. It is corrosive because it compounds: teams built through affinity bias look the same, which makes the affinity bias in the next hire stronger, and so on. Over five years an unmanaged affinity-bias hiring loop produces a monoculture that is very hard to break, and that then does poorly on innovation and market fit.
The interventions with the best evidence are structural. Blind resume review (removing names, schools, and photos from the initial screen). Structured interviews with a fixed question set. Calibrated scoring rubrics that require the interviewer to justify each score against defined criteria. Diverse interview panels. None of these eliminates bias; each measurably reduces it.
Tokenism, and why it usually backfires
Tokenism is the practice of making a surface-level gesture toward inclusion without changing the underlying structure. The single visible hire from an underrepresented group, the one photo on the careers page, the "diverse" panel that consists of one person from the target group and four from the majority. Tokenism is often well-intentioned. It is almost never effective.
The reason it backfires is what it does to the person in the token position. They are asked to be present in every DEI photo, to speak for their entire group, to lead the diversity committee on top of their day job, and to represent the company's inclusion values at every external event. That is a lot of unpaid labour, and it is a common driver of the very attrition the tokenising organisation was trying to prevent.
The right response is structural inclusion, not visible inclusion. That means changing who gets in the door (sourcing pipelines), who gets through interviews (structured scoring), who gets promoted (calibration meetings), and who gets sponsored (formal mentorship). Photos on the careers page can wait until the underlying picture has actually changed.
Affinity groups, and how they fit in
Affinity groups, more commonly called Employee Resource Groups (ERGs), are employee-led communities organised around shared identity or shared experience. Common examples include Women in Engineering, Black Employees at Company, LGBTQ+ ERG, Parents at Work, Veterans ERG, Neurodiversity Group.
ERGs, done well, do three things. They give members a space for community and mutual support. They surface systemic issues to the business (the group's collective view carries more weight than any single employee's complaint). And they advise HR and leadership on policy questions from a member perspective.
ERGs, done badly, become the mechanism by which the organisation offloads DEI work onto the people already carrying the most emotional load. If a company expects ERG leads to run the group in their spare time, with no budget, no executive sponsor, and no explicit recognition, it has not built an ERG. It has built another form of tokenism.
The healthy pattern: named executive sponsor, an annual budget, dedicated time (allocated hours, not "in addition to your job"), and formal input on relevant policy decisions.
Where bias enters the hiring funnel
Bias is not evenly distributed across the hiring process. It concentrates at four decision points.
- Resume screening. Landmark research on name-based bias in the US showed that identical resumes with African-American-sounding names received significantly fewer callbacks than those with white-sounding names. The effect persists in follow-up studies.
- The interview loop. Interviewer notes tend to be written to support the initial impression, not to test it. Unstructured interviews are almost worthless as a predictor of job performance, and they are prime territory for affinity bias.
- Scoring and calibration. The same answer scored by two interviewers can differ by more than a point on a 5-point rubric if there is no calibration. Calibration meetings, held before hiring decisions are finalised, are one of the highest-value bias interventions.
- Offer negotiation. Research consistently shows different negotiation outcomes for otherwise comparable candidates by gender and race. Fixed salary bands, disclosed to candidates, reduce this.
Any bias-reduction programme that ignores these four points and focuses on training alone is unlikely to move the numbers.
What actually reduces bias in hiring
The interventions with the strongest evidence are structural, not attitudinal:
- Structured interviews using the same question set for every candidate for a role, in the same order, scored against pre-defined criteria.
- Calibrated scoring rubrics with defined levels and required justification for each score. Vague rubrics ("communication: strong") do not reduce bias.
- Diverse interview panels where possible, with awareness that being the only member of a group on a panel puts pressure on that person.
- Blind or partially blind resume screening in the earliest funnel stage.
- Written debrief before verbal discussion to prevent the first speaker's opinion from anchoring the room.
Fabric's AI-led Round 1 interview enforces several of these structural interventions by design. Every candidate for a given role gets the same set of role-specific questions in the same conversational format. Scoring is against a defined rubric. Cheating detection is built in as a core part of the product, not an add-on, which matters because inconsistency in how cheating is caught across candidates is itself a source of unfairness. The recruiter or panel then reviews Fabric's output and makes the final call. Fabric does not decide who is hired.
*Fabric's technical-depth and eligibility scoring is designed to flag mismatches and surface them to your recruiter. It's a signal for your team to weigh, not an automatic reject.* Reducing interviewer variance across a large candidate pool is a structural change, not a training-based one, and it is where the largest bias-reduction gains in Round 1 currently live.
Related posts
- DEI, culture, and benefits
- What is DEI?
- Allyship, ERGs, and cultural competence
- Structured vs. situational interviews
- Interview methodology
FAQ
What is the definition of unconscious bias?
Unconscious bias is a set of automatic mental shortcuts, learned from experience and culture, that shape how a person perceives and judges others without them realising it. Everyone has these biases; the question is what an organisation does to counter them in high-stakes decisions like hiring and promotion.
What is affinity bias?
Affinity bias is the specific unconscious tendency to favour people who resemble ourselves in background, education, appearance, or shared interests. It shows up in hiring as the interviewer who rates the candidate from the same alma mater half a point higher without noticing.
What is tokenism in the workplace?
Tokenism is the practice of making a superficial gesture toward inclusion (usually a single visible hire or promotion from an underrepresented group) without changing the structures that produced the underrepresentation. It is often well-intentioned and almost always counterproductive.
How can unconscious bias affect hiring?
Bias shows up at every stage: which resumes get read, which candidates get past a phone screen, which answers get scored generously in an interview, and which offers get pushed harder. Structured interviews and calibrated scoring reduce (though do not eliminate) the effect.
What are affinity groups at work?
Affinity groups, also called Employee Resource Groups or ERGs, are employee-led communities organised around shared identity or experience (for example, Women in Tech, Black at Company, or Parents at Work). They exist to build community, surface issues, and advise the business on inclusion.
Does unconscious bias training actually work?
The research is mixed. Standalone one-off training sessions show small and often short-lived effects, while structural interventions like blind resume review, calibrated scoring, and diverse interview panels show larger and more durable ones. Training is most useful when it changes process, not just awareness.