Name-blind recruitment: the weakest form of blind hiring

Why names became the focus

Because the evidence is unusually clean. Send two applications identical in every respect except the name at the top, and the response rates differ. Bertrand and Mullainathan's field experiment is the canonical version, and it has been replicated in enough markets that the underlying effect is not seriously disputed.

That produced an obvious-looking intervention: take the name off. It is cheap, it is easy to explain, and most applicant tracking systems have a toggle for it.

The problem is that it treats a symptom of the format as the whole disease.

What a name is actually carrying

A name is dense. In one field it frequently signals:

• Ethnicity and national origin — the effect the studies measure
• Gender — for most names, immediately
• Religion — for many
• Generation and rough age — naming fashions move in decades
• Class and region — spelling variants, hyphenation, honorifics
• Migration history — an anglicised first name with a non-anglicised surname tells a story

So removing it removes several inferences at once, which is genuinely worth something. But notice why it carries so much: because it is a personal identifier sitting in a document full of other personal identifiers. The name is the most concentrated one, not the only one.

What survives the redaction

Here is the same CV with the name removed. Ask yourself how anonymous it is:

[REDACTED] Lahore, Pakistan → Dubai, UAE · [REDACTED]@gmail.com BSc Computer Science, University of the Punjab, 2011–2015 Senior Engineer, Careem, 2019–2024 Engineer, Systems Ltd, 2015–2019 github.com/[REDACTED] · Native Urdu, fluent English

Everything about this person is still legible: approximate age from the graduation year, national origin from the education and languages, migration path from the locations, and — in a market where a specific company had a specific team — quite possibly the individual. The GitHub link resolves the rest in one click.

The reviewer does not need to be trying. The inferences arrive unbidden, which is precisely the mechanism name-blind screening was meant to interrupt.

Three ways name-blind fails in practice

Manual redaction fails silently. Somebody has to remove the name from a document, and at volume they will miss some. Nobody audits the misses, because a leaked name looks exactly like a normal application. Worse, partial coverage makes the process less comparable: some candidates were reviewed blind and some were not, so you are now mixing two conditions.

The metadata betrays it. Filenames (priya-sharma-resume-v3.pdf), PDF author fields, and editing history all survive a redaction that only touched the visible text.

It ends at the first email. As soon as the recruiter replies, or a call is scheduled, the name is back — and if that happens before the substantive evaluation, the anonymity covered the least consequential part of the process.

The better question

Instead of "what should we hide?", ask: what does this decision actually need to see?

That inverts the problem, and the answer is usually short. To judge whether someone can do a job you need evidence about their ability to do it. You do not need where they were born, what year they graduated, which company employed them, or what they look like.

Framed that way, the design changes shape. You are not redacting a document that contains too much; you are collecting the evidence the decision requires and never gathering the rest. Redaction is a compensating control for having collected the wrong thing in the first place — and compensating controls leak.

This is the difference between name-blind screening and genuinely blind evaluation: one removes a field from a document, the other changes what reaches the evaluation at all.

If you are going to do name-blind, do it properly

For teams who are not going to rebuild their process this quarter — reasonable — there is a much stronger version of the same idea:

1. Automate it. Manual redaction at volume does not work. If your ATS cannot do it, that is a procurement finding.
2. Extend the field list. Photograph, dates, schools, employers, addresses, links, referral source.
3. Strip filenames and metadata, not just visible text.
4. Keep it blind through the whole screening stage, not just the first pass.
5. Standardise what reviewers see, so an application with more redactions is not visibly different from one with fewer.
6. Do not schedule anything before the evaluation is recorded.
7. Measure selection rates by group before and after. If nothing changed, that is worth knowing rather than assuming.
8. Say what you did, accurately. "We remove names during initial screening" is honest. "We hire blind" is not.

That last point matters more than it looks. Overclaiming is how a genuine improvement becomes a credibility problem when a candidate notices their university was visible the whole time.

How CalHire avoids the problem entirely

There is no name to redact, because a candidate's identity never enters the evaluation.

Candidates declare skills and prove them through one cheat-resistant assessment. The composite score comes from a skills test, a text interview and role-fit — with no résumé-match component. Employers see verified skills and scores under an opaque handle that is never derived from personal information. Personal data is stripped at a hard architectural boundary before any model sees a candidate, across scoring, the interview and ranking. An uploaded résumé only pre-fills declared skills; it is never stored as the record and never shown as one.

Identity — and right-to-work detail — unlocks only when an employer advances a candidate to an allowed stage and the candidate consents, for that one employer, recorded in an immutable ledger.

That is the structural version of what name-blind recruitment gestures at. See the features page for the anonymisation boundary, or for candidates for how the reveal works from the candidate's side.

Redaction asks people to unsee something. Architecture means there was never anything to see.