पाठशाला Pathshala · दल Dal, The team · Lesson 22 · Build

Diversity that improves decisions, not the deck

A team that went to the same colleges and lives in the same three cities makes the same mistakes with confidence. How to find where your hiring funnel narrows, fix it stage by stage and measure the change.

Pathshala, The Founder Library · 11 October 2026 · 7 min read

Spools of sewing thread in many colours arranged in rows on a workshop rack. — Photograph: cottonbro studio · Pexels

Many Indian startups at twenty people look like one person repeated: the same few colleges, the same three cities, the same languages at home, mostly men, mostly under thirty-five. Nobody chose that. It is what a hiring process produces when it runs on referrals and pedigree under time pressure. The cost does not show on a diversity slide. It shows in decisions that a room of similar people makes quickly, confidently and wrongly.

This lesson starts with what a different voice does to a decision, then follows the mechanisms that keep teams homogeneous, the funnel measurement that finds where, the fixes at each stage, and the legal floor every Indian employer has to meet.

What a different voice does to a decision

The useful evidence is about decisions, not about diversity as a value. In a study described by Kellogg Insight, groups of four had to identify the most likely suspect in a murder mystery. Three members always came from the same social group; the fourth, joining a few minutes in, came either from the same group or a different one. Groups with an outsider found the right answer more often. They also reported much lower confidence in their answer. The researchers concluded that the outsider triggered more careful processing of information: the old-timers explained their reasoning, checked their assumptions and noticed what they had missed.

That is the case for a startup. The decisions that kill companies, a market misread, a price set for people like the founders, a product that assumes an English-speaking metro user, are made in rooms where everyone agrees quickly. A team whose members grew up in Indore, Guwahati and Kochi, or who have managed a household budget, a factory floor or a kirana counter, catches different errors. The discomfort the study measured is the mechanism working.

The deck version of diversity does the opposite. A target announced to investors, met by hiring a few people from under-represented groups into junior roles with no say in decisions, changes the photograph and not the room where decisions are made. It also tells the people hired that they were hired for the photograph, which is the fastest way to lose them. The version that improves decisions puts different experience where decisions happen: on the interview panels, in the product reviews, in the pricing discussion and, as the company grows, in the leadership team and on the board. It also needs people who feel able to disagree, because the study’s effect came from the outsider’s presence changing how the group reasoned. An outsider who stays silent changes nothing, and people stay silent in teams where disagreement is punished, which is a management problem before it is a hiring one.

How a homogeneous team builds itself

Four mechanisms do most of the work, and none needs anyone to intend the outcome. Referrals. The first ten employees refer people like themselves, and referral hires feel faster and safer, so the company leans on them harder. Pedigree screens. A filter on college name screens out most of the country’s graduates before anyone reads what they built; the [interviewing lesson](/library/interviewing-for-judgement-not-pedigree) makes the case for judging the work instead. Panels that look alike. Candidates read the panel as a picture of who succeeds here, and interviewers rate familiar styles of speaking higher. Where and when the work happens. A role that requires relocating to Bengaluru, late meetings and no flexibility around care filters out people with families to look after, and they are disproportionately women.

Measure the funnel, stage by stage

Measure before fixing, because the leak is rarely where the founders expect. Track every candidate through five stages: applied, screened, interviewed, offered, joined. For each group you track, compute the pass rate at each stage and compare it with everyone else’s. The US Equal Employment Opportunity Commission’s guidelines use a four-fifths rule of thumb: when one group’s selection rate is below 80 per cent of the highest group’s, the step deserves a closer look. Their worked example has one group hired at 60 per cent and another at 30, a ratio of half. It is not Indian law and not proof of anything; it is a useful threshold for where to look.

A group of people winnowing rice with woven baskets on a mat spread with grain in bright sun. — Every stage of a hiring funnel is a winnowing. Count what each one lets through before deciding which one to change. Photograph: Mehmet Turgut Kirkgoz · Pexels

Collect the data carefully. Ask only for dimensions you can ask about lawfully and respectfully, gender, disability, home state, first-generation graduate, and make every question voluntary with a decline option. Keep the answers apart from the people making the hiring decision and report only in aggregate. Under the Digital Personal Data Protection Act, 2023, consent must be free, specific, informed, unconditional and unambiguous, and limited to the data needed for the stated purpose, so say on the form exactly what the data is for.

At the defaults the group is 30 per cent of applicants and about 16 per cent of the people who join. Almost all of the loss happens at one stage: screening, where the group passes at 25 per cent against everyone else’s 40, a ratio well below four-fifths. Interviews are even. Raise the screening rate to match and the number who join from the group rises by about 60 per cent without touching any later stage. That is the usual finding: one stage does most of the damage, and it is often the cheapest one to change, a résumé filter rather than a culture.

Count the funnel before you fix it. One stage usually does most of the damage, and it is rarely the one the founders would have guessed.

Fixes at each stage

Sourcing. Post roles where the people you are missing look: regional colleges, return-to-work programmes, communities in Indian languages, cities beyond the metros for roles that can be remote. Cap referrals at a share of hires, or pay a higher referral bonus when the referral widens the pool. Screening. Remove the college filter, replace the résumé screen with a short work sample scored blind, and write down the three things the screen is for. Interviews. Use the same questions and a written scorecard for every candidate, and put at least one person on each panel who is not from the dominant profile. Offers. Make every offer from the [compensation bands](/library/compensation-bands-for-startup), so negotiation skill, which is unevenly taught, does not decide pay. Joining. If acceptance is where a group drops out, ask the people who declined why; the answer is usually flexibility, location or a signal from the panel.

Retention is the last stage of the funnel. Track exits by group within the first year. A company that hires a more varied team and then loses it in eight months has changed its intake and not its culture, which the [culture lesson](/library/culture-is-what-you-tolerate) treats as the thing that is tolerated rather than written.

The legal floor and the policies above it

Some of this is not optional. Under the Sexual Harassment of Women at Workplace Act, 2013, every employer must constitute an Internal Committee by written order, presided over by a senior woman employee, with at least half its members women and one external member; complaints from workplaces with fewer than ten workers go to the district’s Local Committee. Failing to constitute the committee is punishable with a fine of up to ₹50,000. Under the Code on Social Security, as the government’s factsheet summarises, maternity leave is up to twenty-six weeks and establishments with fifty or more employees must provide a crèche. Checked 10 October 2026.

Above the floor, the policies that change who stays are few and cheap: flexible hours written down rather than granted as favours, parental leave for fathers, a clear return-to-work plan after maternity leave, accessible offices and documents, and promotions decided in the same calibrated way as ratings, so that the people who ask loudest are not the only ones promoted.

The quarterly diversity review

Every quarter, in one hour: the funnel by stage for each group you track, with the four-fifths ratio at each stage, pooled over the last four quarters so small numbers do not mislead; first-year exits by group; and the pay position within band by group. Pick the one stage with the worst ratio and agree one change to it, with an owner and a date. Every year: the panel composition for every interview loop, the referral share of hires, the Internal Committee’s annual report, and a check that every policy is written down where a new joiner can find it. Report the numbers to the team, not just the board, because a measured change is more convincing than a statement of intent.

Nothing here is legal advice. The four-fifths ratio is a screening heuristic from US guidance, not an Indian legal standard; take advice before collecting any sensitive personal data.