Does diversity training work in India? What two mandates and one experiment show.
Most of the diversity conversation in Indian offices runs on American research. The famous study is American, the categories are American, and the objection writes itself: that's America.
It is a fair objection, and it is also unnecessary, because India has better evidence than America does on this particular question. Not because we studied it harder. Because we legislated it.
India made two kinds of rule about fairness at work, within months of each other. One was a rule about a structure. The other was a rule about an attitude. Both are more than a decade old now, and the results have split so cleanly that they amount to a finding.

What a mandate can buy
The Companies Act 2013 required every listed company above a size threshold to appoint at least one woman director by April 2015. SEBI went further, requiring an independent woman director in the top 500 listed firms by 2019 and the top 1,000 by 2020. India became the first developing country to impose a women-on-boards quota.
It worked. Women's share of directorships in NSE-listed companies rose from 8.2% in 2014 to 12.7% in 2015 — a 54.9% jump in a single year, with 497 women inducted onto boards in FY2015 alone. The feared side effect, that boards would be diluted, did not show up in the data.
If you want a number to move, a mandate moves it. That is not a small thing and we should not be grudging about it. Several hundred women entered Indian boardrooms who would otherwise not have.
What the same mandate could not buy
Then look at what the companies did with the rule.
| What the mandate produced | |
|---|---|
| One woman, and no more | 73% of companies |
| Three or more women | 4.3% of companies |
| Women per listed board | 1.03 on average |
| Share of all board seats | 16.8% |
| Key committees chaired | Under 8% of audit, risk, nomination and remuneration committees |
| Chairperson roles | 0.7% of independent women directors |
| Pay | About half that of male directors — commissions 60% lower, share units 52% lower |
| Tenure | 40% shorter |
And then the detail that settles the argument.
The last possible day
49 women were appointed to boards on 31 March 2015 — the final day of the deadline. 39 were appointed on 1 April 2019, the day SEBI's rule took effect.
Among women serving on more than one board, 40% received their second appointment within six months of the first and 15% within a month — and of those, more than half were to a related company. The study that compiled this reads it as evidence of skirting the mandate.
So the structure arrived: on the last possible day, in the minimum possible quantity, without the chairs, the committees, the pay or the tenure that would make it mean anything.
The rule bought a seat. It did not buy inclusion. But the seat appeared, and it appeared fast, because a structure is countable and a mandate can count.
The other kind of rule
The POSH Act, 2013 did something rarer. It did not mandate a structure. It mandated the training itself — requiring employers to run workshops and awareness programmes at regular intervals to sensitise their employees, with penalties for non-compliance.
As far as we can establish, this makes it the most mandated, most audited, most legally enforced piece of behavioural training in the country.
In May 2023 — ten years after the Act — the Supreme Court of India, hearing Aureliano Fernandes v. State of Goa, observed that it was not known whether the committees the Act requires even existed across departments and districts, and that its directions were not being complied with. It ordered the Union, every State and Union Territory and a long list of public bodies to verify, publish, train and file affidavits. In December 2024 it had to issue follow-up directions for a district-wise compliance survey.
Eleven years after the law, the country's highest court was still asking whether the committees exist.
And the part that should concern anyone who buys this kind of training: we could find no published Indian study measuring whether sensitisation of this sort changes behaviour. The Indian literature examines compliance and committee capability — one recent paper finds that the committee members themselves lack adequate training. That is a finding about the trainers, not the trained.
Mandating a structure changes the numbers. Mandating an attitude changes almost nothing.
Structure responds to a mandate because structure can be counted. Attitude does not, because nobody can audit the inside of a person's head. So what gets produced instead is the artefact of compliance: one woman on every board, on the last day. A workshop in the annual report, with nothing measured on either side of it.
This is the Indian version of what Frank Dobbin and Alexandra Kalev found across more than 800 American firms over thirty years — and it is the stronger version, because it is domestic and because both arms of the experiment are visible from here.
The Indian problem is not an awareness problem
This is where most diversity programmes are solving the wrong thing.
If awareness were the gap, thirteen years of mandated sensitisation would have closed some of it. AIMA and KPMG's 2026 study of women's leadership in corporate India found that 55% of employees say their organisation has a well-defined process for assessing candidates for leadership roles — and only 28% believe that process is fair and transparent. 42% report pay disparities. 36% believe management favours men for the harder roles. 65% point to mid-career exit as the moment women leave.
Nobody inside that gap needs to be informed that bias exists. They are telling you they can already see it. The gap is between the stated system and the lived one — what Chris Argyris called the espoused theory and the theory-in-use — and no amount of describing the espoused one closes it. A slide deck about the policy is a description of the 55%, delivered to people whose complaint is the 28%.
The national data points the same way. India's female labour force participation is 40.0%, against 79.1% for men. Asked the main reason for being outside the labour force, 69.8% of men said they wanted to keep studying; 44.4% of women cited child care or home-making. The binding constraint on Indian women at work is unpaid domestic load, and it is measured nationally. Any programme that treats women's advancement as a confidence deficit is arguing with the Periodic Labour Force Survey.
What the Indian evidence says does work
Here is the part we think almost nobody in this market has read, and it is Indian, randomised and published in one of the top journals in economics.
Matt Lowe ran a field experiment in caste-segregated rural India. From a sample of 1,261 men, 800 were randomised into eight one-month cricket leagues. Some were assigned mixed-caste teammates — that randomised collaborative contact, where people share the goal of winning. Opponents were then assigned randomly — that randomised adversarial contact. Outcomes were measured one to three weeks after each league ended, using incentivised behavioural exercises rather than attitude surveys. Nobody was asked how they felt about caste. They were put in situations where their behaviour cost them something.
| Mixed-caste teammates | Mixed-caste opponents | |
|---|---|---|
| The contact was | Collaborative — a shared goal | Adversarial — competing |
| Effect on cross-caste behaviour | +0.23σ | −0.38σ |
| Own-caste favouritism | Down by up to 33% | — |
| Cross-caste trade | Up to +21%, payouts +18% | Reduced |
| Trust | — | Reduced significantly |
The mechanism is almost blunt: cross-caste interactions with opponents were 50 percentage points more likely to be hostile than interactions with teammates. Lowe's own summary of it is the sentence we would put on a wall — this contact reduces division through working together, not winning together.
A second Indian study points the same way from a different angle. Gautam Rao studied a Delhi policy that required many private schools to admit poor children, and found that wealthy students who had poor classmates became more generous, discriminated less and were more willing to socialise across the line. The mechanism was not the curriculum or the teachers. It was sitting next to someone.
Two limits on that
Lowe's experiment is cricket, among men, in rural India, over a month — not a workshop, and the transfer to anything we run is our inference, not his finding. Rao's is a year of shared classrooms, not a two-day programme. Neither is a study of corporate training.
But put them beside each other and the implication is hard to avoid. Most Indian corporate team-building is structurally adversarial. Teams compete. There are points. There is a winning table. Lowe randomised exactly that distinction and the competitive arm reduced cross-group cooperation and trust. The dominant format in our own category is the one that backfired.
What we have learned running these rooms
We are an applied theatre company, so treat the following as observations from our own sessions rather than findings. They are not controlled, and we say what we don't know at the end.
A behaviour has to be named as a moment, not as a value
"Respect" and "inclusion" are nouns. They can be printed, communicated and recited, and they still have no observable shape — nobody can point at the instant one was or wasn't happening. Their opposites, said out loud in a corridor, can. Every behaviour we have successfully put in front of a room arrived as a moment somebody recognised, not a value somebody endorsed.
The language has to be the one people argue in
Marico's unconscious bias scenes were devised bilingually, Hindi and English, because that is how the moments being staged actually sound. Their HR lead, Urvashi Agarwal, described the result as a journey from unconscious to conscious. A scene in the wrong register is a scene about someone else.
Attendance is not participation, and voluntary is not free
When we performed on office floors for a fintech company rather than booking conference rooms, and sent no mandatory invites, 99.3% of the 137 people surveyed said they wanted to see more. That number is only interesting because nobody was obliged to be there. But the same AIMA/KPMG study found that nearly half of the women surveyed had attended no leadership development programme at all in the past year. Voluntary attendance avoids resistance and trades it for a reach problem. Anyone recommending it — us included — owes you a plan for the people who won't walk past.
Six small rooms beat one big one, but a performance scales and practice doesn't
The Marico work ran as six batches rather than one event. Separately we ran an induction for 450 students at MDI Gurgaon, and it did what a performance can do: a whole cohort seeing the same thing, a shared reference, recognition. What it cannot do is practice. Decide which of the two you are buying.
The room will argue about a character long before it will argue about itself
This is the most reliable thing we know, and it is the reason the method exists. Put the decision in someone else's hands on a stage and people will debate it freely — including the part of it that is theirs.
Half a day can be enough when the problem is already in the room
We ran half a day on open communication and creativity for EdgeTech Air Systems, the day before their two-day business planning session. Nothing was devised for them from scratch and nothing was measured afterwards. What their managing director reported was a planning day that produced more usable ideas than their planning days normally do. Treat that as a client's account, not as our evidence — but it is the shape of the case for going in before the meeting rather than after the problem.
How a scene gathers data a feedback form cannot
This follows directly from the 55/28 problem. If the gap is between the stated system and the lived one, you need an instrument that observes the lived one. A form asks people to report on themselves. A scene makes the organisation show you.
Six things a room returns that a survey does not:
1 · Where the room stops the scene
In forum format anyone can halt the action. Which beat they halt it at is data — it marks the moment your people recognise. Run the same scene in two offices and the stop-points differ; where they differ, you have found a real difference between those two cultures rather than a self-reported one.
2 · What they propose instead
The alternatives the room offers are a live inventory of the behaviours your people believe are actually available to them. If nobody ever proposes going to the skip-level manager, that option does not exist in your organisation, whatever the policy says.
3 · Who intervenes, and who never does
Note the seniority of the first person to stop the scene, and who stays silent throughout. A survey measures claimed willingness to speak up. A room measures it.
4 · Where the laugh is, and where the silence is
Recognition laughter is the most honest signal in the room — the loudest laugh marks the most widely experienced behaviour. The beat where a room goes quiet marks the one that is still live.
5 · The private inference, written down before the reveal
Stop the scene at the moment before the decision and ask everyone to write what they concluded and what they would do. Then show them the other character's version. Those slips give you a distribution of the inferences your organisation makes under ambiguity — a baseline you can repeat.
6 · Recognition, measured properly, before and after
Not "do you value inclusion" on a five-point scale. A forty-second clip and one question: point to the moment this went wrong. That is a comprehension measure with a right answer, it can be run before the session as a genuine baseline, and it beats any satisfaction score.
Then the follow-up that turns it into behaviour data: at 30 days, ask managers for one logged instance where someone named the behaviour in the scene's own language. When a team starts describing a real meeting using a phrase from a scene, without prompting, you are watching a shared vocabulary do work — and vocabulary adoption is countable in a way that an attitude never is.
What we don't know
We have not run a controlled trial. We have no baseline-and-comparison design behind the numbers above, we have not measured anyone's behaviour at 90 days, and we have never had a client agree to a control group — which is a gap in our evidence, not in the method. Nobody in Indian applied theatre has published that study, including us.
What we are confident about is narrower and still useful. Mandating a structure moves a number. Mandating an attitude does not. The Indian problem is not that people cannot see bias; it is that they can see it and do not trust the process that is supposed to handle it. And the only randomised Indian evidence we know of on shifting a bias says the thing that worked was collaboration, and the thing that backfired was competition.
Most of this industry is still buying the competition.
Sources: Companies Act 2013 and SEBI LODR; board-composition analysis of 1,944 NSE-listed companies and 192 women directors (IIM Ahmedabad / FICCI); the POSH Act, 2013; Aureliano Fernandes v. State of Goa, Supreme Court of India, 12 May 2023, with follow-up directions of 3 December 2024; AIMA–KPMG study of women's leadership in corporate India, 2026; Periodic Labour Force Survey 2025, MoSPI; Matt Lowe, Types of Contact: A Field Experiment on Collaborative and Adversarial Caste Integration, American Economic Review 111(6), 2021; Gautam Rao, Familiarity Does Not Breed Contempt, American Economic Review 109(3), 2019; Dobbin and Kalev, Why Diversity Programs Fail, Harvard Business Review, July–August 2016.
Tamasha Entertainment runs applied-theatre workshops for corporate teams from Gurgaon — diversity and inclusion, behaviour change, team building and leadership sessions led by working actors. If you are buying a diversity programme this year, ask whoever is selling it what they measure, and when.
