Pay Equity Audits: A Practical Guide for Indian Employers

Pay Equity Audits: A Practical Guide for Indian Employers

A pay equity audit is the exercise of checking whether the people in your company who do substantially similar work are paid in a way you can actually defend — and, where they are not, understanding why. It is not a survey, not a benchmarking report, and not a single headline number about men and women. It is an internal investigation into your own payroll, done with enough rigour that you would be comfortable explaining the results to your board, your investors, and eventually to your own employees.

Indian employers are being pushed toward this from several directions at once. Pay transparency rules have been spreading across other markets, which means Indian companies with global customers, global investors or global parent entities are increasingly being asked questions about pay distribution that they cannot currently answer. Candidates are more willing to ask about ranges and less willing to disclose current salary. And the consolidation of India's labour legislation into four labour codes, brought into effect from late 2025, has forced almost every employer to revisit compensation structures anyway because of the revised definition of wages. If you are already rebuilding salary structures, it is the cheapest moment you will ever get to check whether those structures are fair.

This guide is written for HR leaders, founders and compensation teams in Indian companies of roughly 50 to 1,000 employees — big enough that pay decisions have drifted across managers and years, small enough that you do not have a dedicated rewards analytics function. It walks through scoping, data preparation, two levels of analysis, remediation, upstream fixes and the staged move toward transparency. One caveat up front and repeated throughout: nothing here is legal advice. Verify current requirements under the applicable law as notified in your states, and involve employment counsel before you act on findings.

What a Pay Equity Audit Actually Is

At its core, a pay equity audit asks a narrow question: among employees doing work of a similar nature and similar value, is pay explained by legitimate, job-related factors — or is some part of it explained by gender, or by another characteristic that should have nothing to do with compensation?

That framing matters because it sets the scope. You are not asking whether everyone is paid the same. You are asking whether differences in pay track differences in the things that are supposed to drive pay: level, scope of responsibility, skill scarcity, sustained performance, location, and so on. Where a difference survives after you have accounted for all of those, you have something that needs an explanation or a correction.

A good audit produces three things:

• A measured picture of pay differences across groups, both before and after controlling for job-related factors.
• A short list of specific individuals or cohorts whose pay cannot be explained.
• A documented method, so that next year's audit is comparable and so that the reasoning behind any correction is recorded.

A bad audit produces a slide with one percentage on it.

Three exercises people confuse with each other

These get mixed up constantly in Indian HR conversations, and the confusion leads to wasted effort.

Market benchmarking compares your pay to the outside world. You buy or participate in a salary survey, map your roles to survey roles, and see whether you sit at the 25th, 50th or 75th percentile. This tells you about competitiveness and retention risk. It tells you nothing about internal fairness — you can be at the 75th percentile overall and still be underpaying a specific group inside your own walls.

A raw gender pay gap number compares average or median pay of all women to all men across the company, with no adjustment. It is a single headline figure. It is genuinely useful, but it measures something quite different from equal pay.

A pay equity audit compares like with like inside your organisation and isolates the portion of any difference that job-related factors cannot explain.

You need all three eventually. They answer different questions and they should not be substituted for one another.

Unadjusted vs Adjusted Gaps — the Single Most Misread Number in Compensation

This distinction is where most internal pay conversations go wrong, so it is worth spending real time on.

The unadjusted gap (sometimes called the raw or headline gap) is the difference between the median or mean pay of one group and another, across the whole population, ignoring role, level and everything else. If your median woman earns less than your median man, you have an unadjusted gap. Full stop, no context.

The adjusted gap (sometimes called the controlled or residual gap) is what remains after you statistically account for legitimate factors — job level, job family, location, tenure, performance, employment type. It attempts to answer: for the same work, at the same level, in the same place, with similar experience and similar performance, is there still a difference?

These two numbers typically tell very different stories, and both stories are true.

Why a large unadjusted gap usually is not "unequal pay for the same work"

In most Indian companies with a meaningful unadjusted gap, the gap is not driven by two people sitting side by side, doing the same job, on different salaries. It is driven by who sits where.

If women are concentrated in support functions and junior levels, and men are concentrated in engineering, sales leadership and senior levels, the company's median female salary will be lower even if every single individual at every single level is paid identically. That is a representation problem and a level-distribution problem — not a same-work pay problem.

This distinction is not a way of dismissing the unadjusted number. It is the opposite. The unadjusted gap is often the more important of the two, because it points at the structural issues — hiring pipelines, promotion rates, function segregation, attrition at mid-career — that actually determine lifetime earnings. The adjusted gap is narrower but more legally and operationally actionable.

Treat them as two findings with two different remediation paths:

Report both. If you report only the adjusted gap you will look evasive. If you report only the unadjusted gap you will misdiagnose the cause and spend money on the wrong fix.

The Indian Legal Backdrop, Kept General

Indian law has carried an equal-remuneration principle for decades, and that principle has now been absorbed into the consolidated wage legislation. In broad terms, the framework covers equal remuneration for the same work or work of a similar nature regardless of gender, and non-discrimination on the ground of gender in recruitment and in conditions of service, subject to the exceptions the law itself provides.

Alongside this, the four labour codes brought into effect from late 2025 reorganised a large amount of Indian employment law and introduced a revised statutory definition of wages. That definition changes how the components of a salary structure relate to each other and to statutory benefit calculations, which is why so many Indian employers spent the last stretch restructuring CTC. Restructuring is exactly when unintentional inequities get created — and exactly when they can be caught.

A few practical points, deliberately general:

• The rules operate at the level of the same work or work of a similar nature, which is a question of job content, not job title. Your internal job architecture is doing legal work whether you realise it or not.
• Obligations around registers, records and disclosures exist and vary by state notification, establishment type and headcount. Confirm what applies to each of your entities with counsel; do not assume a single national answer.
• Anti-discrimination expectations extend beyond pay into recruitment and conditions of service, so an audit that looks only at salary misses part of the picture.
• Data protection obligations under India's DPDP framework apply to the personal data you will be handling throughout this exercise.

I am deliberately not citing sections, penalties, thresholds or cases here. Requirements change, state notifications differ, and the accuracy of the specific citation matters enormously if you are relying on it. Verify the current position with employment counsel before you scope the audit, and again before you act on its findings. Treat this article as a method, not as a legal opinion.

There is also a strategic reason to move before you are required to. Transparency obligations elsewhere have taught a consistent lesson: companies that discovered their gaps only when they had to publish them ended up making expensive, rushed corrections under public scrutiny. Companies that audited quietly, fixed quietly, and then disclosed had a far better story and a far smaller bill.

Scoping the Audit

Scoping is where you decide what you will be able to say at the end. Get it wrong and you will either drown in data or produce a finding so narrow that nobody acts on it.

Which legal entities

Indian groups often run multiple entities — an operating company, a global capability centre, a services arm, sometimes a separate entity for a different state or a different line of business. Decide early whether you are auditing entity by entity or on a consolidated group basis.

Audit entity by entity for compliance-facing analysis, because obligations attach to the establishment. Audit on a consolidated basis for management insight, because that is how careers and internal mobility actually work. Most companies end up doing both views off the same dataset.

Which employee groups

Start with your full-time permanent population on your own payroll. Then consciously decide how you treat:

• Contract and third-party staff — usually out of scope for the pay analysis itself, but worth a separate look, because heavy use of contract labour in certain functions can quietly shape your representation picture.
• Interns, apprentices and trainees — normally excluded, but note the exclusion so the headcount reconciles.
• Part-time and reduced-hours employees — include them, but analyse on a full-time-equivalent basis so you are comparing rates, not hours.
• Employees on long leave — include with their substantive pay, and flag them; career-break effects are a finding in themselves.
• Founders and C-suite — you can exclude them from the statistical model because the numbers are small and idiosyncratic, but review them separately by hand and say that you did.
• Recent joiners — include everyone, but be ready to look separately at people hired in the last few months, since new-hire pay reflects current market rather than historical drift.

Which pay elements

This is the part teams most often under-scope. Salary is not one number in India.

• Fixed pay: basic, allowances, and the rest of the fixed CTC structure. Post-restructuring, compare consistently — either compare gross fixed across the board or compare the same defined build-up for everyone.
• Variable pay: annual bonus, quarterly incentives, sales commissions. Look at both target and actual payout; a gap in payout with identical targets is a performance-rating or opportunity-allocation issue and needs a different fix.
• Equity / ESOPs: grant size at hire, refresh grants, vesting status. In Indian startups this is frequently where the largest unexplained differences live, because grants are negotiated ad hoc, often by founders, often without a grid.
• Joining bonuses and retention bonuses: one-off but material, and usually the least governed of all pay decisions.
• Allowances and perquisites: transport, relocation, housing support, insurance beyond the standard plan, learning budgets. Include anything discretionary.
• Off-cycle corrections: mid-year adjustments, counter-offers, promotions outside the normal cycle. These are the clearest fingerprints of unmanaged pay decisions.

Run the primary analysis on fixed pay, then repeat it on total cash, then on total compensation including equity. If the story changes across those three views, that itself is the finding.

Which characteristics

Gender is the mandatory dimension and the one with clear legal grounding. Beyond gender, many Indian employers look at differences by location, by whether someone joined through an acquisition, by hiring source (referral vs agency vs campus), and by employment type. Be thoughtful about collecting or analysing other personal characteristics — that raises privacy, consent and sensitivity questions that should be worked through with counsel and your data protection lead before, not after, you pull the data.

Data Preparation: This Is 70% of the Work

Every experienced compensation analyst will tell you the same thing: the analysis takes a day, the data takes a month. Budget accordingly, and do not let anyone promise leadership a result in two weeks.

The reason is that most Indian mid-size companies do not have a single clean source of truth for pay. Fixed salary sits in payroll, ESOPs sit in a cap table spreadsheet, performance ratings sit in a review tool or a separate sheet, job titles are whatever the offer letter said, and levels either do not exist or exist in three competing versions.

Build a clean employee master

Pick one snapshot date and pull everyone active on that date. Then assemble one row per employee with these fields.

Two fields on that list deserve a warning label. Prior salary should be collected only to diagnose whether historical pay is leaking into current pay — never as an explanatory variable in your model. And leave of absence data is personal and sensitive; agree the handling with counsel and your data protection lead before pulling it.

Fix your job architecture first

If you take one thing from this guide, take this: you cannot run a pay equity audit without a job architecture, and most companies of this size do not have one.

A job architecture is a structured map of the work: job families (Engineering, Sales, Finance, Customer Success), sub-families where useful, and levels within each family that describe scope, autonomy and impact. Titles are marketing; levels are the skeleton.

You do not need a twelve-level global framework. For 50 to 1,000 people, five to eight levels per family is usually plenty. What matters is that:

• Levels are defined by work characteristics, not by salary. If you define a level as "people earning between X and Y", your audit becomes circular and worthless.
• The same level means the same scope across families. An L4 engineer and an L4 finance analyst should be comparable in autonomy and impact, even though their market rates differ.
• Every employee is mapped, and the mapping is reviewed by someone who knows the actual work — usually the function head, not HR alone.
• The mapping is done before anyone sees the pay analysis. If levels are assigned after people can see the gap numbers, the levelling will be unconsciously fitted to justify existing pay.

That last point is the most important governance rule in the whole exercise. Freeze the architecture, then analyse.

Define comparator groups honestly

A comparator group is the set of people you will compare against each other — your operational definition of "similar work". Typically it is job family plus level, sometimes plus location band.

Good practice:

• Define groups on job content, not on who happens to sit near whom.
• Aim for groups with enough people to be meaningful. Very small groups produce noisy results and tempt you into over-interpreting a single person.
• Where a group is too small, roll up — combine adjacent levels or related sub-families — but document the roll-up rule before you look at results.
• Where a role is genuinely unique (one Head of Legal, one DPO), take it out of the statistical work and review it individually.
• Write the rules down. The written rule is what protects you from the accusation, internal or external, that you drew the boundaries to make the problem disappear.

Handle location differentials explicitly

India has real geographic pay variation — a senior engineer in Bengaluru and the same role in a tier-2 city are usually not on the same number. That is a legitimate factor, and treating it as one is fine. What is not fine is letting location become an unexamined proxy.

Two checks worth running:

1. Is your location policy written down and applied consistently, or is it applied case by case? A policy applied case by case is a discretion channel, and discretion channels are where gaps form.
2. Since the shift to remote and hybrid work, are two people doing identical work at identical levels being paid differently purely because of where they logged in from when they were hired? That is increasingly hard to defend, and it disproportionately affects people whose location was constrained by caregiving responsibilities.

Handle performance ratings with suspicion

Performance rating is a legitimate control variable. It is also, in many organisations, a variable that carries bias forward.

Before you put ratings into a model, look at the distribution of ratings by gender and by level. If women receive systematically lower ratings in your company for reasons your leadership cannot articulate, then "controlling for performance" simply launders the bias into the residual and makes your adjusted gap look smaller than it is.

The honest approach is to run the analysis twice — once with performance included, once without — and report both. If the adjusted gap closes substantially when you add ratings, you have not solved a pay problem. You have found a performance-rating problem.

Methodology Level 1: The Cohort Comparison Any HR Team Can Run

You do not need a data scientist to start. A careful cohort comparison in a spreadsheet will surface most of the serious problems in a company under a few hundred people.

Step by step

1. Pick your pay measure. Start with annualised fixed pay. Use median, not mean — medians are not distorted by one very highly paid person.
2. Group employees by comparator group (job family + level, optionally + location band).
3. Drop groups that are too small to say anything about, but keep a list of who you dropped and why. Do not silently discard them.
4. For each remaining group, calculate the median pay by gender, the count by gender, and the difference as a percentage of the higher median.
5. Calculate range penetration for each individual — where they sit within the defined pay range for their level, expressed as a percentage. This is often more revealing than raw pay, because it normalises for level automatically.
6. Sort groups by the size of the gap, weighted by headcount, so you see where the money and the people actually are.
7. Flag the outliers within groups too — individuals sitting far below the group median, regardless of gender. Some of what you find will be plain inconsistency rather than a protected-characteristic issue, and it still needs fixing.
8. Aggregate to the company level for the unadjusted number, and report the level-by-level picture next to it.

A worked illustrative example

The figures below are illustrative only — invented for the purpose of showing the method, not drawn from any survey, company or dataset. Do not quote them as evidence of anything.

Read that table the way a good analyst would. The company-level gap of roughly 25% looks alarming, but almost none of it appears inside the comparator groups. Most of it comes from where people sit: women are concentrated in Customer Success L3, men are concentrated in Engineering L5. That is a representation and progression finding, and throwing money at it will not fix it.

Meanwhile the two genuinely actionable items are Engineering L5 and Customer Success L3. Engineering L5 has only three women, so the median is fragile — you look at those three individuals by name and understand each case. Customer Success L3 has twenty people, so an 8% gap there is worth a proper investigation into starting offers and increment history.

And Sales L4, which looks clean on fixed pay, may not be clean at all — you still have to check commission structures, territory allocation and actual payout against target.

What the simple method cannot do

Cohort medians cannot separate overlapping factors. If women in a group happen to have shorter tenure on average, a median comparison will show a gap that tenure explains — and you will not know that from the table. To untangle several factors at once, you need a model.

Methodology Level 2: The Regression-Based Approach

A regression estimates how much of pay is explained by each factor simultaneously, and reports what is left over. In practice you regress the logarithm of pay on a set of explanatory variables plus a gender indicator; the coefficient on the gender indicator is your estimate of the adjusted gap.

You can do this in R, Python, or even a well-built spreadsheet with a statistics add-in. The technique is standard. The judgement is entirely in which variables you include.

Variables it is reasonable to include

• Job family or function — different markets, different rates.
• Level or grade — the strongest single predictor in a well-built architecture.
• Location band — if your differential policy is written and consistently applied.
• Company tenure and role tenure — and consider a squared term, since the pay-tenure relationship usually flattens out.
• Relevant prior experience — if you record it reliably, which most companies do not.
• Employment type — full-time, part-time, fixed-term.
• People-management responsibility — genuine span, not just a title.
• Performance rating history — with the caveats above, and reported both with and without.

Variables you must not include

This is the trap, and it is subtle enough that well-intentioned teams fall into it regularly.

Prior salary. If you control for what someone earned at their last employer, and their last employer underpaid them for reasons connected to gender, you have imported that bias and declared it explained. Your adjusted gap will shrink beautifully and mean nothing. Never put prior salary on the right-hand side.

Negotiated starting pay or offer premium. Same logic. Negotiation outcomes are shaped by who negotiates, who is expected to negotiate, and who is penalised for negotiating. "He asked for more" is a description of the problem, not a justification.

Current pay or CTC band. Obviously circular, but it sneaks in via variables like "salary grade" when grades were assigned based on pay rather than on work.

Department when department is a gender proxy. If a function is 90% one gender, a department control can absorb exactly the effect you are trying to measure. Use job family defined by work content instead, and look at cross-function differences separately.

Anything assigned after the pay data was visible. Levels, job families and comparator groups must be frozen first. Otherwise the model is fitting your own post-hoc rationalisations.

The general test: would this variable be acceptable as a written reason on a pay decision memo? "Because his previous employer paid him more" fails that test. "Because she manages a team of twelve and he manages none" passes it.

Practical cautions on the regression

• Sample size. Under roughly 100 employees, a whole-company regression is fragile. Use it as a cross-check on your cohort work, not as the headline.
• Use log of pay so coefficients read as approximate percentage effects and the distribution behaves better.
• Report the confidence interval, not just the point estimate. A gap of 3% with an interval spanning zero to seven says something quite different from a tight 3%.
• Look at the residuals for individuals. The model's real value is identifying specific people whose pay the model cannot explain — those are your remediation candidates.
• Run it separately by function where you have the numbers. A company-wide coefficient can hide one badly affected function inside an otherwise clean population.
• Watch multicollinearity. Level, tenure and title move together; if they are too tangled, the model will produce unstable coefficients that swing wildly when you add or drop a variable.

Setting a Materiality Threshold

At some point someone will ask: how big does a gap have to be before we do something about it? You need an answer agreed in advance, in writing, before the numbers arrive.

Without a pre-agreed threshold, two bad things happen. Small gaps get dismissed as noise whenever fixing them would be expensive, and the definition of "material" drifts to wherever the budget is comfortable.

A workable threshold has three parts:

1. A statistical component. Is the difference distinguishable from random variation given the group size? A three-person group tells you very little; a thirty-person group tells you a lot.
2. A size component. A gap below a low single-digit percentage is usually within the noise of ordinary compensation administration. Above that, it needs a reason.
3. A pattern component. A gap that appears in the same direction, in the same function, across two or three consecutive audits is meaningful even if each year individually looks small. Consistency of direction is a signal in its own right.

Be careful about the framing though. A statistical threshold tells you whether a group-level pattern is real. It does not tell you that a specific individual is fairly paid. If one named person is 15% below every comparable colleague and nobody can explain why, you fix that regardless of whether the group-level coefficient reaches significance.

So run two tracks in parallel: a group-level track governed by the threshold, and an individual-outlier track governed by "can we write down a defensible reason for this person's pay?" The second track usually produces more corrections than the first, and they are cheaper.

Interpreting Results Honestly — and the p-Hacking Trap

Here is the uncomfortable part. Every pay equity audit creates an incentive to find a small number, because a small number means a small remediation bill and a comfortable board meeting. That incentive is quiet, and it works on well-intentioned people without their noticing.

The technical name for what it produces is p-hacking: trying specifications until one gives you the answer you wanted. In pay equity work it takes recognisable forms.

• Regrouping comparators until gaps disappear. Splitting a group with a gap into two smaller groups, each now too small to be significant. Nothing has changed except the arithmetic.
• Adding controls until the coefficient shrinks. Especially controls that encode the bias — prior salary, negotiated premium, or a "criticality" score invented after the fact.
• Excluding inconvenient people. Dropping the outliers, the new joiners, the acquired population, the one function with a problem — each exclusion individually defensible, collectively fatal.
• Switching pay measures. Running fixed pay when total comp looks worse, or vice versa, and only reporting the flattering one.
• Switching statistic. Moving between mean and median depending on which is kinder this quarter.
• Rounding the threshold. Setting materiality at exactly the level that puts your largest gap just outside it.

Protections that actually work:

• Pre-register the method. Write the specification, comparator rules, exclusions and threshold down, get them signed off, and date the document — before the data is analysed.
• Report every specification you ran. Not just the final one. If you ran it five ways, show the five results and explain why the headline one was chosen.
• Separate the roles. The person who runs the analysis should not be the person who owns the remediation budget.
• Use an independent reviewer. External counsel, an external consultant, or at minimum someone from a different function who will ask awkward questions.
• Keep every version. If the number changed between draft three and draft seven, you should be able to explain exactly why.

An honest audit that reports an uncomfortable gap and a credible plan is worth far more than a flattering audit that quietly falls apart the first time anyone examines it.

Building the Remediation Plan

You now have a list of people whose pay you cannot explain. Turning that into action is a project with money, sequencing, legal and communications dimensions.

Costing it

Build a schedule of every proposed adjustment with the current pay, the proposed pay, the increase in absolute and percentage terms, and the rationale. Then compute:

• Total annual cost of the adjustments, and cost as a percentage of total payroll. In most companies that run a serious audit for the first time, this number is smaller than the leadership feared — the corrections concentrate in a modest number of individuals.
• Flow-through costs. In India this matters more than people expect: statutory contributions, gratuity accrual, bonus calculations and variable pay all move when fixed pay moves. Work the fully loaded number with finance and payroll, not the headline increase.
• Second-order effects. If you lift a cohort, you may compress the differential with the level above and create a new problem. Model the post-correction distribution before you commit.

Sequencing

There are two philosophies and both are legitimate.

Correct in one go. Fix everything at a single effective date. It is clean, it is fast, it removes the ongoing exposure, and it signals seriousness. It costs more upfront and is harder to absorb in a tight year.

Correct over cycles. Spread corrections across two or three increment cycles, largest gaps first. It is easier to fund and easier to fold into normal process. The risk is that the plan gets quietly abandoned when the next budget pressure arrives, leaving you with documented knowledge of a gap and no fix — which is a worse position than not knowing.

If you phase it, protect the plan: a board-approved schedule, a named owner, a ring-fenced budget line, and a documented completion date. Discuss the choice with counsel, because the record of what you knew and when you knew it matters.

Whichever route you take, one principle is non-negotiable: you adjust upward, never downward. Reducing someone's pay to close a gap is a legal, contractual and cultural disaster. Correct by raising the underpaid, and control the overpaid through slower future increments if the gap is genuinely indefensible.

Communicating a correction to an individual

This conversation carries real risk, and the wording should be reviewed by employment counsel before any manager has it. General principles that usually apply:

• Keep it forward-looking and factual. Something like: "We reviewed pay across comparable roles and are adjusting your salary to better align with the range for your level, effective from this date."
• Do not characterise the past. Managers should not say "you were underpaid", "this is to make up for a mistake", or "we should have done this years ago". These are admissions, and they are usually not even accurate — most gaps come from accumulated process drift rather than a decision anyone made.
• Do not promise back pay or imply it is owed unless counsel has specifically advised that route.
• Brief managers with a script and a Q&A, and tell them exactly what to do when asked a question outside it: escalate, do not improvise.
• Do not link it explicitly to gender. Saying "we adjusted your pay because you're a woman" is both legally risky and, for many recipients, insulting.
• Do not make it conditional. No waivers, no release documents attached to a pay correction, unless counsel is driving that process for specific reasons.
• Document the decision internally with the rationale and the approval trail, in the file that is meant to hold it.

Handle the individual whose pay is high relative to peers separately and gently. Usually you do nothing immediate and let future increments bring the structure into line. You do not reduce pay, and you certainly do not tell them they are overpaid.

Fixing the Upstream Causes

If you remediate without changing the processes that produced the gap, you will find the same gap again in three years — having spent the money twice. The upstream fixes are where the durable value is.

Stop asking for salary history

Basing an offer on what someone earned before is the single most reliable mechanism for carrying historical underpayment from employer to employer. If a candidate was underpaid at their last job for reasons connected to gender, and you set your offer as a percentage uplift on that number, you have adopted someone else's bias as your own compensation policy.

Practical steps:

• Remove current CTC and expected CTC fields from your application form and your ATS.
• Train recruiters to make offers against the internal range for the level, not against the candidate's current number.
• Where you must discuss expectations, anchor on your range first: "This level pays between X and Y — does that work for you?"
• Tell your recruitment vendors this is the policy. Agencies collect current CTC by habit and will keep feeding it to you unless you stop them.

This is worth doing regardless of what the law requires, and the practice is disappearing in market after market. Verify your own obligations, but do not wait for a rule.

Build real pay ranges and use range penetration

For each level in each job family, define a minimum, midpoint and maximum. Anchor the midpoint on market data for the level and the location band, and set the width wide enough to allow for genuine experience differences but narrow enough to actually constrain decisions.

Then measure range penetration — where each employee sits between the minimum and maximum, expressed as a percentage. This is the most useful single equity metric available to a mid-size HR team. Comparing penetration by gender within each level strips out the level-mix effect automatically and points straight at the decisions that need examining.

If your women cluster at 40% penetration and your men at 60% within the same level, you have a finding, and you did not need a regression to see it.

Put guardrails on offers

Most inequity is created at the moment of hire, not during increments. Increments compound whatever the starting number was.

• Default to the range midpoint for a fully qualified candidate, with a written justification required for anything above.
• Set an approval threshold. Offers above a defined penetration point need a second approver — the compensation lead or the function head.
• Review offers against the incumbents. Before a new joiner's offer is signed, check it against existing employees at the same level in the same location. New hires being paid above tenured internal staff is the most common and most corrosive source of drift.
• Cap the negotiation premium. Decide the maximum you will move from the initial offer, and apply it consistently rather than case by case.
• Log the reason for every off-midpoint offer. That log is your audit trail next year.

Put guardrails on increments and calibrate

• Run increment cycles against a matrix that maps performance rating and range penetration to a recommended increase. Someone high in the range gets a smaller percentage than someone low in the range at the same rating, which naturally compresses drift.
• Calibrate ratings across managers before increments are decided, not after. Uncalibrated ratings are the main channel by which manager bias enters pay.
• Run an equity check on the proposed increments before they are released. Pull the recommended increases by gender within each level. If the pattern is off, you can fix it during the cycle at near-zero incremental cost — this is by far the cheapest remediation you will ever do.
• Track promotion increases separately. Promotion is often where the largest single jumps happen and where the least governance exists.

Make promotion equitable

Pay equity at a point in time is a snapshot of years of progression decisions. Track, by gender and level:

• Promotion rate — proportion of each group promoted per cycle.
• Time in level before promotion.
• Proportion of each group in the promotion nomination pool versus the eligible population.
• Outcome rate — of those nominated, who actually gets through.

If women are nominated at the same rate but promoted less often, the problem is in the committee. If they are nominated less often, the problem is with managers and the nomination criteria. Those are different fixes.

Deal with parental leave and career breaks

Career interruptions have a measurable long-term effect on pay, and in India they fall overwhelmingly on women. Things you can decide as policy:

• Do not penalise the review cycle. Someone on leave for part of the year should be rated on the period worked, not implicitly marked down for absence.
• Make sure they receive the increment cycle rather than being skipped because they were not present when decisions were made.
• Structure returns properly — phased return, a real role rather than a holding pattern, and a scope conversation within the first month back.
• Track pay trajectory before and after leave as a standing metric. If people's increment percentages drop systematically in the two cycles after a return, you have found something important.
• Watch the "mummy track" effect, where returners are quietly moved into lower-visibility work that then justifies slower progression.

The Representation Dimension

Most of the money in a pay gap is not in what people are paid; it is in which jobs they hold. Any audit that ignores representation will keep producing the same unadjusted number year after year while reporting a clean adjusted gap, and nobody will understand why.

Analyse, by gender:

• Distribution by level. The classic pyramid. What proportion of each level is female, from entry to leadership?
• Distribution by function, especially between higher-paying technical and revenue functions and lower-paying support functions.
• The hiring pipeline — applications, shortlists, interviews, offers, acceptances. Where does the proportion drop? A pipeline that is 40% women at application and 15% at offer has a selection problem, not a supply problem.
• Attrition by level and tenure band. Mid-career attrition among women is a common and expensive pattern; exit-interview data rarely explains it honestly.
• Access to high-value work — who gets the strategic projects, the client-facing roles, the P&L responsibility. This determines who is promotable in three years.

Set representation targets the way you set any other business target: a baseline, an aspiration, an owner, a timeframe, and a review cadence. Take advice on how you frame and implement such targets, because the line between broadening a pipeline and making a decision on a prohibited ground is one to walk carefully and with counsel.

Privilege, Confidentiality and Data Protection

You are about to handle the most sensitive dataset in the company. Treat it that way from day one, not after the first leak.

Legal privilege

In many jurisdictions, conducting an audit under the direction of legal counsel can help protect the working papers from disclosure. Whether and how that applies to your situation in India is a question for your own counsel. What is universally true is that the decision has to be made before you start — you cannot retrospectively make an analysis privileged after it has circulated on email to fifteen people.

Ask counsel early: who should commission the audit, who should receive the outputs, how should documents be labelled, and what should go in email versus a controlled document.

Access control

• Keep the analysis dataset in one controlled location with a named, minimal access list.
• Use a pseudonymised working file — employee ID, gender, level, function, location, pay — for the analysis itself. Names are only needed at the remediation stage.
• Do not email spreadsheets of individual salaries. Every forwarded attachment is a permanent uncontrolled copy.
• Managers see their own team's remediation outcomes, not the company dataset.
• Log who accessed what and when.
• Agree a retention period and actually delete the intermediate files when it expires.

DPDP considerations, kept general

India's data protection framework imposes obligations on how personal data is collected, processed, stored and shared, including around notice, purpose limitation, security safeguards and handling by any processor acting on your behalf. Salary data, performance data and any leave or health-adjacent information sit at the sensitive end of what you hold.

Practical hygiene, which you should confirm against current requirements with your counsel and data protection lead:

• Be clear about the purpose for which you are processing the data and do not quietly reuse the dataset for something else.
• Collect only what the analysis needs. Curiosity is not a purpose.
• If an external consultant touches the data, put the processing terms, security obligations and deletion requirements in a written contract before any data moves.
• Check where the data will be stored and processed, particularly if your HR system or your consultant is outside India.
• Have a plan for employee queries about their own data, since individuals have rights over it.
• Minimise and delete. Keep the final report and the method; do not keep seventeen intermediate spreadsheets on someone's laptop.

The reputational damage from leaking a salary file exceeds the damage from almost any gap you might find in it.

From Audit to Transparency, in Stages

Transparency is not a switch. Companies that publish ranges before they have defensible ranges create more problems than they solve. The sequence that works is: get your structure right, get your data right, fix what is broken, then open up — in stages, each one internal before it is external.

Three rules for moving between stages.

Never publish a range you will routinely exceed. If every offer lands above the posted maximum, you have not published a range, you have published a fiction, and your existing employees will notice first.

Fix before you publish. Stage 3 exposes every unexplained difference to the affected person immediately. Do the remediation first.

Move one stage at a time and let it settle. Each stage generates questions. Answer them before adding more disclosure.

What to Tell Employees — and What Not To

The communication strategy is a real decision, not an afterthought. Get it wrong and a genuine good-faith exercise reads as either a cover-up or an admission.

Reasonable to share

• That the audit happened and why. "We review pay across comparable roles annually to make sure our decisions are consistent and defensible."
• The method, in plain language. What you compared, what factors you accounted for, who reviewed it.
• What you are changing — the process improvements, the guardrails, the range structure. Process changes are safe to communicate and demonstrate seriousness.
• The overall commitment, including the cadence of future reviews.
• Aggregate representation data, if you are prepared to discuss what you are doing about it.

Handle with care or not at all

• Individual results for anyone other than the person concerned. Never.
• Precise gap figures, until you have taken advice and have a plan attached. A number without a plan is an accusation you have handed to everyone.
• Language that characterises the past — "we discovered we had been underpaying", "historic discrimination". Take counsel's wording.
• Promises you cannot keep — "we will close the gap by next year" is a hostage to fortune if your remediation depends on future budgets.
• Comparisons between named teams or managers. That is a blame exercise, not a fairness exercise.

The general pattern that works: be transparent about process and commitment, careful about numbers, and legally reviewed on anything retrospective. Employees respond well to "here is how we make pay decisions and here is how we check ourselves". They respond badly to silence, and badly to a number with no follow-through.

Running It Annually and the Metrics to Track

A one-off audit is a project. A repeated audit is a control. The second is far more valuable, and much cheaper after the first year because the data plumbing is already built.

A sensible annual rhythm for an Indian company running an April or July increment cycle:

• Two to three months before the cycle: refresh the employee master, re-map any new or changed roles, refresh range midpoints against market data.
• One to two months before: run the analysis, review findings with counsel, build the remediation list.
• During the cycle: fold corrections into the increment process wherever possible — it is cheaper, quieter and easier to explain than a separate off-cycle exercise.
• Immediately before release: run the pre-release equity check on proposed increments and promotions, and fix anything that has drifted.
• After the cycle: report to the board or leadership on the trend, not just the level.

Metrics worth putting on a standing dashboard

• Unadjusted median gap, company-wide and by function.
• Adjusted gap with confidence interval, reported both with and without performance controls.
• Number and cost of corrections made, cumulative.
• Range penetration by gender, by level.
• Proportion of offers made above range midpoint, split by gender.
• Proportion of offers requiring exception approval.
• Representation by level, with year-on-year movement.
• Promotion rate and time-in-level by gender.
• Increment percentage by gender within rating band.
• Voluntary attrition by gender and level.
• Percentage of employees mapped to a current, reviewed level.

Watch the trend and the direction of the residual, not just the headline. A gap that is stable at a modest level with a clean process is a healthier position than a gap that fell sharply because of one large correction and is already creeping back.

Readiness Checklist

Before you start, honestly score yourself on these. If more than three are missing, spend the first phase building foundations rather than running analysis.

• [ ] A single source of truth for current fixed pay across all entities
• [ ] Variable pay target and actual payout recorded per employee
• [ ] Equity grants accessible to the audit team
• [ ] Joining and retention bonuses recorded somewhere other than offer letter PDFs
• [ ] A job architecture with defined families and levels
• [ ] Every employee mapped to a level, reviewed by the function head
• [ ] Levels defined by work content, not by salary band
• [ ] Gender recorded consistently for the whole population
• [ ] Date of joining and date in current role both available
• [ ] Work location recorded accurately, including for remote employees
• [ ] Performance ratings for at least the last two cycles, in comparable form
• [ ] A written location differential policy
• [ ] Pay ranges defined per level, or a plan to define them
• [ ] Counsel engaged on privilege, scope and communications
• [ ] Data protection review completed on the dataset and any external processor
• [ ] Executive sponsor named and a remediation budget conversation started
• [ ] Method, comparator rules, exclusions and materiality threshold documented and signed off before analysis

A 90-Day Plan

Days 1-30: Foundations and scope

• Secure an executive sponsor and agree the objective in one sentence.
• Engage employment counsel on privilege, scope and communication constraints.
• Complete the data protection review; set up the controlled workspace and access list.
• Decide entities, populations and pay elements in scope; document exclusions with reasons.
• Audit the job architecture. If levels do not exist or are inconsistent, this becomes the month's main work.
• Pull the raw data and build the first draft employee master; log every quality problem you hit.
• Write and sign off the method document: comparator rules, variables, materiality threshold, specifications you will run.

Days 31-60: Data and analysis

• Clean the employee master. Reconcile headcount against payroll to the individual.
• Finalise and freeze level mappings with function heads — before anyone sees pay results.
• Run the cohort analysis: medians and range penetration by comparator group.
• Run the regression if your numbers support it; run the pre-agreed alternative specifications too.
• Test performance ratings for their own distributional issues.
• Pull representation, promotion and pipeline data.
• Draft findings with counsel. Build the individual remediation list with rationale per person.

Days 61-90: Decisions and action

• Cost the remediation fully loaded, including statutory flow-through.
• Take the findings and options to the sponsor and leadership; agree scope, budget and sequencing.
• Get communications reviewed by counsel; build the manager script and Q&A.
• Execute the first tranche of corrections, ideally inside the normal cycle.
• Agree the upstream fixes with owners and dates: salary-history removal, offer guardrails, increment matrix, calibration, promotion tracking.
• Set the annual cadence and the standing dashboard.
• Archive the method, the dataset and the decision log properly; delete intermediate files per your retention rule.

What Your HR Software Should Do Here

Most of the pain in a first pay equity audit is not analytical. It is archaeological — reconstructing salary history from offer letters, increment spreadsheets and half-remembered conversations. That is a systems problem, and it is fixable.

An HRMS that actually supports equity work should provide:

• A real job architecture: job families and levels as first-class objects, with every employee mapped and the mapping history retained.
• Pay ranges attached to levels, with automatic range penetration calculation per employee.
• A complete change audit trail on pay: every revision with effective date, old value, new value, reason code, who requested it and who approved it. This is the single highest-value feature for audit work, and the one most often missing.
• All pay elements in one place — fixed, variable target and actual, joining and retention bonuses, allowances, and ideally equity grant references.
• Offer guardrails in the recruitment flow: range validation, midpoint defaults, approval routing for exceptions, and no salary-history field.
• An increment workflow with a matrix, calibration support, and an equity check before release.
• Role-based access and access logging on compensation data, so the controlled dataset is genuinely controlled.
• Reporting that slices pay by gender, level, function and location without exporting everything to a spreadsheet.

If your system does these things, the audit becomes a quarterly report rather than an annual crisis. If it does not, the first audit will cost you a month of someone's life before any analysis begins.

Frequently Asked Questions

How often should we run a pay equity audit?

Annually, timed to land shortly before your main increment cycle so corrections can be folded into normal process. Some companies add a lighter mid-year check on new hires and off-cycle changes, which is where drift appears fastest. The first audit is the expensive one; once the data pipeline and job architecture exist, subsequent runs are a fraction of the effort.

Can we run a pay equity audit with fewer than 100 employees?

Yes, though the method changes. Regression needs sample size, so below roughly 100 people you rely on cohort comparisons, range penetration and individual review — essentially asking, for each person, whether you can write a defensible reason for their pay relative to comparable colleagues. That question is answerable at any headcount, and doing it early is far easier than untangling years of drift later.

Should we use an external consultant?

It depends on your internal capability and your risk posture. External support helps where you lack analytical skills, where you want independence in the findings, and where counsel advises that an external exercise better supports the confidentiality position. If you do engage one, put the data protection terms, security obligations and deletion requirements in writing before any data moves, and insist that the method is documented well enough for you to reproduce it yourself next year.

What if we find a gap we cannot afford to fix immediately?

Document the finding, build a costed multi-cycle plan with a named owner and a completion date, get it approved at board or leadership level, and start with the largest and least defensible gaps. Discuss this with counsel — knowing about a gap and having no plan is a materially different position from knowing about it and executing a funded plan. What you must not do is quietly shelve it and hope the numbers move on their own.

Is it legal to pay two people in the same role differently in India?

Differences based on legitimate, job-related factors — experience, sustained performance, scope, location under a consistent policy — are generally defensible. Differences based on gender, for the same work or work of a similar nature, run against the equal-remuneration principle carried into Indian wage legislation. The test is whether you can articulate and evidence a job-related reason. Verify the current position and its application to your establishments with employment counsel rather than relying on a general description.

Should we stop asking candidates for their current salary?

As a matter of practice, yes. Anchoring offers on prior pay imports other employers' pay decisions, including their inequities, into your structure, and it is the most reliable way to recreate gaps you have just spent money fixing. Make offers against your internal range for the level instead. Check your own legal obligations, but this is worth doing on the merits regardless of what is mandated.

What is the difference between a pay equity audit and pay transparency?

The audit is the diagnostic — a private internal analysis of whether your pay is defensible. Transparency is the disclosure — publishing your philosophy, levels, ranges or outcomes to employees and candidates. The audit should come first, always. Publishing ranges before you know whether your existing pay fits inside them exposes every unexplained difference simultaneously, with no plan attached.

Conclusion

A pay equity audit is not primarily a compliance exercise, though it has compliance value. It is a test of whether your compensation decisions are the product of a system or the product of accumulated improvisation. Almost every company that runs one honestly finds a mixture of both, and the useful output is rarely a single number — it is the list of specific decisions, guardrails and structures that were missing.

Start with the unglamorous parts. Build the job architecture. Clean the employee master. Freeze the method before you look at results, and write down your materiality threshold while you still have no idea what the numbers will say. Involve counsel early on privilege, scope and communications, and involve them again before you remediate or communicate anything. Report the unadjusted gap and the adjusted gap side by side, because they describe two different problems with two different fixes. Then spend most of your energy upstream — on offers, increments, calibration and promotion — because that is what determines whether you have to do this again in three years or merely refresh a report.

The external pressure is only going one direction. Transparency requirements keep spreading in other markets, investors and candidates keep asking sharper questions, and India's post-labour-code compensation restructuring has already put every salary structure on the table. Employers who get this right quietly, ahead of any domestic mandate, will be answering questions from a position of evidence. The rest will be improvising under scrutiny.

If the biggest obstacle is that your job levels live in one spreadsheet, your ranges in another and your pay history nowhere at all, that is the part CozyHR is built to remove. Job architecture, pay ranges with range penetration, and a full audit trail on every pay change sit in one system, so equity analysis becomes a report you run rather than a month of spreadsheet archaeology. Try CozyHR and see what your pay data looks like when it is all in one place.