Benchmarking a salary means comparing what you pay for a job with what other employers pay for the same job. Put that way, it sounds like a lookup: find the role in a survey, read the number, done. In practice most benchmarking errors happen before the number is read. They come from matching the wrong job, choosing the wrong comparison group, picking a percentile without a reason, or using data that was already out of date when it arrived.

Where market data comes from

In India, larger employers mostly buy compensation surveys from consulting firms such as Aon, Mercer, WTW and Korn Ferry. Participating companies submit anonymised pay data for their employees, mapped to the survey’s job catalogue and levels, and in return receive aggregated results. Smaller firms rely on cheaper sources: industry-association surveys, recruitment-firm salary guides, and self-reported data from sites such as AmbitionBox or Glassdoor.

The sources differ in quality for structural reasons. Employer-submitted surveys use payroll data, so they are accurate about what was paid, but they cover only participating companies. Self-reported sites have broad coverage but no verification and an obvious bias: people who feel underpaid or recently switched jobs are more likely to post. They are useful for spotting direction, not for setting a pay band.

Step one: match the job, not the title

Surveys describe each benchmark job in a short paragraph covering scope, typical experience and level. Matching means comparing your role’s actual content with that description. The usual rule of thumb is that a match is acceptable when roughly 70 to 80% of the content overlaps.

Titles are unreliable. A “Manager” at a 200-person product startup may own a function end to end; a “Manager” at a large IT services firm may lead a team of eight inside a well-defined process. Matching on title would compare very different jobs.

When a role straddles two survey jobs, analysts sometimes blend them. A role that is 60% data engineering and 40% database administration might be priced as 0.6 times the data engineer figure plus 0.4 times the DBA figure. With illustrative medians of ₹20 lakh and ₹15 lakh, the blended reference is ₹18 lakh. Blending is legitimate, but it should be documented and used sparingly; a role that needs blending every year may be a hybrid that the market prices differently from either parent.

Step two: pick the comparison group deliberately

The same job can have very different medians depending on the peer group selected. A cloud engineer’s pay among global capability centres in Bengaluru is not the same as among domestic mid-size IT firms across Tier 2 cities. The right group is the set of employers you actually lose people to and hire people from. Exit interview data and offer-decline reasons are the best evidence for who those competitors are.

Location, industry and company size are the usual cuts. Each cut makes the sample smaller, so there is a trade-off between relevance and reliability. Survey providers typically suppress results when too few companies or incumbents contribute, which is a sign to widen the cut rather than trust a thin number.

Step three: choose a percentile on purpose

Surveys report percentiles: the 25th (P25), median (P50), 75th (P75) and sometimes 90th. P50 means half of the matched incumbents earn less and half earn more. A company’s pay philosophy states which percentile it aims for, and it need not be one number for everyone.

  • A company may target P50 for most roles, P75 for scarce skills such as security architects, and P40 for roles with abundant supply.
  • It may target P50 on fixed pay but P75 on total cash, so that high performers who earn full bonuses end up well above market while average performers do not.

Paying at P75 everywhere is expensive and rarely necessary; paying at P25 everywhere guarantees that the best candidates go elsewhere. The point of a philosophy is to spend more where it changes outcomes.

Step four: age the data

Survey data has an effective date, often several months before publication. If the survey reflects salaries as of 1 April 2026 and your new structure applies from 1 April 2027, the market will have moved in between. Analysts “age” the data forward using an expected rate of market movement.

With a median of ₹18 lakh and an expected movement of 9% a year, ageing forward twelve months gives ₹19.62 lakh. Ageing forward nine months on a compounding basis gives about ₹19.2 lakh. The rate itself comes from salary increase projections published by the same survey providers. It differs by industry and by role family, and in hot skill markets the actual movement can far exceed the average.

Skipping this step is a common reason why a company believes it pays at market while its attrition says otherwise.

From benchmark to band

The aged reference at the target percentile usually becomes the midpoint of the pay band, and the minimum and maximum are set around it. Benchmark several jobs in each grade and use a smoothed line through them, rather than letting one unusual job dictate the whole grade.

Be clear about which pay element is being compared. Base salary, fixed cash, total cash (fixed plus actual or target bonus) and total remuneration (including long-term incentives and benefits) are different numbers. Comparing your fixed pay with the market’s total cash will make you look underpaid; comparing your target-inclusive CTC with market base pay will make you look generous. Neither is true.

Benchmarking is a judgement supported by data, not a calculation. The judgement shows up in the match, the peer group, the percentile and the ageing assumption. Writing those four choices down, every time, is what makes the result defensible when a manager or employee asks why a role pays what it does.

Written by the RP Strativa Team. Salary figures and rates in the examples are illustrative.

Leave a Reply

Trending