In this guide
Find out which team it is actually happening in, change the shape of that team’s week, and build something that tells you next time before the resignation rather than after.
A 90-day plan for a team of 20 to 200, written for the case where you cannot simply pay more, which is most cases, most of the time.
The short answer
Attrition in a small company is almost never evenly spread. A company at 14% usually has three or four departments in single digits and one at thirty, and the company-wide number exists mainly to hide that.
So the first job is not a retention strategy. It is arithmetic: work out where it is concentrated, and whether the people leaving are the ones you wanted to keep. Most of what gets spent on retention is spent company-wide on a problem that lives in one team.
Why the raise does not hold
The counter-offer is the most common response and the least durable. It works, briefly, and then the thing that made them look elsewhere is still there and now costs you more.
This is not an argument that pay does not matter. If you are paying meaningfully below market, nothing on this page will save you and you should fix that first. But above that floor, people rarely leave a job they find sustainable for a 15% increase, and they frequently leave a job they find unsustainable for nothing at all.
Gallup’s 2019 research put the cost of replacing someone at one-half to two times their annual salary. On a ₹18 lakh engineer that is somewhere between ₹9 lakh and ₹36 lakh in recruiter fees, interview time, the six months before a replacement is useful, and the work that did not happen in between. Measured against that, almost everything below is cheap.
Days 1 to 30: find the real number
Four things, none of which needs a budget.
- Calculate it properly, by team. Last twelve months, average headcount, voluntary exits only. The formula and calculator take twenty minutes. Do it per department for any department with enough people that the number means something.
- Mark every exit regretted or not. Ask the manager who lost them, not HR, and ask about exits going back a year. You are looking for the subset you actually care about, which is usually between a third and a half.
- Bucket by tenure. Under 90 days, under a year, over a year. If a lot of it is under 90 days, your problem is in hiring and onboarding and the rest of this plan is aimed at the wrong thing.
- Ask eight people what their week is like. Not a survey. Eight half-hour conversations, spread across the worst department and a good one for contrast. You are listening for the same sentence said three different ways.
At the end of thirty days you should be able to finish this sentence with specifics: most of the people we did not want to lose came from this team, at around this tenure, and when I ask people there what their week is like they all mention this.
Days 31 to 60: change the load
Now change something, visibly, in the team you identified. Not a policy. A condition.
- Audit the calendar before anything else. Add up the hours in meetings for the five busiest people. If someone is in meetings for more than half their week, they are not doing the job they were hired for, and no amount of wellbeing budget fixes that. Cut the recurring ones first; they are the ones nobody reconsiders.
- Find the fragmented weeks. Twenty hours of meetings spread across every afternoon is far worse than twenty hours in two days. Protect two uninterrupted days a week for the people doing deep work and defend it yourself for a month, or it will not hold.
- Look at who is carrying on-call and escalations. It is almost always the same two reliable people, and they are usually the two you can least afford to lose.
- Give the managers one specific thing to do. Most first-time managers in startups were promoted for being excellent at the work. “Have a real one-to-one every fortnight, and the agenda is theirs not yours” is more useful than a training budget.
Say out loud what you changed and why. The change matters; the fact that it visibly followed from people saying something matters more.
A note on sequencing, because this is where budget usually goes first and should not. A stress management session is worth running, but after you have fixed the calendar, not instead of it. Teaching people to cope with a week you could have changed is the most expensive way to be seen to care.
Days 61 to 90: build the loop
The first sixty days fix the team you already know about. The last thirty are about not being surprised again.
- Start asking, weekly and anonymously. Something short enough that people actually answer. The absolute level matters much less than the direction over six weeks, and the response rate itself is a signal: people stop answering before they stop caring.
- Put stay interviews in the calendar. Twice a year per person, run by their manager, not attached to the appraisal cycle. The questions that get an honest answer are on the next page over.
- Make help private and easy. If using the counsellor means telling your manager, it will not be used. If booking takes more than two minutes, it will not be used.
- Review it monthly, with the same four numbers. Voluntary LTM by team, regretted share, tenure buckets, and whatever weekly measure you started. Thirty minutes, same slot.
If you would rather this ran itself, that is what Wanderfly does: team scores from check-ins, the monthly PSS-10 and calendar load, forecasts at 14, 30 and 60 days, and private expert sessions paid for by the company. If you would rather do it with a form and a recurring invite, that genuinely works too for a while.
The three things that move it most
Across the companies we work with, the same three come up far more often than anything else.
- The shape of the week. Not total hours: fragmentation and unpredictability. People sustain hard work. They do not sustain a week they cannot plan.
- The direct manager. Nothing else in a small company has the same effect on whether a person stays, and nothing else gets as little deliberate attention.
- Whether saying something changes anything. The belief that raising a problem leads to a change is the thing that keeps people raising problems. Once it goes, everything looks fine right up until the resignation.
What to stop spending money on
Being plain about this, as a company that sells into this budget.
- One-off engagement events. An offsite does not change the Monday after it. People enjoy them; they do not stay for them.
- Benefits with low usage. Pull the usage numbers for anything you are paying per-seat for. Something used by 4% of staff is a line item, not a programme.
- Recognition, as a platform purchase. Recognition itself is not the problem. Buying software to administer it usually is, because the thing that works is specific and timely and comes from the person you report to. If you are going to run a programme, our piece on making recognition land covers what separates the ones people believe from the ones they ignore.
- The annual survey, if nothing follows it. It is worse than nothing, because it asks people to describe a problem and then teaches them that describing it achieves nothing.
- Activity monitoring, in any form. It is sold as visibility. It reliably reduces discretionary effort and trust, which is precisely the thing you are trying to protect. More on that in detection versus monitoring.
If you run a contact centre or a services team
Everything above still applies, with three adjustments, because the rhythm is different.
- Report monthly and annualise, but split early exits out. A large share of contact-centre attrition happens inside 90 days. Mixed into one figure it swamps everything and moves for reasons the floor cannot control.
- The roster is the retention lever. Shift predictability, how far ahead rosters are published, and how often someone is moved between shifts will beat almost any engagement initiative.
- Watch the team-leader layer hardest. Spans of twenty or more, promoted from the floor, usually with no training. When a team leader goes, a visible share of their team tends to follow within two quarters.
One adjacent thing worth getting right at this size: the statutory floor underneath all of it. Late full-and-final settlements and unclear leave policy generate a surprising share of the bad feeling that shows up later as attrition, and they are entirely fixable. Our India statutory compliance checklist goes through it Act by Act.
What this will not fix
If you are paying well below market, if the company is in trouble and people can tell, if there is a senior person everyone is quietly leaving because of, or if the work itself has stopped being interesting. None of this reaches that. Those are real causes and they need real answers, not a measurement programme.
And some attrition should happen. A company where nobody leaves voluntarily is usually one that has stopped being honest about performance, or has made leaving feel impossible. The goal was never zero. It is to stop losing the people you did not want to lose, for reasons you could have seen.
If you want the argument for why burnout data is the most practical early measure a small company has, and where that argument stops, it is in burnout and attrition. The wider picture is on employee retention.
Questions people ask
How do you reduce attrition without increasing salaries?
By removing the things that make the job harder than it needs to be, which is usually workload shape rather than workload volume: too many meetings, unclear ownership, an on-call rota that falls on the same people, and managers who have never been trained to manage. A counter-offer buys a quarter. Fixing the week the person is actually living buys longer, and costs less.
What causes high attrition in Indian startups?
In our experience the pattern is rarely a single cause. It is usually a small number of teams carrying disproportionate load, a manager layer promoted for being good at the craft rather than at managing, and no mechanism for anyone to say any of this before they resign. Pay is often cited at the exit interview because it is the safest thing to say.
How long does it take to reduce attrition?
You will see leading measures move in four to eight weeks if you change something real. The attrition rate itself takes two to three quarters, because it is calculated over the last twelve months and because people already serving a 60- or 90-day notice are still in the numerator. Judge the first quarter on what changed, not on the rate.
Does an employee wellbeing programme reduce attrition?
Only the parts that change working conditions or get people real help. Benefits nobody uses do not, and a platform nobody acts on does not either. What helps is measuring often enough to see which team is struggling, acting on it within a fortnight, and making expert support genuinely private so people use it. The measurement is worth nothing without the second part.