How to measure early attrition in the first 90 days
New-hire departures are usually visible weeks before they happen. This draft sets out a measurement plan, but the research base it should rest on is still missing.
Early attrition is the cost most companies notice late and measure least. A new hire who leaves in the first three months takes recruiting effort, manager time and weeks of output with them, and the exit interview seldom explains why. This explainer sets out how to design a first-90-day measurement plan. It is a working draft. None of the sources we checked contains onboarding or new-hire turnover research, so the reasoning below is practitioner judgement. It should be tested against published studies before any board sees it.
How it works
Treat the first 90 days as a series of checkpoints rather than a single event. A new hire forms a view of the role, the manager and the team early, and much of that view is set in the first few weeks. Does the laptop arrive working? Is the first task real work or a training exercise? Does anyone explain the unwritten rules about who decides what? When the answers are poor, the person usually notices within days, even if they say nothing for months.
The useful distinction is between signals and outcomes. A resignation is an outcome, and it arrives too late to be useful. Signals come before it. A training milestone is missed. A new starter stops asking questions in meetings. A manager cannot name the first three things the person is working on. Signals are harder to read than resignations, but they can be acted on while the person is still deciding whether to stay. A measurement plan should record them at fixed points so that patterns show up across cohorts, not only in one painful case.
Keep the signals few. A plan with twenty measures will be filled in carelessly by busy managers, and the data will be useless by the second quarter. Choose a handful that a manager can answer in a few minutes and that someone outside the team can check. Each measure should answer a single question: is the person able to do the job, do they know who to ask, and do they expect to be here in six months? If a measure does not answer one of those, leave it out.
Who it affects
Every organisation that hires in volume feels this problem, but the cost falls unevenly. A business with three sites and one people lead may have no single person who sees all the new starters at once, so a pattern in one plant stays invisible to head office. A small firm may lose a hire and never learn whether the role, the manager or the pay was the cause, because one departure is too few to read. Larger organisations can see the pattern and still miss it, because the people who run onboarding rarely own the reasons people leave. Those reasons sit with line managers, who are measured on output.
Founders and chief executives in small companies often carry the onboarding themselves, which makes the signals easy to read but hard to record. The risk there is different. The person who notices a problem is also the person who is too busy to write it down, and the company learns about it when the hire hands in notice. For a company of that size, a short written log kept by one named person may be the whole plan.
What leaders get wrong
- They measure satisfaction at day 90 and call it onboarding. By then the decision to leave has often been made, and the survey records the result rather than the cause.
- They treat the exit interview as the main source of evidence. Leavers are polite, and the reason they give is often the one that is easiest to say at the door.
- They count resignations without a denominator. Ten leavers from forty starters and ten leavers from ninety starters are different problems, and a bare count hides the difference.
- They set one target for every role. A sales hire and a plant engineer do not follow the same first month, and a single threshold will flag the wrong people.
- They collect the data and nobody reads it. A dashboard with no named owner and no agreed threshold is decoration, however carefully it was built.
A subtler error is to measure what is easy to count. Attendance at induction sessions is simple to record, and it tells you very little about whether a new hire can do the work. Completion of a compliance course is recorded in the learning system, and it says nothing about whether the person knows who signs off their expenses. Good measures are often awkward to collect, because they require a manager to answer a question they would rather not be asked. That awkwardness is usually a sign the measure is worth having.
What to do about it
Start by setting five checkpoints: day one, day 14, day 30, day 60 and day 90. At each one, the manager answers a short set of questions and the answers go into a shared record. Keep the questions concrete. Can the new hire name their first three deliverables? Have they had a one-to-one with their manager that was not about administration? Do they know the two people they should ask when something goes wrong? A yes or no is easier to compare across cohorts than a score out of five, and a manager can answer it honestly without feeling graded.
Assign one owner for the whole record. That person is not the manager, and it is not the recruiter. In a company of a few hundred people it is often someone in people operations with the authority to ask a manager why a checkpoint was skipped. Their job is to notice patterns: three starters in one team who all could not name a first deliverable at day 30 is a finding, and it should reach whoever owns the team's hiring plan before the next offer is made.
Set the thresholds in advance, and make them specific to the role. For each checkpoint, decide what a worrying answer looks like and what happens next. A reasonable first rule is that two or more missed answers at day 30 trigger a conversation between the manager and the owner. The number is a starting point and should be adjusted once the company has a few cohorts of data. Write down the reason for each threshold so that, when it is revised, the team can see what changed and why.
Always report departures against starters in the same cohort. Divide the number who left within 90 days by the number who started in that period, and show the result by team, site and role. Where cohorts are small, say so on the page. A single departure from a team of four is a conversation, not a trend, and a board that is shown the percentage without the count will read more into it than the data allows.
Before any of this goes into a board pack, someone should find the published studies on onboarding, socialisation and early-tenure turnover and check whether they support the signals chosen here. We have not done that work in this draft. The sources we were given contain no research on the subject, and so this article offers judgement rather than evidence. A researcher should add the relevant studies, confirm each figure at its original source, and then the plan can be tested against them. Until then, run the five checkpoints in one team as a pilot, keep the record honest, and treat the first two quarters as a test of the measures themselves.