Employee Engagement Surveys That Actually Predict Attrition

Most engagement surveys ask the wrong question first. The right question, if you care about predicting voluntary attrition over the next 90 days, is not "are you satisfied with your job?" It is some version of "would you recommend this company as a place to work to a friend in your role?" — and even that one is overrated relative to three others.

Across a cohort of 47 mid-market Kestrel deployments analyzed in Q1 2026, four engagement survey questions accounted for 71 percent of the predictive lift on 90-day voluntary attrition. The remaining 34 questions accounted for the other 29 percent combined.

The four questions that do most of the work

From a pool of 38 commonly-asked engagement questions across the surveys our customers run (Culture Amp, Lattice, Glint, custom Qualtrics builds, and a handful of internal tools), four consistently showed the strongest correlation with 90-day voluntary exits when controlled for tenure, role family, and manager quality:

1. "My manager and I have meaningful conversations about my career growth."

The strongest single predictor across our customer base, with a standardized coefficient of 0.34 against 90-day exit. Employees who responded below 4 on a 5-point scale were 2.8 times more likely to leave voluntarily in the next quarter than those who responded 4 or 5.

2. "I have a clear understanding of how my work contributes to the company's goals."

Often dismissed as a soft alignment question, this one had the second-largest effect size in our analysis. The interpretation that fit our data best is that this question is a proxy for whether the employee believes the work itself is valuable, not just whether they understand the strategy.

3. "I have the resources I need to do my job effectively."

The simplest of the four. Employees who answered below 4 here were 2.1 times more likely to exit. Importantly, this question's predictive power was almost entirely concentrated in the bottom two quartiles of responders — once people scored 4 or above, further increases produced no additional predictive value.

4. "In the past month, I have considered leaving this company."

The bluntest of the four. Some HR teams resist asking this directly out of concern that it primes the response. Our data does not support that concern. Companies that asked this question saw response patterns consistent with companies that did not, and the question's predictive power was substantial — a coefficient of 0.41 against 90-day exit when asked.

What the rest of the survey is for

This is not an argument for cutting your engagement survey down to four questions. The remaining questions on a typical engagement survey do useful work — they help diagnose why the predictive signals are flashing, they surface team-level cultural issues, and they build a longitudinal record that lets HR leaders track sentiment over time.

The argument is narrower: if your goal is to predict and act on attrition risk at the team level in the next 90 days, four questions get you most of the way there. Everything else is supporting work.

How we ran the analysis

We pulled engagement survey data and 90-day post-survey attrition outcomes from 47 Kestrel customer deployments active in 2025, totaling 312,000 individual survey responses tied to 11,400 voluntary exits. Standardized question scores were modeled against binary 90-day exit outcomes using a logistic regression with fixed effects for company, role family, tenure bucket, and manager.

The four questions above retained statistical significance (p < 0.001) across every cut of the data we ran. The next strongest question (a variant of "I trust senior leadership to make good decisions") had a meaningful effect at the company level but did not retain significance once company fixed effects were applied — suggesting it tells you about the company, not the individual's exit risk.

Practical implications for HR teams

If you are building or revising an engagement survey for 2026, three concrete recommendations:

1. Include all four of the above questions. If you can only ask one, ask the "considered leaving" question and accept that you will lose some diagnostic signal in exchange for the cleanest possible predictor.
2. Watch the bottom of the distribution, not the average. Engagement scores are usually reported as averages. The predictive signal is in the proportion of responses below a threshold (4 on a 5-point scale, in our data), not in the average score.
3. Re-run the survey at the cadence that matches your attrition window. A once-per-year survey gives you a yearly attrition prediction. If you care about 90-day risk, you need pulses at least quarterly.

If you want to see what these four questions look like applied to your own data, our team can run a one-time analysis as part of a Kestrel demo. Get in touch.