How to Reduce Employee Turnover: Test What Changes

Reducing employee turnover starts with a specific, repeated voluntary-departure pattern and a working condition the organization can change. Diagnose the mechanism, choose an accountable action, verify delivery, and compare outcomes over a suitable period. A lower rate is encouraging, but it does not by itself prove that the action caused the change.

Define the outcome before choosing a program

Avoid starting with a generic instruction such as "improve culture" or "train managers." Name the population, event, period, and decision.

For example:

Decide whether a more predictable first-two-week schedule and a named day-three contact should continue for new service employees after two comparable intake periods.

The statement identifies the group, action, horizon, and decision. It can be evaluated. "Reduce turnover across the company" cannot.

Use definitions that match the decision. The U.S. Bureau of Labor Statistics JOLTS guidance distinguishes voluntary quits from layoffs and discharges and other separations. An internal employer analysis may need still narrower definitions, such as voluntary departures within 90 days of starting. State the numerator, denominator, and exposure period.

Do not turn a group pattern into a score for a named employee. The unit of action may be an onboarding cohort, role, site, work process, or manager population. The purpose is to improve a condition, not infer who will leave next.

Diagnose the mechanism, not only the rate

A turnover rate shows where to investigate. It does not explain why people left. Combine several evidence types:

The current CIPD turnover factsheet recommends understanding why employees leave before selecting retention initiatives. Its retention guidance covers areas such as flexibility, fair treatment, wellbeing, development, and work experience. Treat these as lines of inquiry, not a universal package.

When work pressure is part of the evidence, the UK Health and Safety Executive's Management Standards offer six useful work-design areas: demands, control, support, relationships, role, and change. The HSE framework also emphasizes employee participation in identifying practical improvements. It is a risk-management source, not evidence that one of those changes will reduce turnover in every workplace.

Write a testable action hypothesis

Use this form:

For population, evidence suggests working condition contributes to defined outcome through mechanism. Owner will deliver action by date. We will verify implementation, review near-term evidence, and observe turnover outcome over period. We will adapt, continue, scale, or stop under pre-agreed criteria.

Separate four questions:

1. Need: Is the problem material and repeated?
2. Delivery: Did the promised action reach the intended group?
3. Mechanism: Did the work experience change in the expected way?
4. Outcome: Did voluntary turnover move, and what else changed?

If delivery failed, the result does not test the action. If delivery succeeded and turnover did not move, the mechanism may be weak, the period too short, or another cause more important. If turnover fell, alternative explanations still need review.

Choose an action the owner can deliver

Match the action to the evidence and the person with authority.

A conversation cannot substitute for a staffing, pay, safety, or work-design decision. It can help define the issue and check how a change is experienced. Escalate urgent, legal, health, safety, or conduct concerns through the appropriate route rather than placing them in a routine retention experiment.

Create the validation worksheet before launch

Complete these fields:

Build evaluation into delivery. The UK government's Quality in Policy Impact Evaluation guidance distinguishes observing an outcome from attributing that outcome to an intervention. A credible comparison or counterfactual improves confidence, but the design should be proportionate to the decision. A simple before-and-after result is monitoring evidence, not causal proof.

Work through a fictional test

This example is fictional and deliberately small. It shows the calculation, not a guaranteed result.

Four service sites hire similar entry-level roles in two intake periods. Two sites introduce an earlier schedule and day-three contact. Two comparison sites keep the existing onboarding process during the same period. The groups were not randomized, so manager mix, applicant mix, workload, and local labor conditions may still differ.

Baseline cohort

• action sites: 15 voluntary departures within 90 days among 60 starters, or 15 / 60 = 25%;
• comparison sites: 12 among 60, or 12 / 60 = 20%;
• schedule delivered at least seven days before start at action sites: 27 of 60, or 45%;
• day-three contact completed at action sites: 31 of 60, or 51.7%.

Test cohort

• action sites: 9 voluntary departures within 90 days among 60 starters, or 9 / 60 = 15%;
• comparison sites: 9 among 60, or 9 / 60 = 15%;
• schedule delivered at least seven days before start at action sites: 52 of 60, or 86.7%;
• day-three contact completed at action sites: 49 of 60, or 81.7%.

The action-site turnover rate changed by 15% - 25% = -10 percentage points. The comparison-site rate changed by 15% - 20% = -5 percentage points. The exploratory difference in changes is:

-10 percentage points - (-5 percentage points) = -5 percentage points

There were six fewer early voluntary departures at the action sites than in their previous cohort. That is an observed count, not six departures proven to have been prevented. The comparison suggests that conditions beyond the action may explain part of the movement. Small counts, non-random allocation, site differences, and simultaneous changes prevent a causal claim.

Delivery also remained incomplete. Eight of 60 starters did not receive the schedule by the promised date, and 11 did not receive the day-three contact. The team should investigate those misses, review employee accounts of schedule predictability and support, check overtime and service quality, and continue for another comparable cohort before deciding whether to scale.

Review benefits, costs, and adverse effects

Turnover is one outcome. Also examine:

• whether the action reached the intended population fairly;
• employee experience of the targeted condition;
• manager and operations capacity needed to sustain it;
• overtime, absence, quality, safety, and customer effects;
• transfers, dismissals, and contractor use that could distort the rate;
• cost of delivery and any displaced work;
• whether the effect appears in different sites, roles, or cohorts;
• whether the original problem moved elsewhere.

Set review dates that fit the mechanism. Schedule publication can be checked before a start date. Role clarity may be reviewed after the employee has done real work. A 90-day turnover outcome needs complete 90-day exposure for each cohort. Do not call partial exposure a final result.

Decide what the evidence supports

Use disciplined language:

• Delivered: the action reached the planned group.
• Associated: the outcome moved alongside the action.
• Contributed: several evidence sources support the mechanism, with known alternatives.
• Caused: reserved for a design strong enough to establish attribution.

Most local HR tests will support delivery, association, or contribution claims. That can still improve a decision. Record uncertainty and choose the next evidence step instead of turning a promising result into a guarantee.

The employee retention strategies guide helps choose an initial pair of owned actions over 30, 60, and 90 days. This guide goes deeper on validating one action after it is chosen. Use the turnover cost worksheet when the decision also needs separate cash, capacity, and output views.

Where Lontra may fit

Lontra can be evaluated when the test needs structured employee or manager context about a defined work condition. Review the current platform capabilities, then use one trial campaign with up to 30 invitations over 60 days and no credit card to examine that question before a wider decision.

Use conversation evidence with operating records and human review. It does not predict a named departure or prove that an intervention reduced turnover. Managers receive a brief rather than raw employee responses, and aggregate HR views require at least five respondents.

Sources

• U.S. Bureau of Labor Statistics: JOLTS definitions
• CIPD: Employee turnover and retention factsheet
• CIPD: Retention guidance for people professionals
• UK Health and Safety Executive: Management Standards
• UK Government: Quality in Policy Impact Evaluation