Attrition Prediction uses your survey and employee data to assign attrition risk level to your company segments. To learn more about using and interpreting results, see Attrition Prediction reporting.
This article focuses on methodology, and includes:
Calculation
Attrition Prediction uses a statistical model trained on the leaving behaviour across millions of survey data points in the Peakon database. In determining the attrition risk per segment, the model also uses 5 key factors and follows this order:
- The model calculates attrition risk per employee.
- The model then uses employee-level attrition risk to calculate the average attrition risk for each segment, as well as for the whole company.
- The model compares the average risk of each segment to the average risk of the company, to assign an attrition risk level. Example: Attrition risk in the Marketing segment is in the top 10% of your organization.
Key factors
| Factor | Description |
|---|---|
|
Engagement score |
Difference to benchmark for the eNPS question: How likely are you to recommend [your organization] as a place to work?. We observed that employees who provide a 0-6 score to our eNPS question are more than 3 times more likely to resign than those who score the question 9-10. Employees are typically the most engaged when they first start a new job. On average, engagement declines by up to 1 point during the first 2 years of employment, regardless of whether the employee has any intention of leaving. This is expected and the decline often levels out after 1 year of tenure. |
|
Loyalty score |
Difference to benchmark for the Loyalty question: How likely is it that you would stay with [organization] if you were offered the same job at another organization?. The average loyalty score of employees that remain with a business is 20% higher than the average loyalty score of departing employees. |
|
Growth score |
Difference to benchmark for the Growth question: I feel that I’m growing professionally. This question relates to an employee's perceived opportunities to improve their personal and career growth. |
|
Participation |
Aggregated and latest survey round participation rate. The response rate of employees that resign is 15% lower compared to the employees who remain. Our analysis of survey responses indicates that employees that become disinterested in answering the survey also become disinterested in trying to improve their workplace, and therefore are at a higher risk of leaving your organization. |
|
Tenure |
Time spent at the organization, as per the Tenure (or equivalent) attribute. Generally, employees are more likely to leave in the first 2 years rather than later. Right after 3 months of tenure is also when the biggest drop in engagement occurs. This presents a paradox, as on the surface you have a more engaged employee, relative to other tenure groups, but they're also more likely to leave. Factoring tenure length into the calculation presents a more accurate indication of attrition risk. Although the Attrition Prediction feature displays 3 tenure ranges in the user interface, the prediction model maps your tenure segments against these tenure ranges:
For the attrition model to map segments to a tenure range, ensure that any customization of the tenure ranges still falls within the above ranges. |
The Attrition Prediction heatmap also displays a column with past resignations for context, although it doesn't contribute to the attrition risk calculation. The Resigned column displays data from the Resigned segment within the Separation Reason attribute (either the exact segment, or an equivalent segment, mapped by your Customer Success Manager).
Attrition risk levels
In short, the attrition model answers this question: "How does the attrition risk of each segment compare to the average attrition risk in your organization?". For interpretation and recommendations, see Attrition Prediction reporting.
| Risk | Meaning |
|---|---|
|
Severe |
The risk of attrition in this segment is in the top 10% of your organization. |
|
High |
The risk of attrition in this segment is in the top 25% of your organization. |
|
Elevated |
The risk of attrition in this segment is in the top 50% of your organization. |
|
Reduced |
The risk of attrition in this segment is in the bottom 50% of your organization. |
|
Low |
The risk of attrition in this segment is in the bottom 25% of your organization. Engagement in this segment is high. |
|
Minimal |
The risk of attrition in this segment is in the bottom 10% of your organization. Engagement in this segment is high. |
Note
Consult with your legal counsel to determine whether your configuration of segments, and thus the attrition risk by segment, satisfies your organization’s compliance requirements. Customers can configure and provide instructions to their workforce on the use of this feature to ensure it doesn't directly or indirectly cause discrimination or discriminatory results, whether intentional or not. Customers are responsible for understanding and complying with any legal obligations arising from their use of the Attrition Prediction model and attrition risks, including any assessment, testing, or documentation that may be required under anti-discrimination laws.
Improving prediction accuracy
Peakon’s Attrition Risk model works most accurately when you:
- Increase active survey participation whilst also actioning feedback.
- Move to a higher survey frequency to identify trends and risks earlier - the more recent the answers, the more accurate the prediction of current risk.
- Enable the standard Loyalty and Growth questions.
- Use standard tenure ranges, or ensure customized ranges still fit within the ranges that the Attrition Prediction model uses. Example: If you change the segments that were originally within the 3-12 month interval to 3-6 months and 6-12 months, both segments still fit within the interval. However, if you change the range to 3-13 months, the attrition model can't map the segment to any applicable interval.
The above points are not required for the Attrition Risk feature to work, however they will improve score accuracy and contribute to the data set used in the model.
Comments
0 comments
Article is closed for comments.