Attrition Modelling: How to Use Pay Data to Predict and Prevent Turnover

The Core Design Challenge

Predictive attrition modelling shifts the intervention point from the exit conversation to the earlier signals that consistently precede exit decisions. The compensation signals are the most accessible of these: compa-ratio below 80% with tenure above 24 months, consecutive zero merit increases, high-performance ratings with below-median pay positioning, and role type mobility (a characteristic of the role, not the employee — Software Engineers have more and faster external job opportunities than most other role types). These signals do not guarantee that an employee is about to leave; they identify the population where the risk is elevated enough to warrant a proactive conversation.

The business case for predictive attrition investment rests on the cost differential between retention and replacement. Replacing an employee typically costs 50-150% of their annual salary when recruitment, onboarding, and the productivity gap while the new hire reaches full effectiveness are all included. For a £60,000 Software Engineer, the replacement cost is £30,000-90,000. A proactive salary adjustment of £3,000-5,000 that retains the employee is 6-30× more cost-effective than a recruitment cycle. The arithmetic strongly favours early intervention — and the earlier the intervention, the more options are available. A salary adjustment proposed before the employee has started job-searching is significantly more effective than one proposed after they have received an external offer.

Building an attrition risk model does not require advanced machine learning or specialist data science. A simple additive scoring model — assigning point values to each risk factor based on its predictive strength, summing them to a total risk score, and banding the score into low/medium/high risk categories — is practical, interpretable, and deployable in a standard spreadsheet. The key is including the right factors (compa-ratio, performance-to-pay alignment, consecutive merit misses, tenure in grade, role type mobility) and weighting them appropriately for the organisation's specific context. A model that identifies the right employees and generates the right conversations is more valuable than a sophisticated model that no one trusts or acts on.

How the Approach Works

Role type mobility is the most commonly overlooked factor in compensation-focused attrition models. A below-midpoint compa-ratio represents a higher attrition risk for a role with abundant external alternatives (Software Engineering, financial analysis, data science) than for a role with fewer direct external equivalents. The speed with which a motivated employee can convert an exit decision into an external offer depends heavily on the external market for their skills — and this market characteristic is specific to the role, not the employee's tenure or performance. Weighting role type mobility appropriately in the risk model produces significantly better predictions than models that treat all roles as equivalent in their external opportunity context.

The intervention protocol is what converts a risk score into a retention outcome. A high-risk score that triggers a report but no action is a data project. A high-risk score that triggers a defined sequence — HBP conversation with the manager, salary review within 30 days, career conversation with the employee — is a retention programme. The sequence must be defined, resourced, and governed before the model goes live: who receives the high-risk list, who is responsible for each action, what the timeline is, and how progress is tracked. Without this, the model will identify the problem accurately and fail to prevent the outcome it predicted.