Why early attrition in tech is more related to career progression than company culture.

      TL;DRA A study on People Analytics involving 205 tech professionals revealed that early employee turnover is primarily influenced by a lack of career advancement rather than workplace culture. The most significant factors for retention were promotions, internal movement, and observable growth opportunities, while team social interactions had minimal measurable effect.

      Entering this research, I believed I already understood the reasons behind early turnover. After spending over a decade in People Analytics, particularly at Meta, I developed a theory regarding why tech employees leave their positions within a year. I thought two main factors contributed significantly: the rate at which employees received promotions and their frequency of socializing with teammates outside of work. The first factor seemed obvious, while the second appeared to be an overlooked human aspect in the industry.

      I was partially correct. Upon surveying 205 global tech professionals and using a machine learning model to predict early turnover, it became evident that promotions were the most crucial indicator in the dataset. However, socializing had little impact. Furthermore, other significant factors associated with promotions pointed to a trend I hadn’t fully anticipated: early attrition in tech is fundamentally a career momentum issue rather than a cultural one.

      This realization has transformed my perspective on retention, and I believe it may do the same for you.

      The tech industry has historically faced high turnover rates. Many tech companies have a median employee tenure of about one year, independent of the company's size. This phenomenon isn't a result of the post-pandemic job landscape or merely a trend in a hot job market; rather, it has been a persistent issue since the industry's inception, which has never been properly addressed.

      The costs associated with turnover are well established. Replacing an employee can cost up to 2.5 times their salary when considering factors like recruitment, onboarding, lost productivity, and the critical knowledge that leaves with them. Research indicates that a one standard deviation increase in attrition correlates with an 8.9% drop in profits. In a time when tech firms are investing billions in AI infrastructure and closely reviewing costs, the financial hit from preventable attrition is more difficult to justify than ever.

      What is less understood is why this issue continues despite significant investments to resolve it. Tech companies spend extensively on benefits, engagement programs, cultural initiatives, and manager training. While some of these efforts yield marginal improvements, none have shifted the overall trend significantly.

      I would argue that a key reason for this challenge is that most retention initiatives are reactive. When an employee indicates dissatisfaction or resigns, that's when interventions occur—but often it’s too late to change their mind. My interest has always been in whether it’s possible to foresee turnover early enough to act on it.

      To explore this question, I created a dataset. The first challenge was finding data. Although numerous public datasets on employee turnover exist, most lack industry specificity. The most commonly referenced is a fictional HR dataset developed by IBM data scientists, frequently utilized in academic research but offering no insights into the tech sector.

      To bridge this gap, I developed a 24-question survey and distributed it globally among tech industry professionals, requiring that both their current and former employers be technology companies. After filtering out duplicates and incomplete responses, I ended up with 205 usable records. While this isn't a large dataset by industry standards, it’s clean, specific, and tailored to the question at hand.

      I defined “early attrition” as leaving a job within the first year. Each respondent was classified according to this criterion, forming the target for the machine learning model’s predictions.

      I trained five machine learning algorithms on this dataset, testing various configurations. I used the F1 score to gauge performance as it provides a more accurate picture than simple accuracy, which can be misleading. A model could achieve high accuracy by predicting that everyone would stay longer, as that is the common outcome. The F1 score considers this imbalance. The optimal model combined an Extra Trees Classifier algorithm with SMOTE, which generated synthetic examples to correct the dataset imbalance, achieving an F1 score of 0.97.

      The model was successful, revealing important insights.

      Promotions emerged as the most significant predictor of early turnover. The correlation was -0.54, which indicates that fewer promotions equate to a higher likelihood of early attrition. This finding aligns with expectations in the tech industry. Promotions signify that a company recognizes and invests in an employee’s future; the absence of this signal often leads employees to seek opportunities elsewhere.

      Nearly half of the respondents (49%) reported never having been promoted in their previous roles. This statistic was striking for an industry that prides itself on meritocracy. The model highlighted a noticeable issue that hadn’t been acknowledged before.

      Additionally, three other factors surfaced as noteworthy indicators of early attrition. Each deserves individual consideration as their implications may be counterintuitive.

      - Age: Younger workers were significantly more prone to early turnover, with a correlation of -0.49. This suggests that older employees are less likely to leave within the first year. From