People Analytics: Your Boss Has a Flight-Risk Score on You — Here's How They Calculated It

People analytics platforms aggregate hundreds of behavioral signals into a single flight-risk score

Employee turnover costs US companies $1 trillion annually — a number so large it becomes abstract. Let's make it concrete: replacing a single senior software engineer at a major tech company costs between $150,000 and $250,000 when you factor in recruiting fees ($30-50K), lost productivity during vacancy (3-6 months of a $180K salary), onboarding and ramp-up time (3-6 months at 50% productivity), and institutional knowledge that walks out the door and cannot be replaced at any price. A mid-level manager departure costs 100-150% of annual salary. A frontline retail worker turnover costs about 30-50% of annual wages, but at scale — retail turnover averages 60% annually — it adds up to billions.

Despite this staggering cost, 67% of HR departments still rely on exit interviews to understand why people leave — a method that is, by definition, too late. People analytics, powered by machine learning, flips this equation. Instead of asking departing employees what went wrong, predictive models identify who is likely to leave 3 to 12 months before they resign, giving managers a window to intervene. The technology is not experimental. Google's People Analytics team, which has been operating since 2006, was one of the first to prove that data could predict turnover. Today, platforms like Workday, Microsoft Viva, and Culture Amp sell flight-risk prediction as a standard feature.

The Data That Predicts Your Departure

Turnover prediction models work by identifying statistical patterns in historical HR data. The inputs are extensive, and most employees have no idea how many signals they're generating. Workday's predictive model, trained on data from 65 million worker profiles across its platform, analyzes over 200 distinct signals. Here are the strongest predictors:

• Tenure sweet spot: The highest flight-risk window is 12-18 months after hire, when the initial enthusiasm fades and before stock vesting or promotion cycles create retention incentives. 30% of all voluntary departures occur in this window.
• Compensation relative to market: Employees whose pay falls below the 40th percentile for their role and location are 2.3x more likely to leave within 6 months. But raw underpayment isn't the only factor — perceived pay equity (comparison to peers) is an even stronger predictor.
• Manager quality score: Google's famous Project Oxygen identified that manager quality is the single strongest predictor of team retention. Teams scoring in the bottom quartile of manager quality have 2.5x higher turnover than top-quartile teams. The model predicted turnover with 75% accuracy up to 12 months in advance.
• Collaboration network centrality: Employees who are socially isolated within their organization — measured by email frequency, meeting attendance, and Slack/Teams interaction patterns — are 1.8x more likely to resign. This signal often appears 4-6 months before departure.
• Commute time changes: A 2025 study from the University of Chicago found that employees who move to locations with 20+ minute longer commutes are 34% more likely to quit within a year. Remote/hybrid work arrangements moderate this effect but don't eliminate it.
• Promotion velocity: Employees who haven't been promoted or given a significant scope increase in 24+ months, particularly at the 3-5 year tenure mark, show a sharp increase in flight risk. The absence of forward career movement is one of the most reliable departure signals.

Predictive models synthesize behavioral, structural, and sentiment data into actionable retention insights

NLP: Reading the Room at Enterprise Scale

Structured HR data tells only part of the story. The richest signals — frustration, disengagement, brewing conflict — live in unstructured text: open-ended survey responses, Slack messages, Glassdoor reviews, internal communication patterns. Natural language processing tools now extract sentiment and theme signals from these sources at scale.

Culture Amp's NLP engine analyzes millions of open-ended survey comments across its platform, detecting sentiment shifts across teams and departments. In a 2025 deployment at a Fortune 100 financial services company, the system identified a 40% spike in negative sentiment across teams reporting to recently departed managers — a "cascading turnover" signal that affected 12% of the organization. The company intervened with leadership support for affected teams and prevented an estimated 340 additional departures over the following 6 months, saving an estimated $51 million in replacement costs.

Microsoft Workplace Analytics, which draws on Teams, Outlook, and SharePoint metadata (not content — Microsoft explicitly does not analyze email or chat content), monitors collaboration patterns as burnout indicators. The system flags employees who send more than 30% of their emails outside working hours, whose meeting load exceeds 25 hours per week, or whose collaboration network is shrinking (signaling withdrawal). A 2025 study across 120 organizations using Workplace Analytics found that burnout-flagged employees had a 3.1x higher resignation rate within 90 days compared to non-flagged peers. Companies that acted on these signals — adjusting workloads, adding resources, or initiating manager conversations — reduced turnover among flagged employees by 28%.

AI-Driven Retention: From Prediction to Intervention

Prediction without intervention is voyeurism. The most sophisticated people analytics platforms don't just flag flight risks — they recommend specific retention actions. The logic is straightforward: different employees leave for different reasons, and the most effective intervention depends on the predicted cause.

Betterworks and Degreed, two platforms that focus on AI-driven development paths, use predictive analytics to recommend personalized career development plans. Betterworks' system matches employees to internal roles based on skills gaps and career aspirations, increasing internal mobility rates by 2.1x at companies that deploy it. The logic is compelling: internal moves cost 20% of what external hires cost, and employees who make internal transitions are 75% more likely to remain with the company 2 years later compared to those in the same role who don't move.

The Surveillance Question: How Much Monitoring Is Too Much?

The line between analytics and surveillance is thinner than most companies admit

Here's where people analytics gets uncomfortable. The same data that predicts flight risk — email patterns, meeting frequency, collaboration graphs, sentiment scores — can also function as workplace surveillance. A 2025 Gartner survey of 3,000 employees found that 51% would consider leaving their employer if they discovered their work behavior data was being used for predictive analytics without explicit consent. And 62% said they distrust employer claims that analytics data is anonymized and used only for aggregate insights.

Their skepticism is justified. Several high-profile incidents have eroded trust. In 2024, a leaked document from a major tech company (widely reported but never officially confirmed) revealed that its people analytics system generated individual flight-risk scores for every employee, including scores visible to their direct managers. The backlash was immediate. The company pulled the feature within 48 hours. More subtly, Microsoft Workplace Analytics generated controversy when privacy researchers discovered that the system could de-anonymize collaboration data by cross-referencing meeting schedules and email metadata, even though the system was designed to show only aggregate team-level insights.

Regulatory pressure is building. California's CPRA requires businesses to disclose the categories of personal information collected and the purposes for which it is used. The EU's GDPR restricts automated decision-making that "significantly affects" individuals, and several EU member states have issued guidance stating that flight-risk scoring of individual employees without explicit consent may violate the regulation. Illinois' Artificial Intelligence Video Interview Act and New York's Local Law 144 extend similar protections to specific AI hiring and monitoring contexts.

Best-practice organizations are navigating this tension by limiting people analytics to anonymized, aggregate data for trend identification and requiring manager consent and HR review before any individual-level data is surfaced. Workday, to its credit, designed its flight-risk feature to show only team-level risk scores, not individual scores, by default — though enterprises can configure it otherwise. The question isn't whether companies will continue analyzing workforce data (they will), but whether they'll do it transparently enough to maintain employee trust.

The ROI Is Real — But So Is the Risk

Despite the privacy concerns, the business case for people analytics is overwhelming. Companies in the top quartile of people analytics maturity — as measured by a 2025 Deloitte study of 800 organizations — report 82% higher retention rates, 3x faster internal mobility, and $2,400 per employee in annual savings from reduced recruitment costs. The same study found that only 24% of organizations have achieved this level of maturity, suggesting significant competitive advantage for those that invest early.

Google's People Analytics team, the pioneer in this field, demonstrated the power of data-driven HR over a decade ago. Its Project Aristotle — a multi-year study of what makes teams effective — identified psychological safety as the strongest predictor of team performance, a finding that reshaped Google's management development program. Project Oxygen codified the specific behaviors of effective managers and used them to create a manager coaching program that improved team satisfaction by 10% and reduced turnover by 25% in manager-changed teams. These weren't AI-driven at the time, but they established the principle that systematic analysis of workforce data produces actionable, measurable results.

Your employer knows more about your likelihood of leaving than you do. They have your salary benchmarked against market rates, your collaboration graph mapped against organizational health metrics, your sentiment trajectory from survey responses, and your promotion velocity compared to peers at the same tenure. The question isn't whether this data exists. It's whether anyone is using it to help you — or just to manage you.

People analytics is a double-edged sword. Used well, it identifies disengaged employees before they become departed employees, surfaces pay inequity before it becomes a retention crisis, and helps managers understand what their teams actually need. Used poorly, it becomes surveillance by another name, generating individual risk scores that employees never see and managers use to make decisions about careers they don't fully understand. The technology itself is neutral. The governance around it is not.