Case study · Predictive and prescriptive analytics

Reducing 90-Day Attrition

After a stretch of fast hiring, early attrition spiked. Leadership had a limited budget and a real question: redesign onboarding, train managers, offer incentives, or some mix?

28%
lower early attrition
~40%
less manual reporting
17%
faster new-hire ramp-up

What changed

Before and after the interventions

Early attrition is shown in percent; reporting effort and ramp-up are indexed so that before = 100.

90-day attrition after rollout

Attrition fell from 35% to 25% within six months of the onboarding redesign and manager enablement.

Which levers worked

Estimated uplift by intervention

Mentor assignment and a structured first day did the most for early-career cohorts; manager training mattered most where specific managers were the issue.

Data and method

  • Data foundation: HRIS, ATS, LMS, survey and scheduling data joined into one governed dataset with PII masking, role-based access and small-cell suppression.
  • Predictive: survival analysis (Cox proportional hazards) to see when risk peaks, and gradient boosting with SHAP values to surface drivers such as first-week schedule adherence, mentor assignment and pay-band misalignment.
  • Uplift: a T-learner estimated the causal effect of each intervention by segment.
  • Prescriptive: a linear program allocated budget across sites and cohorts to maximize retained days within cost and fairness limits. It recommended about 60% to onboarding redesign for the highest-risk cohorts and 40% to manager training.

Model outputs

The original analysis outputs, shown as published (synthetic data).

SHAP feature importance: first-week schedule adherence, mentor assigned and LMS completion rate lead; offer salary band mismatch and manager span push risk up
SHAP drivers of early attrition. Green features lower risk when present; red features raise it.
Survival curves over 90 days: mentored cohort stays near 47 percent at day 90 versus about 23 percent without a mentor
Survival analysis: probability of staying over the first 90 days, mentored versus not.
Bar chart of optimal budget share: Cohort A 50 percent, Cohort B 30 percent, Cohort C 20 percent
Prescriptive optimization: recommended budget share by cohort.

Results

Within one quarter of launch, early attrition fell 28% across the pilot groups, manual reporting time dropped about 40% thanks to automated pipelines and scorecards, and new-hire ramp-up improved 17% on productivity KPIs. Leaders got a clear answer on where to invest.

About the data. Charts use the sample data from the original interactive dashboard (the old site had two near-identical copies; this uses the version that also includes the uplift chart). Outcome figures are carried over unchanged from the original case study.

Rebuilt from:

  • attrition_case_study.html (write-up, outcomes)
  • assets/img/shap_summary.png, survival_curve.png, optimization_allocation.png (original model outputs)
  • interactive_dashboards/reducing_90_day_attrition_dashboard.html (all three charts)
  • interactive_cases_refined/ninety_day_attrition_case.html and prescriptive_attrition_case.html (earlier copies)

More reports

Report · Workforce planning

Workforce Planning & Forecast

Twelve-month headcount scenarios by department under different attrition and hiring rates, plus a live Tableau forecast.

Open report
Analysis · Engagement

Measuring Engagement Impact

Does engagement move productivity and turnover? A department-by-department look at the data, with the honest answer.

Open report
Report · Workforce composition

Diversity Metrics

Representation by department across race, gender and job level, in one consistent view.

Open report