People Analytics · Ideation and architecture
RHappy Nexu
Academic project in the architecture phase that investigates how to organise historical HR data to identify signals associated with voluntary exits. The proposal foresees analyses that support case prioritisation and human decisions, with attention to privacy and explainability.
- Type
- Individual academic project · Data Science / FIAP
- Status
- Ideation and architecture
- Scope
- Voluntary-exit risk reading, as a proposal
The challenge
The signs that someone is about to leave show up before the final decision, but stay scattered across data and routines. The project aims to turn the historical HR base into an early reading of voluntary-exit risk among operational store staff.
The reading would cover two horizons, 90 and 180 days, and exists to buy time for a conversation, not to predict for its own sake.
What does people data reveal about who stays and who leaves?
Proposed architecture
The design starts from a historical Excel and CSV base that is cleaned, turned into variables available before the horizon, and feeds the models. Nothing recorded after the exit is used, to avoid data leakage.
- Historical HR base
- Cleaning and standardisation
- Variables from before the horizon
- Planned models
- Explainable risk signals
- Dashboard for human review
What the dashboard would do
Prioritise
Risk estimate per horizon and segmentation by profile and unit, as far as the data allows.
Explain
Analysis of the factors behind each reading, so a leader can understand the criterion and disagree with it.
Guide action
A recommendation such as “review the context with the responsible leader and record the action taken”. The reading supports the conversation; it does not make the decision.
Follow up
Turnover dashboard and tracking of the cases under attention, including in busier periods for the stores.
Visual example with illustrative data
Executive view
- 12prioritised signals
- 90 dayshorizon
- 4areas in focus
Areas under attention
- Store operations
- Customer service
- Logistics
- Cashier
Profile 04-7A, anonymised
Recommendation
Review the context with the responsible leader and record the action taken. The reading supports the conversation; it does not make the decision.
Responsibility
Transparency
Explain signals, criteria and limits.
Privacy
Use only what is needed, in line with Brazil’s data protection law (LGPD), with controlled access and a bias review to avoid discrimination.
Human oversight
The decision stays with responsible people. It is neither a verdict nor a punishment.
Planned technologies
- Data and preparationplanned
- Modelling, to be testedplanned
- Visualisation and explainabilityplanned