How it works
This page describes the design of ReturnOnTalent AI at the level a product or engineering reader needs to evaluate it — the pipeline, the auditability rules, the deployment options and the data principles. It deliberately does not describe the algorithms, the prompts, the scoring weights or the evaluation logic; those are proprietary and are not part of this public repository.
The run pipeline
One workflow, from raw data to board-ready answer.
- Data. Import your people data the way you already hold it — map the columns once, validate before anything runs. Nothing is pre-populated, nothing is trained on anyone else’s workforce.
- Run. Choose the population and the modules, see the estimated cost, then launch. Your data, your results.
- Attrition. A score per employee, with risk bands and the factors behind each score. See who is at risk months before a resignation.
- Hidden Talent. A nine-dimension view of performance and potential, mapped to what each person wants next. Find the talent you already paid for before you hire from outside.
- Sentiment. An employee sentiment index that separates what people feel from what the environment causes — with the verbatim evidence behind every score. Listen without surveilling.
- Executive Report. One consolidated, auditable summary that unites attrition, mobility and sentiment — the CHRO and the CFO read from the same page.
Auditability: two rules, built in from the start
- No score without proof. Every score cites its source — the field and the evidence behind it. You can challenge a number and get an answer. That is the difference between a retention dashboard and a retention decision.
- No score without a version. Every output carries the model, the prompt version, the weights and the date behind it. Results stay auditable, run after run.
These two rules are the product. A retention number you cannot trace is an opinion; a retention number you can trace is a decision.
Your model, your choice
ReturnOnTalent AI runs on the language model provider you already trust.
- Bring your own keys. Keep your people data where your compliance requires it, and your provider bills you directly.
- Or let us manage the models. We run them for you; the engine is the same either way.
Either way, your data is not used to train anyone else’s model.
Data principles
- Nothing pre-populated. The modules run on the workforce data you already hold.
- Nothing trained on anyone else’s people. Your data stays your data.
- Identifiers are pseudonymised at ingestion, so no individual is exposed in the pipeline.
- Compliance first. ReturnOnTalent is designed to comply with PDPA (Singapore) and GDPR (European Union). HR governs the process.
What is not disclosed here
The following remain proprietary to WeLinkTalent Pte Ltd and are intentionally absent from this repository:
- the machine-learning algorithms and feature engineering;
- the prompts and prompt versions;
- the scoring weights and evaluation logic;
- the internal service architecture.
Where this fits
ReturnOnTalent AI sits alongside your HRMS — Workday, SuccessFactors and others — and adds the retention layer those systems do not cover. It is not an ATS and not a job board.