Methodology

Career Trajectory Underwriting

A full account of how TalentOracle generates retention risk scores, earnings trajectories, and intervention playbooks — including the signal sources, model design, and limitations.

What is career trajectory underwriting?

Traditional retention tools score employees on internal history: tenure, performance reviews, engagement survey responses. These are backward-looking and organization-specific. They cannot see what the market will offer this person in six months.

Career trajectory underwriting works from the external career record — the sequence of roles, companies, and skills a person has accumulated — and underwrites their future: who is likely to leave, when, and what it would cost to prevent it. The methodology is borrowed from insurance and credit underwriting: assess the risk, quantify the confidence interval, and price the intervention.

The six-agent pipeline

Each report passes through six specialist agents in sequence. Every agent has a strict output schema — no unstructured text, no hallucinated claims. The final report includes the complete chain of reasoning.

01

Career Sequence Extractor

Structural parser

Converts raw resume text into a canonical JSON career record: positions (title, company, tenure, seniority level), skills, total experience years, estimated salary band, and education.

Why: Downstream agents need structured data, not prose. This step normalizes across resume formats and writing styles so signal extraction is consistent.

02

Signal Analyzer

Pattern recognizer

Scores 8–12 career signals across four dimensions: velocity (rate of title/comp progression), stability (tenure volatility), market desirability (skills in demand), and trajectory momentum (upward vs. plateauing).

Why: These signals are the empirical predictors of departure. Velocity below cohort average, combined with high market desirability, is the single strongest departure indicator we observe.

03

Trajectory Predictor

Forward-looking model

Produces the 3-year earnings trajectory (conservative, base, aggressive bands with CAGR) and the retention risk score (0–100). Also predicts the next most likely career move and estimated timing.

Why: Point estimates are false precision. We output confidence intervals. The risk score is calibrated against outcomes collected through the follow-up loop — not a naive heuristic.

04

Retention Intelligence Engine

Intervention designer

Generates a ranked playbook of 3–5 retention interventions specific to this person's risk signals. Each intervention includes: the action, rationale, estimated risk reduction, cost, and timeline.

Why: A risk score without an action plan is a dashboard metric. The intervention playbook is what makes the report operationally useful to an HR leader or operating partner.

05

Narrative Explainer

Plain-language translator

Writes the full report narrative — risk summary, signal walkthrough, earnings context, and departure window — in plain English, grounded in the specific data points from this resume.

Why: Every claim in the narrative traces to a specific signal. This is the audit trail. Vague language is prohibited by the output schema.

06

Report Architect

Assembly and audit

Combines all agent outputs into the final FinalReport structure. Records model version, pipeline version, input hash (SHA-256), and timestamp. Pure assembly — no additional inference.

Why: The audit trail is a first-class output, not an afterthought. Every report is verifiable: the same resume, same model, same pipeline version should produce the same conclusions.

Signal definitions

Velocity Score

How fast this person has progressed relative to their cohort. Accelerating velocity = lower risk. Plateauing velocity = higher risk.

Stability Score

Tenure consistency across roles. Frequent job-hopping at senior levels is a stronger departure signal than at junior levels.

Market Desirability

How in-demand are this person's skills today. High desirability with below-market comp is the most reliable departure predictor.

Trajectory Momentum

Whether the career arc is accelerating, stable, or decelerating. A senior IC who hasn't moved in 3 years at a growing company is a flight risk.

Comp Lag

Estimated gap between current comp and market rate for this role and experience level. >20% lag significantly elevates departure probability.

Career Ceiling Proximity

How close this person is to their estimated career ceiling in the current organization. At-ceiling employees leave for lateral roles at better-compensating companies.

Tenure vs. Title Mismatch

Years in role vs. title progression. Overdue promotions are a leading indicator of departure within 6–12 months.

Industry Signal

Whether this person's career has been in contracting or expanding industries. Contraction increases externally-motivated departure rates.

How the risk score is calculated

The retention risk score (0–100) is a composite of the eight signals above, weighted by their empirical predictive power based on confirmed outcomes in our dataset. Higher scores indicate higher departure probability within 12 months.

Low

0–39

Engaged and unlikely to depart without material change.

Elevated

40–59

Monitor. One or more signals warrant attention.

High

60–74

Proactive intervention is recommended within 30–60 days.

Critical

75–100

Departure likely within 3–6 months without immediate action.

Limitations

  • Resume-bound. The pipeline only sees what is in the resume. It cannot observe internal performance reviews, manager relationships, or real-time comp. These are significant predictors that the model does not have access to.
  • Probabilistic, not deterministic. A score of 80 does not mean this person will leave. It means that, in the population of people with this career signal profile, departure rates are significantly elevated. Any individual outcome may differ.
  • Cold-start model. Weights improve as confirmed outcomes accumulate. Early customers will see less calibrated scores than customers who have been closing the feedback loop for 12+ months.
  • Not a basis for adverse employment action. Reports must not be used as the sole or primary basis for terminations, demotions, or other adverse decisions. They are decision-support tools for humans, not automated verdicts.