The most expensive, most volatile asset a company holds has never been priced. Here is what changes when it is.
For a century, three disciplines learned to price the future. Actuaries priced mortality and built insurance. Statisticians priced default and built modern credit. Analysts priced cash flows and built the securities markets. Each took an uncertain future event, bounded it inside a confidence interval, and made it underwritable — assessable, traceable, and priced.
Talent never received this layer. The single most expensive and most volatile asset a company holds — the forward trajectory of its people — is still governed by intuition and backward-looking sentiment. This paper introduces career trajectory underwriting: the practice of reading a career sequence and underwriting its future — who is likely to leave, when, and what it would cost to keep them. We argue that the résumé is not a record but a forward curve; we present a model in which departure is governed by three independent clocks; and we describe a fully auditable pipeline that returns a priced, traceable verdict on a person's trajectory in minutes.
A modern enterprise underwrites almost everything. It prices the risk that a building burns, that a debtor defaults, that a currency moves, that a shipment is lost at sea. It buys cover, sets reserves, and quantifies its exposure to a precision of basis points. And then its best engineer — the one carrying the roadmap in their head — updates a résumé on a Tuesday night, and the firm has no number for it at all.
This is the strange blind spot at the centre of the org chart. The asset that is hardest to replace, slowest to ramp, and most expensive to lose is the one asset the firm has never learned to price. Headcount appears on no balance sheet as a forward curve. It appears as a cost line and a hope.
The reason is historical, not philosophical. Underwriting requires a record that points forward — a signal the future leaves in the present. Insurance had mortality tables. Credit had repayment histories. Talent appeared to have only the résumé, and the résumé was read as a backward artifact: a list of things already done. Read that way, it predicts nothing. Read correctly, it is the most forward-looking document a person produces.
There is a quiet asymmetry in how a company learns that someone is leaving. The employee knows at T − 90: the recruiter's call, the quiet weekend of updating a profile, the second interview booked as a "dentist appointment." The company learns at T = 0: the calendar invite titled "quick chat," the letter, the two weeks' notice that is already too late to matter.
The ninety days in between are not silent. They are loud with signal — tenure pushing past the person's own exit cadence, compensation drifting below market, a title that has stopped moving while scope keeps growing. These are not feelings. They are facts visible in the public career record, legible to anyone reading forward instead of back. The company simply was not reading.
Career trajectory underwriting collapses the latency. It reads the signals that are already present at T − 90 and returns the ninety days — the only window in which retention is still cheaper than replacement. After the letter lands, every lever costs more and works less. Before it, the same levers are an ordinary line item in a one-on-one.
Decisions about people have moved through two eras and are entering a third. The era of intuition asked a manager how they felt about their team and trusted the answer. The era of sentiment replaced the hunch with the engagement survey — a genuine advance, but a measurement that is both lagging and endogenous. It records how a person feels about the past, inside one organisation, on the day they answered. It cannot see the offer the market will extend in six months, because that offer has nothing to do with how they feel and everything to do with what they have become.
The manager's read. Unrecorded, unfalsifiable, and almost never wrong about the people it already likes.
The engagement survey. Backward-looking, organisation-specific, and blind to the external market that actually does the poaching.
The forward curve. Reads the external career record to price who leaves, when, and what it costs — probabilistically, and with the reasoning shown.
The third era inverts the question. It does not ask how does this person feel about us. It asks what will the market offer this person next, and how does that compare to what we are giving them. The first is sentiment. The second is underwriting. Only one of them predicts a departure that has not happened yet.
Treat a career the way a desk treats any other asset and the analysis writes itself. The sequence of roles, companies, tenures and skills is not a biography — it is a price history. From a price history one builds a forward curve: where this trajectory is most likely to go, and how confident one can be at each step. TalentOracle outputs that curve as a three-year earnings band, conservative through aggressive, with a compound growth range rather than a single false-precise number.
The confidence band is the entire point. A point estimate — "this person will earn $300,000 in three years" — is a guess wearing a suit. A band — "$255K to $358K, base case $300K, growing 11.5% to 17.8% a year" — is an underwriting statement. It tells a leader not just where the trajectory is heading but how much a compensation adjustment actually buys, and how much uncertainty they are pricing against. You cannot manage what you can only guess. You can manage what you can bound.
No single signal predicts a departure. A person can be highly desirable to the market and still stay for years; a person can be underpaid and still have nowhere better to go. Departure is not a level — it is an alignment. We model it as three independent clocks, each running on its own signals. A clock striking alone is noise. Departure becomes probable when two or more strike together.
What the outside world will pay to take them — and how far below it they currently sit.
How much runway remains where they are before the trajectory flattens against the org's ceiling.
Where they sit in their own historical pattern of movement — overdue, or only just settled.
Underwriting that cannot be audited is not underwriting — it is a guess with better production values. TalentOracle's pipeline is built so that every conclusion traces back through a chain of discrete reasoning agents, each with a strict output contract: no unstructured prose, no claim without a signal beneath it. The chain that produced the score is preserved on the report itself.
Normalises raw résumé text into a canonical career record — positions, tenures, seniority, skills, estimated comp band.
Why downstream reasoning needs structured data, not prose. One consistent input regardless of how the résumé was written.
Scores eight to twelve signals across velocity, stability, market desirability and trajectory momentum — the empirical predictors that feed the three clocks.
Why velocity below cohort combined with high market desirability is the single strongest departure indicator observed.
Produces the three-year earnings bands and the 0–100 retention-risk score, plus the most likely next move and its timing.
Why point estimates are false precision. The output is a calibrated interval, tuned against confirmed outcomes — not a naïve heuristic.
Generates a ranked playbook of three to five interventions specific to the person's signals — action, rationale, estimated risk reduction, cost, timeline.
Why a risk score without an action plan is a dashboard metric. The playbook is what makes the report operational.
Writes the report in plain English, grounded in the specific data points from this résumé. Vague language is prohibited by the schema.
Why every sentence in the narrative is the audit trail in human form. Each claim traces to a signal.
Compiles the final report and records model version, pipeline version, input hash and timestamp. Pure assembly — no new inference.
Why the audit trail is a first-class output. Same résumé, same model, same pipeline → same verdict, verifiably.
What the underwriter returns — illustrative, on a single anonymous career sequence. Six questions, one document: will they leave, when and why, what it costs to keep them, where they go next, what the trajectory is worth, and how you can check the work.
Most likely next move — Staff Engineer at a late-stage growth-equity-backed infrastructure company, on a 9–14 month horizon, at a 20–30% total-comp step-up.
Alternate path — lateral IC move to FAANG-tier comp band, lower probability, faster timeline. You are not competing with a counter-offer. You are competing with a title.
| Action | Rationale | Δ Risk | Cost |
|---|---|---|---|
| Promote to Staff (scope + title) | Resolves the title-compression signal — the loudest clock striking. | −26 pts | $0–title |
| Close 18% comp lag | Removes the market pull before the recruiter prices it for them. | −14 pts | ~$36K/yr |
| Charter-level project ownership | Extends visible runway against the ceiling clock. | −9 pts | Low |
Underwriting is, finally, a pricing exercise — and the price here is almost embarrassing. A report costs the price of a working lunch. The event it helps prevent costs more than a million dollars in recruiting fees, lost ramp, slipped milestones and the quiet contagion of churn that investors learn to read in a board deck. The arithmetic does not require optimism.
This is what underwriting has always done: it lets you spend a known, trivial amount now to bound an unknown, ruinous amount later. The report is the cheapest cover a company can buy on the one asset it has never insured — and the only cover that also hands back a playbook for keeping the claim from ever being filed.
Underwriting earns its authority by stating its limits in the same breath as its conclusions. A model that will not name what it cannot see is not institutional-grade — it is marketing. These constraints are not caveats appended to the work; they are the doctrine that makes the work trustworthy enough to act on.
The pipeline sees only the external career record. It cannot observe internal performance reviews, manager relationships, or real-time compensation — significant predictors it deliberately does not claim to hold. It underwrites what is legible, and says so.
A score of 80 does not mean this person will leave. It means that, across the population with this signal profile, departure rates are materially elevated. The unit of truth is the cohort; any individual outcome may differ.
It is a cold-start model: weights sharpen as confirmed outcomes accumulate. Early reports are less calibrated than those of a customer who has closed the loop for twelve months. Honest underwriting improves with evidence and admits its youth.
Reports are decision support for humans, never automated verdicts. They must not be the sole or primary basis for a termination, demotion, or any adverse decision. The underwriter informs the judgment. The person makes it.