Every candidate on TalentGPT carries predictions: will they move, when can they join, will they stay, what will they cost. This page is the sneak peek into how those numbers are made, what evidence feeds them, and where their limits are.
Click any signal on any profile and you see the evidence behind it: the roles, tenures, patterns and company context it was computed from. No black-box numbers.
When a profile does not carry enough evidence for a signal, we say No Data instead of inventing a score. Missing information is never silently punished or papered over.
A notice period starts as an estimate from employer patterns. The moment a candidate states it in a reply, the profile is upgraded to what they actually told us.
Pay in LPA, notice periods in days, employer archetypes from IT services to funded startups, and institution signals like IIT and IIM are native to the models, not bolted on.
A look inside the models. Not the full recipe, but enough to see that there is a recipe.
How likely is this person to take your call?
Most tools show you who looks good. We also model who is actually reachable right now, because a perfect profile that will not move wastes your week.
Patterns we read
plus 20+ supporting patterns weighed together
Why you can trust it: A single flag is never enough. "Very Likely" requires converging evidence, and if the profile contradicts itself (a stale open-to-work badge after a fresh role change) we cap the rating and re-verify before letting it climb.
If they say yes, when can they actually start?
In India the gap between offer and joining is where hires die. We estimate a start window for every candidate before you have even spoken to them.
Patterns we read
benchmarked across 90+ named Indian employers and employer archetypes
Why you can trust it: Windows are labeled as estimates until the candidate confirms. The first reply usually states the real notice period, and when it does, the profile switches from "estimated" to "told us" automatically.
Will they stay once they join?
A great hire who leaves in eight months is a failed search. We read the career history the way a seasoned recruiter would, at scale.
Patterns we read
plus trajectory patterns learned from millions of Indian career paths
Why you can trust it: Risk is shown as a level with the reasoning attached, so you can disagree with the model and see exactly where it got its read.
What will this person cost, in LPA, before the first call?
Expected CTC is modeled for every candidate in the pool, so budget fit is a filter you apply upfront instead of a surprise in round three.
Patterns we read
two numbers per candidate: market rate and their likely expectation, plus the gap between them
Why you can trust it: Validated against real, disclosed packages: our estimates land within roughly 11% of actuals on average, with the true figure inside our band in about 9 of 10 cases. Every estimate still shows its inputs.
Who will actually move the needle after joining?
Beyond the four headline signals, every profile is scanned for 36 power signals: the patterns that separate people who ship from people who attend.
Patterns we read
36 signals per profile, each with a level and the evidence behind it
Why you can trust it: Signals are computed from what the career record shows, then refined by an AI read of what the person actually shipped. Self-reported buzzwords do not score.
Is this profile what it claims to be?
Inflated profiles waste interviews. Before you invest a call, we pressure-test the story the profile tells.
Patterns we read
checks run automatically, with a deeper AI verification available on demand
Why you can trust it: Authenticity concerns flag a profile for your review. We never silently reject anyone; you always see the flag and the reason, and you make the call.
What is happening at their employer right now?
Candidate signals are only half the picture. We track the companies too, because the best time to reach someone is written in their employer’s trajectory.
Patterns we read
company context feeds every candidate signal above, automatically
Why you can trust it: When a company contracts, its people’s Open to Move and Joining Window update on their own. You see the shift without watching the news.
No. The core signals are computed from evidence in the career record: roles, tenures, promotions, company context and stated availability. AI is used to read and summarize what a person shipped, not to invent numbers. Every score shows the reasoning behind it, and you can open it on any profile.
We abstain. A signal without enough evidence shows No Data instead of a made-up score, and overall ratings are discounted for data completeness rather than quietly filled in. A candidate is never ranked up or down on information that does not exist.
Validated against real disclosed packages, estimates land within roughly 11% of actual CTC on average, and the actual figure falls inside our predicted band in about 9 out of 10 cases. Estimates are shown as a band with a market rate and a likely expectation, never as a fake-precise single number.
It starts as an estimate: stated phrases in the profile, employment gaps, and notice patterns by employer type (IT services majors run 60 to 90 days, funded startups 15 to 30, product MNCs 30 to 60), adjusted for seniority and events like layoffs. The first reply usually states the real number, and the profile is then upgraded from "estimated" to "told us".
Public professional profiles and licensed data providers, enriched with our own company intelligence built from millions of firmographic records. We do not scrape private accounts, read personal inboxes, or use any data a candidate has not made professionally visible.
Yes. Profiles are re-read when they are re-sourced or re-opened, company intelligence accrues continuously, and candidate replies upgrade estimates to confirmed facts. Scores you have already used for a shortlist stay stable, so your rankings do not silently reshuffle under you.
Type a role you are hiring for and watch every candidate arrive with move likelihood, joining window, expected CTC and the evidence behind each. Free, no card required.