Talent intelligence, explained

Numbers you can interrogate, not just trust

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.

Four rules the intelligence lives by

Every score shows its work

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.

We abstain, we never guess

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.

Estimates get verified

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.

Built for how India hires

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.

What we look at, signal by signal

A look inside the models. Not the full recipe, but enough to see that there is a recipe.

Open to Move

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

  • Open-to-work announcements and availability phrases in the profile itself
  • Time in the current role vs that person’s own typical stint
  • Natural transition windows: people move in patterns, and we model them
  • Company momentum: teams that are shrinking or frozen loosen people up
  • Profile freshness: a recently polished profile is a tell
  • Just-moved detection: someone three months into a new job is not a lead

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.

Joining Window

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

  • Stated notice periods anywhere in the profile ("serving notice", "LWD", "immediate joiner")
  • Employer-type notice patterns: IT services majors run 60 to 90 days, funded startups 15 to 30, product MNCs 30 to 60
  • Seniority stretch: leadership transitions take longer and the window reflects it
  • Employment gaps: no current role usually means they can join now
  • Layoff and restructuring signals that compress real notice served
  • Notice buyout likelihood, so you know when money can shorten the wait

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.

Retention & Flight Risk

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

  • Tenure stability: the full distribution of stints, not just the average
  • Short-stint history and whether it is a phase or a pattern
  • Internal continuity: promotions and role growth inside one employer vs serial hopping
  • Employer stability: risk looks different at a stable MNC vs a runway-limited startup
  • Career stage: the same tenure reads differently at year 3 and year 13

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.

Compensation Estimates

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

  • Function, level and years of experience as the base band
  • City and market: the same role prices differently in Bangalore, Gurgaon and Indore
  • Company tier: top-tier tech, product MNCs, GCCs and services shops pay on different curves
  • Industry pay tiers across 140+ industries, because big tech and big edtech are not the same "big"
  • Institution premium: IIT, IIM, BITS and top-tier signals move the band
  • Career trajectory: fast risers price above their band, and we model it

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.

Impact Potential & 36 Power Signals

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

  • Promotions and scope growth: teams, budgets and remits getting bigger
  • Ownership language and quantified outcomes in how they describe their work
  • Startup experience, zero-to-one exposure and comfort with ambiguity
  • Leadership potential, cross-functional range and stakeholder mastery
  • Learning velocity: certifications, new stacks and skill refresh over time
  • Execution rigor and impact density across every role, not just the latest

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.

Authenticity

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

  • Peer benchmarking: does this trajectory look like real peers at the same level and market?
  • Industry benchmarks for titles, tenures and progression speed
  • Internal consistency: overlapping timelines, multiple "current" roles, dates that do not add up
  • Claim-to-evidence ratio: senior titles with junior-shaped histories get flagged
  • Staleness and shell-profile tells, so dormant copies are not read as fresh facts

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.

Company Intelligence

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

  • Headcount trajectory over time: growing, stable or contracting
  • Funding recency and runway pressure at venture-backed employers
  • Layoff and restructuring events, cross-checked from multiple signals
  • Firmographics on millions of companies: size, industry, stage and pay tier

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.

Fair questions, straight answers

Are these scores just AI guesses?

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.

What happens when a profile is thin or incomplete?

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.

How accurate are the compensation estimates?

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.

How can you know someone’s notice period before talking to them?

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".

Where does the data come from?

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.

Do the signals update over time?

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.

See the intelligence on a real search

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.