Employee Lifecycle Check-ins
Find out who's about to leave while you can still keep them.
Goodfit calls every new joiner around day 90, holds a real conversation in the language they choose, and hands HR a named at-risk queue: who might leave, why, and what they asked for.
Trusted by HR teams at 200+ companies
The problem
Attrition is decided in month three. Most companies find out in month nine.
The signals are there early: a role that isn't what was promised, a manager who never checks in, a pay gap nobody has named. Nothing in most onboarding programs is built to hear them.
Exit interviews are post-mortems
By the time someone is on an exit call, nothing can be done about them. The reasons were there at month three. Nobody asked.
Surveys get the polite answer
A pulse form fired at a new joiner gets a four out of five and no words. It cannot probe, and it cannot tell "fine" from "fine for now".
Managers hear it last
The person most likely to notice a wobble is often the reason for it. Concerns about a supervisor do not travel through the supervisor.
Why it matters
A check-in is a live case. An exit interview is a post-mortem.
Both are worth running. Only one of them can still change the outcome for the person on the call.
EXIT INTERVIEW
Someone who has already left
- →Explains last quarter’s attrition
- →Produces a report about patterns
- →Fixes the next person, not this one
LIFECYCLE CHECK-IN
Someone you can still keep
- →Catches the wobble at day 90
- →Produces a queue with names on it
- →Tells you what would make them stay
How it works
From joiner list to at-risk queue, without anyone chasing calls.
Load a cohort once. Every person flows through the same call, the same analysis, and the same risk rule.
Share
Send your joiner list as a spreadsheet: name, phone, role, team, manager, joining date. Goodfit builds the cohort from it.
Call
Around day 90, the AI counselor phones each person and holds a five-to-ten-minute conversation in the language they choose.
Analyze
Every call is read the same way: commitment, risk and stay drivers, what they asked for, training, manager support, a recommend score.
Act
You get a live at-risk queue with names on it, a report per person, and a cohort workbook with the action list already written.
01 · The call
One gate question decides the rest of the call.
“Looking at the next twelve months, how do you feel about staying on here?” The answer routes the conversation. The employee just hears a conversation.
A real conversation, not a form: the counselor works through a structured script and follows up wherever an answer is thin. Five to ten minutes: long enough to get past the polite answer, short enough that a frontline employee takes the call.
Every question is open: options are never read aloud. People pick whichever option they heard last, or hang up. Answers are mapped to your categories afterwards by the analysis, not by the caller.
The branch is the design: one gate question about the next twelve months routes the rest of the call into settled, unsure, or may-leave. The employee never hears that a branch was taken.
Recorded and transcribed: every call is stored with a speaker-tagged transcript and a high-accuracy second pass, so the analysis reads what was actually said.
Day-90 check-in · branch map
Gate question
Looking at the next twelve months, how do you feel about staying on here?What makes them want to stay, and what we should keep doing.
The one thing holding them back, whether they have raised it, what would change their mind.
What is driving it, whether anyone here knows, the single thing that would keep them.
Everyone, on every branch: manager support, training, and the 0 to 10 recommend score.
02 · Attribution
Named, and told so up front.
You cannot intervene on an anonymous row. So the check-in is attributed, and the only honest design is to say so before the first question rather than after the fact.
AI Counselor
Do you want to continue in English, or would you prefer another language?Employee
English is fine.AI Counselor
Before we start: what you share goes to senior HR leadership, so please feel comfortable sharing your thoughts openly. How has your experience been so far, these first few months?Said, not asked: employees agree to the program before the call, so the line reassures rather than putting the same question a second time.
One objection ends it: if someone asks not to be named or asks to be anonymous, the call ends there and that person is never dialed again. Not by a retry, not by a later pass.
Read from their words: an objection is recorded from what the employee said, never guessed from the shape of the transcript. A refusal and a dropped line look identical in two turns and mean opposite things.
Language first, then the notice: the call opens by asking which language the employee prefers, then delivers the attribution line in that language. The one sentence that has to be understood is never spoken in a language they may not have.
03 · Reach
Someone who never picked up is unmeasured, not fine.
Most non-response is a connection problem, and connection problems are solvable. Nobody who was never reached gets quietly filed as low risk.
Three attempts, inside working hours: spaced calls, never at 9pm. You set the window once and the system does the chasing.
Callback capture: "call me tomorrow at four" is understood and scheduled, never filed as a hang-up.
Outcome per attempt: connected, no answer, unresponsive, callback requested, declined. Every attempt lands on an append-only timeline per person.
A short call is not a completed one: a call that never got past the opening stays in the queue as unreached, rather than being filed as a reading nobody gave.
Day-90 cohort · reach funnel sample data
04 · The risk rule
A published rule, not a black-box score.
Risk is a rule applied to what the employee said. Never a model's opinion, never a 0 to 100 number. Every screen, report, and export agrees because they all run the same rule, and the level can be defended out loud. It always arrives with its reason: a sentence repeating something the employee actually said.
The rule
Said they may not stay, or raised something flagged for review
Unsure, or settled but would not recommend working here
Settled, and would recommend
Never reached, or chose not to take part. Never counted as Low.
At-risk queue sample data
M. Alvarez
Account Manager
Account Manager
Day 92
High
Said they may not stay
S. Okafor
Store Associate
Store Associate
Day 88
Medium
Unsure about staying
J. Fernandes
Operations Associate
Operations Associate
Day 95
Low
Sees themselves staying
P. Iyer
Customer Support
Customer Support
Day 91
Unknown
Could not be reached
Filter by team, manager, driver, and time in the job. Every row opens the person.
05 · The analysis
Every call is read the same way.
Each conversation is analyzed into the same fields, mapped onto fixed lists so patterns roll up cleanly across hundreds of people, with verbatim quotes kept as evidence.
Commitment
Committed, unsure, or at risk, with one sentence of evidence from what they said.
Risk drivers
A primary and up to two secondary: better opportunity, expectation mismatch, pay, supervisor, culture, work-life balance, personal.
Stay drivers
Asked of the settled, never inferred from the absence of a complaint: growth, learning, pay, recognition, manager, culture, hours.
Requested actions
What they actually asked for: one-on-ones, clearer goals, skill development, a move inside the company, a fairer workload, recognition, a pay review.
Raised with anyone?
Often the most useful field on the record. A concern nobody has heard is a different problem from one that was raised and ignored.
Training and manager support
Whether onboarding actually prepared them for the job, and how supported they felt in the first months.
Recommend score
The 0 to 10 question, understood as a spoken number in any supported language, rolled up to eNPS for the cohort.
Flagged disclosures
Harassment, abuse, or coercion raised unprompted is flagged for review and forces the person to high risk on that alone.
06 · What you get out
A report per person. A workbook per cohort. Names on both.
Every aggregate is clickable through to the people behind it. A number a head of function cannot open is a number they cannot act on.
Cohort workbook, four tabs
Built to be opened cold by someone who was not in the briefing.
- →Dashboard: risk mix, reach funnel, commitment, drivers, eNPS
- →Your own sheet returned, original columns, our results appended
- →What they said: the quotes, tagged by driver
- →Action list: who needs a conversation, and about what
Per-employee check-in report
A downloadable PDF for every person on the list.
- →Risk band with the sentence that caused it
- →Their answers and their words, question by question
- →Recording, transcript, and how we reached them
Protecting people in the aggregates
The check-in is attributed, and everyone on the call was told so. A chart captioned “aggregate” that describes three people is not. So any group under five is not charted. Withheld groups are named with their headcount, never silently dropped.
Languages
Meet people in their own language.
Honest answers shouldn't depend on speaking the head-office language. Every call opens by asking which language the employee prefers, then continues in it, in the everyday register of a workplace rather than the textbook one.
12+
languages, chosen by the employee
Mid-call
Switch languages any time, no restart
0 to 10
Spoken scores understood in any language
Why Goodfit
Why it works where pulse surveys and skip-levels don't.
Early enough to be a fix
At day 90 a fix is still a fix. Nine months later it is a counter-offer, and counter-offers rarely hold.
A queue, not a report
The output has names on it, because every row is someone who can still be kept. A chart about last quarter cannot be actioned. A person can.
A rule you can defend
"The model said so" ends a program in its first review meeting. Every risk level here comes from a published rule and travels with the sentence that caused it.
Built for the whole workforce
The call happens in the language the employee picks, in the everyday register of a workplace. Frontline and field staff get heard the same way head office does.
FAQ
Frequently asked questions
How is this different from an exit interview?
Is the check-in anonymous?
When does the call happen and how long is it?
How is risk decided?
Which languages are supported?
Can we use our own questions?
What do we get at the end?
How do you protect people in the aggregate charts?
Does it connect to our HRMS?
Goodfit Employee Lifecycle Check-ins
Hear it at day 90, not at the exit interview.
See Goodfit call your own joiner cohort and hand you the queue with names on it.