One offer above band is paid for by everyone already in it.
Remy drafts the offer from the interview outcome and the job grade, checks the figure against the band before it goes anywhere, tracks it from sent to accepted, and tells you every quarter why offers are being declined. It never sends one, and it never changes a number.
- Never sends an offer
- Never changes a figure
- Every draft goes to your queue
Pay inequity isn’t a policy failure. It’s a series of Fridays.
Nobody sets out to pay two people differently for the same job. It happens one offer at a time, at the end of a long process, when the role has been open for eleven weeks and the candidate has another conversation going and the number that closes it is slightly above what everyone else in that grade is on.
Each of those decisions is defensible on its own. The pattern they add up to is not, and it surfaces two years later when somebody compares payslips.
The other half of the problem is quieter: when candidates decline, almost nobody aggregates the reasons. Each one is filed as a lost negotiation rather than as evidence that the band is wrong.
Five tasks. Every one of them reversible.
Nothing on this list creates a commitment. That’s not a coincidence.
Draft the offer letter
Builds it from the interview outcome and the job grade, using your template. The draft lands in your queue.
Never: sends it.
Check it against the band
Compares the proposed figure with the salary band for that grade and flags anything outside it — including who else in that grade would now be paid less.
Never: changes the figure.
Track what happens next
Sent, opened, accepted, declined, expired. You see where every outstanding offer stands without asking anyone.
Never: chases a candidate without your approval.
Aggregate why offers are declined
Quarterly, in aggregate, fed back into salary benchmarking and into workforce planning.
Never: reports an individual’s stated reason.
Propose options when someone negotiates
Lays out what’s available within the band, what’s outside it, and what each choice implies for the rest of the grade.
Never: responds to the candidate itself.
Sending an offer and setting a figure are Binding actions. They belong to a person permanently, and no configuration changes that.
“Above band” is easy to say. This says who it costs.
A warning that a figure is outside the band is mildly useful. A warning that names how many people already at that grade are now paid less is a different conversation, and it’s the one that actually changes the decision.
It runs before the draft leaves your hands.
Not at approval, not at payroll, not in a quarterly audit. At the moment the figure is typed.
It names the consequence, not just the breach.
How far outside, and how many people at that grade sit below the proposed figure. You can still go ahead — with the cost in front of you rather than discovered later.
It checks the floor too.
Below-band offers are flagged as well, because underpaying a new hire relative to the market is how you lose them in month seven.
It flags. You decide.
There are good reasons to go above a band — a scarce skill, a market that moved, a person who is genuinely exceptional. The agent has no view on any of that and doesn’t try to have one. It puts the number, the boundary and the consequence on one screen and stops.
Salary bands come from your own job grade structure. Career GPT doesn’t set them and doesn’t recommend figures.
Every agent action falls into one of three tiers.
The useful question isn’t whether an agent is autonomous. It’s what specifically it can do without you, and that has a precise answer for every task.
All five of this agent’s tasks are Reversible.
Which is unusual for the stage of the process it works in, and deliberate. The offer stage is where an automated mistake is least recoverable, so the agent’s entire job is preparation and checking. The two actions that matter — sending, and choosing the number — never left your hands to begin with.
Sent, opened, accepted, declined, expired.
Offers go quiet. Somebody sends one on a Tuesday and nobody is quite sure a week later whether it’s been read, whether the deadline passed, or who was supposed to follow up.
Every outstanding offer shows its state and how long it’s been there, with the response deadline visible. When a deadline is close, the agent tells you — and asks before it reminds the candidate, because that’s an External action.
A declined offer is shown neutrally. Someone choosing a different job is a normal outcome, not a failure, and coloring it red trains everyone to treat it as one.
- Live status on every outstanding offer
- Deadlines surfaced before they pass
- Reminders ask first
Ten declined offers is a pattern. One is a story.
Every declined offer gets explained away individually — they had another option, the commute, the timing. Put a quarter of them side by side and something else usually appears, and it’s often that the band hasn’t moved since it was set.
The agent aggregates the reasons quarterly and feeds them into your salary benchmarking and your workforce planning. Not to tell you your pay is wrong — to show you where candidates keep saying so, in a form you can take to a compensation review.
It’s aggregate only. An individual’s stated reason for declining money is sensitive, and the report never attributes one to a named person.
- Quarterly, aggregated
- Never attributed to an individual
- Feeds compensation benchmarking and planning
In Grade 5, the median expectation across declined offers was $11,000 above the band ceiling.
Options laid out. No position taken.
A candidate comes back with a number. The agent lays out what’s available — where their request sits against the band, what’s possible within it, what going outside it would mean for the four people already at that grade, and what non-salary elements exist in your own structure.
It doesn’t recommend one. It doesn’t reply to the candidate. It doesn’t have a view on whether this person is worth it, because that judgment requires knowing things about the role and the team that aren’t in any of its data sources.
What it removes is the part where somebody has to work out the band implications under time pressure, in the twenty minutes before a phone call.
- Options, never a recommendation
- Band consequence shown per option
- You do the talking
Candidate proposed $76,000 · Grade 5 band tops at $68,000
Four sources, three switches, one log.
Read-only. It never writes to your band structure.
Reminders are External. Off unless you turn them on.
Agent runs use your monthly allowance.
Most of Career GPT is unlimited. The agents are one of the three things that aren’t.
Draws on your AI allowance
- Band checks on an existing draft don’t count
- Drafts and analyses stay available at zero remaining
- Credits are self-serve
Unlimited and free
- Offer status tracking
- Reading the run log
- Reviewing the queue
- Sending offers yourself
- Your band structure
- Everything in Pipeline, Talent Pool, Assessments, Competency Framework, Pre-Onboarding and Reports
At zero remaining, drafting pauses. Outstanding offers keep tracking, and you can still send, change and manage everything by hand.
This is one of five.
Zola
Adapts and publishes roles across every channel, tracks cost per qualified applicant.
Learn more →Iris
Reads resumes against your competency framework and routes candidates to the right next step.
Learn more →Orin
Conducts a scored first-round interview against the role’s competency requirements.
Learn more →Remy
Drafts and tracks offer letters, with salary band checks before anything goes out.
You’re reading this oneHana
Manages the checklist from signed offer through the first ninety days.
Learn more →One allowance, one log, one queue, and the same three tiers across all of them.
Measured inside the product.
Food retail chain · 22 stores
of offers made within band after three months
Logistics company · 3 warehouses in the south
offer acceptance rate
Hotel group · 4 properties
from interview outcome to offer sent
Questions people ask.
Can it send an offer?
No. Sending is a Binding action and always requires a person. There’s no setting that changes this.
Can it change a salary figure?
No. It can flag a figure against the band and lay out options. Choosing the number is yours.
Where do the bands come from?
Your own job grade structure. Career GPT doesn’t set bands and doesn’t recommend figures.
What does the band warning actually tell us?
How far outside the band the figure sits, and how many people currently at that grade would be paid less than the proposal.
Can we still go above band?
Yes, whenever you decide to. The point is that you decide it with the consequence visible rather than discovering it later.
Are candidates’ negotiation positions shared?
Only within the hiring team. Never outside it.
Is a candidate’s decline reason attributed to them?
No. The analysis is quarterly and aggregate only.
Will it chase candidates about deadlines?
Only if you turn reminders on, and it asks before each one, because contacting a candidate is an External action.
How many runs are included?
500 runs per month, shared across the agents.
Who is responsible for the offer letter template?
You are. The agent fills your template with data from the interview outcome and the grade; the wording and its legal review are yours.
See the consequence before you send the number.
Drafted from the interview outcome, checked against your band, waiting in your queue for a person to decide.
Start free- Never sends
- Never changes a figure
- Every Binding action names a person