Keep offered candidates warm until they actually join

The build schedules contact across the window rather than leaving it to whoever remembers. A cadence is generated from the notice length and the joining date, with touchpoints weighted toward the periods where drop-outs cluster: the first week after acceptance, the week the resignation is submitted, the midpoint, and the ten days before joining.

Each touchpoint has a purpose rather than being a check-in for its own sake. Confirm the resignation was submitted and accepted. Confirm the relieving date. Ask directly whether a counter-offer has been made. Confirm the joining formalities and the reporting details. Recruiters can send these personally or let the cadence handle the routine ones over email or WhatsApp, on the channel the candidate consented to.

The most valuable question on that list is the direct one about the counter-offer, and it works better asked early and openly than discovered late.

The notice period is also when the joining paperwork has to be gathered, and chasing it is the least valuable use of a recruiter's week. The build collects the documents the client requires through a candidate-facing upload, tracks what is outstanding, and chases on a schedule rather than through a recruiter's memory.

Which documents are required is configured per client and per role, because a background-check package for a banking client and one for a mid-market product company are not the same list. Expiry-dated documents carry their expiry, and identity or right-to-work documents follow the same capture and reminder handling described on the consent and retention page.

Document progress also happens to be one of the more useful engagement signals, which is why it feeds the risk indicator described below.

The drop-out indicator reads behaviour: whether the candidate replies and how quickly, whether the resignation was confirmed as submitted, whether documents are being uploaded, whether calls are being answered, whether the tone of replies has moved from specific to vague, and whether a counter-offer has been disclosed.

It does not read the candidate's current employer, their salary, their location, their gender, their age, their college, their community or any other attribute of who they are. Those are the inputs a model would find statistically convenient and they are precisely the inputs that turn a retention prompt into a discrimination problem. The distinction is deliberate and it is a design constraint rather than a setting.

The signals are shown as signals. A recruiter sees that the candidate has not replied in nine days and has uploaded none of the four documents requested, alongside the state of the indicator, so the reasoning is inspectable rather than asserted.

This needs saying plainly. A raised risk indicator does not withdraw an offer, does not notify anyone automatically that a candidate is unreliable, does not write anything to the candidate's permanent record beyond the observed events, and does not feed any scoring that affects the candidate in a future process. It puts a task on a recruiter that says call this person today.

Human judgement is what resolves a wobbling offer. A candidate who has gone quiet may be dealing with an obstructive employer, a family matter, or a genuine change of mind, and only a conversation distinguishes them. The product's job is to make sure the conversation happens in week three rather than on the joining date.

For the same reason the indicator is not exposed to the client as a verdict about the candidate. What the client receives is a factual status and, where warranted, a flag that the recruiter is following up.

Two audiences need to know different things. The recruiter needs early, granular alerting: no reply in seven days, documents outstanding at the midpoint, a counter-offer disclosed. Those arrive as tasks with the candidate, the offer and the last exchange attached.

The client needs a smaller, later and more careful signal. A hiring manager who receives a raw risk score will start hedging, and hedging clients cause the drop-outs they are worried about. What the client sees through the portal is joining-date confirmation, document completion status and, where the recruiter judges it warranted, a stated flag that joining is at risk with a recommended action such as an early welcome call from the hiring manager.

That last intervention is one of the more effective ones available, and it is a reason to give the client a controlled signal rather than none at all.

Some will, and the system should make that recoverable rather than merely recorded. A drop-out closes the placement with a coded reason, reopens the requisition with the previous shortlist intact, and links the replacement search to the original placement so any rebate or replacement obligation is tracked to its conclusion.

The coded reasons matter for the same argument made about dispositions elsewhere on this site. Counter-offer accepted, competing offer accepted, relieving refused, personal reasons and no contact are different failures with different fixes. Free text cannot be counted across a year; a coded list can, and it will tell you whether your problem is the offer, the timeline or the brief.