Most descriptions of the staffing industry are about recruitment: finding people, matching skills, building candidate databases. That is the part that looks like a business model.
The part that decides profitability doesn't look like anything at all. It happens on the phone between five and seven in the morning, and it consists of one person calling down a list.
05:40 – the absence comes in
A care worker reports that she will not be in for the day shift. The message goes to the agency's on-call desk, not to the nursing home, because it is the agency that promised the shift would be covered.
From that moment there are 80 minutes until the shift change. In that window somebody has to find a qualified substitute, reach her, get a yes, and still notify the client before the first resident sits down to breakfast.
No part of that task is professionally difficult. All of it is time-critical.
05:55 – calling down the list
Whoever is on the on-call phone has a list sorted by qualification, geography, who worked most recently, and who has said they are happy to pick up extra shifts.
Then the calling starts. Most people don't answer, because it is six in the morning. Some answer and say no. Some answer and say yes, but can't be there before half past eight.
What actually happens in that hour is that one person makes twenty attempts one after another, because she can only talk to one person at a time. Each attempt takes between forty seconds and four minutes. The bottleneck is not how many candidates the agency has. It is that the phone is a serial channel.
06:40 – the calls nobody picks up
While she is calling out, other people are calling in.
A second client reports an absence of their own. A candidate returns a missed call from yesterday. A nurse in the middle of an assignment has a question about a timesheet.
None of them gets an answer, because the one person who could have answered is on a call about a different shift.
The first cost is easy to count: a shift that went uncovered. The second is invisible in the accounts. The candidate who called back was never reached, and took a shift with a competing agency instead. The client who got no answer called the second name on their list.
What that one shift is actually worth
| Outcome | Immediate effect | What costs more |
|---|---|---|
| Shift covered before 07:00 | Billable shift | Nothing |
| Shift covered at 09:00 | Partly billable | The client has already sorted it internally |
| Shift not covered | No revenue | The framework agreement gets reviewed at the next tender |
| Candidate never reached | No revenue | The candidate takes their next shift somewhere else |
The bottom row is the most expensive one, and the only one that isn't recorded anywhere. An agency rarely loses a framework agreement over a single uncovered shift. It loses it over a pattern of them across six months that nobody could see while it was happening.
Three streams of calls, one person
Where an online shop or a garage mostly receives enquiries, three streams run at once in a staffing agency, and they carry very different levels of urgency.
Absences coming in from clients. Short, formal, extremely time-critical. Always in clusters, always at the same times of day.
Outbound calls to candidates. Many attempts, a low answer rate, and the entire value lies in how many get made per hour.
Ongoing questions from people on assignment. Timesheets, expenses, when payroll lands. Not urgent, but they don't go away – they simply come back later in the day, and they consume the same person.
The three compete for one resource, and that resource is the phone line of one person at six in the morning.
What can be done in parallel, and what shouldn't be
Outbound requests to candidates are the part that takes automation best. An agent can call twenty candidates in parallel rather than one after another, read out the shift, take a yes or a no, and stop calling the rest the moment somebody says yes.
That is not primarily a language problem. It is a concurrency problem – and it is the one part of the morning where a machine has a property a human cannot match.
Inbound absence reports take it well too, provided the agent can write straight into the scheduling system. An absence report is four facts: who, which shift, which site, from when.
Two things it should not do.
It should not decide who is allowed to take the shift. The labour-hire rules and licensing requirements sit with the agency no matter who places the call, and rest periods, qualification requirements and professional registration are not things a voice agent should be weighing up mid-conversation. They belong in the filtering of the list before the agent ever gets it. If the agent calls on the wrong basis, the agency hasn't saved time – it has moved an error closer to the customer.
It should not negotiate either. A candidate who says "I can, but only if I'm off the Thursday evening shift" has opened a conversation about terms. That should go to a human with everything said so far, not be answered with a guess.
The goal is not fewer calls
It is tempting to measure this as an efficiency project: calls handled per morning.
That number will look good almost regardless, and it says very little about what actually got resolved. The staffing industry already has a better number sitting in the scheduling system: the share of absences reported before six that are covered before the shift change. That is the client's number too. It is the one they remember at the next tender.
One more is worth tracking alongside it: how many candidates were attempted and not reached. That is the agency's own leakage, and today it rarely appears in any report.
A note on the outbound call
When an agent calls a candidate, it has to disclose that it is AI. The obligation applies from the opening line, and it is easier to live with here than almost anywhere else. A candidate woken at six cares very little about who is asking. She cares which shift it is, where it is, and when she has to be there.
It is the same observation that holds at the other end of the phone: what people react to is rarely who called. It is whether they got enough to make a decision.
The hour that can't be staffed
A staffing agency cannot put more people on the on-call phone at 05:40. The volume only exists in that hour, and a role that is open from 05:30 to 07:00 is not a role.
That is why this is one of the few industries where the question is not whether automation can make something cheaper. It is whether anything at all can be done at the same time.
Threll.ai builds voice agents in Norwegian, Swedish and Danish. The shift that got covered at 06:50 is not a saved cost. It is an invoice that otherwise wasn't sent.
