“Three gaps on the early shift. Who can I call?”
Illustrative firm, not a client
A worked example: one agency over three months.
This agency isn't a client. It's put together from patterns that come up again and again in agencies placing temporary workers at volume.
The agency.
Places: Warehouse and light industrial workers
Size: About 60 staff, 22 of them recruiters
On assignment: About 900 workers in a typical week
Clients: One large client brings in about two fifths of revenue
Holding it together: A duty manager, a payroll lead and two senior recruiters
The month before.
“That's the third short shift this month. We need to talk about the contract.”
“Fifty timesheets still missing, and payroll closes at six.”
“Placed four people today. Spent the evening on paperwork.”
What the review found.
The review covered one month of records and three conversations. Ranked by cost, the leaks were:
- 1. Short shifts, because confirmations came too late
- 2. Timesheets missing at payroll cutoff
- 3. New starters leaving in their first fortnight
- 4. Worker documents about to expire
- 5. Invoice lines that didn't match the client's hours
- 6. Duplicate candidates across recruiters
AI would help with confirmations, timesheets and new starter check ins. It wouldn't help with deciding which worker suits which site, which stays with the recruiters, or with the renewal conversation, which stays with the owner.
Why shift confirmation came first.
Everyone wanted better candidate matching. But short shifts were costing the most, they were damaging the largest client relationship, and progress would show within a week. So that's where it started, and matching came third.
Three months on.
Duty manager
Used to call down a list at 5 AM.
Now starts with two gaps flagged and replacements suggested.
Recruiters
Used to confirm their own workers every evening.
Now spend that time with candidates and clients.
Payroll lead
Used to chase timesheets on closing day.
Now works from a short list, two days early.
The result: shifts started full far more often, fewer people left in their first fortnight, the largest client renewed, and nobody's job was cut.
What went sideways.
A fifth of worker phone numbers were out of date. Two weeks of cleanup came before anything went live.
Some workers found the check ins too frequent. After asking them, it changed to one message the evening before, and none at dawn unless they hadn't replied.
One site supervisor kept his own list. Rather than replace it, the workflow reads it.
What came next.
Timesheet collection was added once shift confirmation had held for a quarter. New starter check ins followed.