A year of feedback from recruiters and site leads comes in as tickets, calls and comments left on shortlists. Read together, it is less varied than you would expect. The same few problems came back from care homes, warehouses, contact centres and shop floors.

Very little of it asked for more automation. Most of it asked for the same thing in different words: read everyone, show the reasoning, and let a person decide. What follows are the five themes that came up most, from five teams that hire at volume.

The backlog was the shortage

Almvik Care staffs night shifts across its care homes. For a long time the assumption was that nobody wanted those shifts. The head of staffing put it more precisely: “Night shifts used to sit open for three weeks. It was the screening backlog holding them up, not a shortage of applicants.”

That pattern turned up well beyond care. Applications arrive faster than anyone can read them, so they get read in the order they came in until the week runs out. The applicant who fits a night rota may be number two hundred in the queue.

The fix Almvik asked for was not a faster reader. It was a complete one. Every applicant is read against the brief, and the site lead sees a ranked shortlist with the reasoning next to each name. The shift question stops being who applied first and becomes who can work the rota.

Describe the work, and the CV stops being the gate

Norhem hires for warehouse roles, and many good applicants have no CV to upload. The old process filtered on one anyway. The operations lead described what changed: “The brief describes the work now, not the CV, and my site managers trust the shortlist that comes back.”

Two things made that work, and both came out of feedback from site managers:

• The brief names the job itself. Lifting, shift patterns, a forklift licence where the role needs one.

• Structured questions replace the upload. Interview agents ask every applicant the same questions, so someone without a CV still has something to be read on.

If you hire for roles like this, the useful test is simple. Could a strong applicant with nothing to upload reach your shortlist? If not, the brief is screening on a document the job does not need.

One rubric has to survive six languages

Vessik runs contact centres in six markets. Its recruiters were clear from the first month that translation alone was not the goal. Each market had drifted into its own questions and its own sense of a good answer.

The contact centre manager summed up the fix: “Every applicant answers the same questions in the language they applied in. Six markets, one rubric, one shortlist.”

What made recruiters trust it was the evidence link on every rating. A reviewer in one market can open a rating from another and read the answer it came from. Disagreements became specific. Instead of “this market scores too high”, a recruiter could point to a line in the rubric and ask for it to change.

Every applicant is owed an answer

Tarnow’s feedback was about the people who do not get hired. Seasonal retail hiring means a thousand applicants for a much smaller number of places, and most of them used to hear nothing.

The recruitment lead named it directly: “A thousand seasonal applicants, and every one of them hears back in the first round. That was not true last year.”

A first-round answer is not a job offer, and nobody at Tarnow treats it as one. It is a reply, sent after a person approves it. The approval step on every send mattered to Tarnow’s store managers. A message goes out only when someone on the team has signed it off, even at seasonal volume.

Oversight is something you show

Vantera’s works council wanted to know how AI screening would be used before it was switched on. Their questions set the tone for many conversations since.

“The council never asked whether the tool was clever. They asked who presses send, and whether they could read the reasons afterwards.”

HR business partner, Vantera

The answers they needed were concrete. A named person approves every decision. The bias test report is produced for each role rather than once for the product. The audit export lets someone outside the tool read what was assessed and what came back. Candidates are told when screening is used. Data stays in the EU.

None of this is legal advice, and a works council’s questions depend on the country and the agreement in place. The general lesson holds, though. If you cannot show oversight, a council will reasonably treat it as absent.

What the feedback did not settle

Not every theme has a clean answer, and a year of feedback also showed where the limits are.

• The first brief is rarely the last. Several teams said their first shortlist was wrong in a consistent way. Rewriting the brief fixed it, but it took a second pass.

• Site leads have to read the reasoning. A ranked list is quick to accept without opening the evidence. The teams that got the most from it made reading the case part of the review.

• Screening does not hire anyone. It gets the right people to an interview sooner. The interview, the reference and the decision still belong to a person.

If you are starting out, those three points are worth more than any feature list. Write the brief as if a colleague will argue with it. Read the case before you approve a name. Keep the decision where it belongs.