A typical engineering loop runs seven to nine interviews over a month.
It still produces decisions the panel cannot explain a week later.
I defined the hiring process for a team at Zalando that now covers 27 markets. Three changes did most of the work.
Change 1: cut the loop to four rounds
Seven rounds is not seven times the signal. It is mostly the same signal, measured again by tired people.
Four rounds cover it.
Screen, 30 minutes. Experience, motivation, level. Does this person want this job or any job.
Practical coding, 60 minutes. A problem from the actual codebase, stripped down. Let them use AI, documentation, whatever they use at work. You are hiring the person who will work that way on Monday.
System or technical depth, 60 minutes. One design problem with clear constraints, scored against criteria written in advance.
Hiring manager and team fit, 45 minutes. How they work with people, how they handle disagreement, what they need from a manager.
Everything a fifth round would add is already in those four, or it is noise.
Change 2: run it in one day
A loop spread over three weeks loses candidates to companies that decided faster.
One day, on site or remote, with breaks. The panel meets the same afternoon while the memory is fresh.
The cost is coordination. The benefit is that you stop losing the people you actually wanted, and your engineers spend fewer total hours because nobody re-reads notes from twelve days ago.
Vacancy cost runs around 1,000 euro a day for an open engineering role. Three extra weeks of process is not free.
Change 3: scorecards before the debrief
This is the change with the highest return and the lowest cost.
Every interviewer submits their score and evidence before anyone speaks. No exceptions, including for the most senior person in the room.
Opinions contaminate each other within seconds. Whoever speaks first sets an anchor and the rest of the panel drifts toward it without noticing.
When the scores are already written, the debrief becomes an examination of disagreement rather than an averaging exercise.
What this does to the numbers
Structured assessment has about 0.51 validity against job performance. Unstructured conversation has about 0.20.
Cutting from nine interviews to four saves roughly 5 to 10 engineering hours per candidate. Across five hires that is close to 288 hours back.
The point is not the saving. The point is that you get better decisions and the saving comes along with it.
About AI in the coding round
Stop trying to catch candidates using AI.
Give them a task that requires it. Watch what they ask it, what they accept without checking, and how they validate the output.
An engineer who prompts carelessly and ships the result is a real risk you can now observe directly. That is a far stronger signal than whether they could write a binary search from memory.
What most teams get wrong when they try this
They change the process without changing the interviewers.
More than 80% of people involved in hiring have had little or no formal training. Being a strong engineer does not make someone a strong interviewer. Building systems and evaluating people are different skills.
If you shorten the loop without calibrating the panel, you have just made a noisy process faster.
Where to start
Pick one change. The scorecard rule is free and you can apply it to the next debrief.
Then look at the loop length, then at interviewer training.
If you want the whole thing done properly, that is what SmartHiring is for. Process audit at 500 euro, workshop sprint at 2,500.
The short version
Four rounds, one day, scorecards submitted before anyone talks.
Train the interviewers or the rest is decoration.