How to measure candidate experience, and what to do with the score

• Measure candidate experience by sending one rating question after every decision, to everyone, and segmenting the answers by the stage they reached and whether they were hired. The arithmetic is the net promoter calculation: scores of 9 and 10 are promoters, 7 and 8 are passives, 0 to 6 are detractors, and candidate NPS is the promoter percentage minus the detractor percentage.

• The formula is the easy half. What makes measurement hard is that a single company-wide score is almost uninterpretable, and the published benchmarks show why.

Two benchmark sets that disagree

• Ashby's overall average of 37 and Starred's -23 for one rejected segment describe different populations. Ashby's figures come from 11.9 thousand responses to 67.4 thousand requests sent between January 2022 and October 2023, with every outcome mixed together. Starred's segmentation isolates the rejection stages that an all-respondent average absorbs.

• Both agree on the direction that matters. Hired candidates score in the eighties, rejected candidates near or below zero. Any company-wide number you publish is therefore mostly a report on your hire-to-reject ratio and your survey response mix.

• So report a small grid instead of one number.

The grid worth reporting

• Segment on two axes: the stage the candidate reached, and the outcome. Six cells cover most processes.

1. Rejected at application review.
2. Rejected after a screen.
3. Rejected after an assessment.
4. Rejected after a final interview.
5. Withdrew.
6. Hired.

• Each cell has a different owner and a different fix. A poor score in cell 1 points at the advert, the form and the speed of the rejection, which is where [your careers page](/spaces/careers-page-conversion-problem) sits. A poor score in cell 3 points at an assessment that asks too much for the information it returns. A poor score in cell 5 usually means people found out something late that they should have known early, such as the salary or the on-site requirement.

Getting people to answer

• Ashby's data settles two arguments that consume a lot of meeting time.

• Survey length barely matters. One to five questions returned 20.1%, six to ten returned 20.1%, and eleven or more returned 18.2%.

• Anonymity does matter, and in the direction that surprises people. Non-anonymous surveys returned 20.6%, optional anonymity 17.7%, and enforced anonymity 15.3%. Enforced anonymity also strips out the stage, recruiter and role attribution that makes the grid above possible.

• Timing decides when you can act on any of it. Ashby found 94% of responses arrive within nine days of the survey going out. Close the survey at fourteen days, report, and move on.

Delay costs more than wording

• Ashby measured candidate NPS against time in process. Past 60 days, the score for rejected candidates fell by 54.9%. For hired candidates it fell by 16.5%.

• The people you turn down carry the full cost of a slow process. That makes cycle time a candidate experience metric as much as a recruiting one, and it is why the rejection email is less important than the date it goes out. Wording is covered in [rejection emails that protect your employer brand](/spaces/rejection-emails-employer-brand).

Pair the score with rates you already have

• A score on its own gives you a mood. Put it next to three operational numbers and it gives you a cause.

• Apply completion rate tells you whether the form is the problem. CareerPlug's 2024 analysis of more than 10 million applications across over 60,000 small businesses found an average click-to-apply rate of 6%, so most careers pages lose the majority of interested people before the survey is ever sent.

• Time in stage tells you where people wait. Median days from application to first response is the single number most correlated with the complaints you will read in the free-text box.

• Applicants per hire tells you how many rejections your process produces. The same dataset gives an average of 180 per hire, with 3% of applicants reaching interview. If your screening discards people who could do the job, you generate more detractors and more advertising spend to make the same hire. That mechanism is in [the cost of ATS false rejections](/spaces/cost-of-ats-false-rejections).

What to do with a bad cell

• Route the number to the person who can change it, with the free-text comments attached.

• Application-stage scores go to whoever owns the careers page and the auto-responses. Assessment-stage scores go to the person who designed the exercise, with the question of whether the exercise earns the hours it asks for. Final-stage scores go to the hiring manager, because at that point the experience is a person, not a system. Withdrawals go to whoever writes the job advert, since most withdrawals trace to a fact discovered late.

• Set one target per quarter, on one cell. A process that improves its assessment-stage score from -23 toward zero has done something real and can prove it.

A 30-day setup

1. Add one 0 to 10 question plus one free-text box, fired on every status change to rejected, withdrawn or hired.
2. Attach stage, role, recruiter and outcome to each response. Skip enforced anonymity.
3. Close each survey at fourteen days.
4. Report the six-cell grid monthly, never a single company average.
5. Put median days-to-first-response next to it.
6. Pick one cell and one owner for the next quarter.

• Ranking candidates on evidence they can do the work, which we describe in [hiring on proof of fit](/spaces/hiring-on-proof-of-fit), tends to move several cells at once, because fewer capable people get rejected by a rule nobody would defend out loud.