Get early access
  1. Home
  2. Knowledge
  3. From longlist to shortlist, quickly and fairly

From longlist to shortlist, quickly and fairly

Updated

A longlist of about 200 candidates is too long to read properly and too important to skim. This is a way through it that is quick without being arbitrary: decide what counts before you read, look for evidence one requirement at a time, calibrate early, and keep every decision your own.

Decide what counts before you read

Write the requirements down before you open the first profile, and split them in two: what must be true, and what would be nice. Then write down what counts as proof. "Experience as a controller in manufacturing" is a requirement. "A job titled controller at a manufacturer, for at least three years" is the evidence you're looking for.

It sounds like paperwork, but it protects you from yourself. In a well-known 2005 experiment, participants redefined the criteria for a job to fit the applicant they already preferred. When they had to commit to the criteria before they knew who the applicants were, the discrimination disappeared. Research on selection points the same way. In a large 2022 review, structured interviews, where everyone is assessed on the same points, were the method that best predicted how people did in the job.

Evidence, one requirement at a time

Don't read a profile as a whole. Read it as answers to your requirements. For each one: which job in the history shows it? For how long? How recently? Hold back the overall verdict until you've been through the requirements. Daniel Kahneman and two co-authors write that intuitive judgment gets much more accurate when the global evaluation waits until the end of a structured process.

Three rules keep the review honest:

  • Evidence is a specific job. "Controller at an industrial company, 2019 to date" is evidence. A headline like "Finance professional" isn't.
  • Missing data isn't a no. If languages or education aren't on the profile, the answer is "to clarify", not "missing". You check it once the candidate is on the shortlist, or in the first call.
  • A no has to be explainable. If you can't point to the requirement the candidate doesn't meet, it may not be a no.

Calibrate on a few yes and no

Before you start on the full list, take three candidates: the strongest, and two that differ from it in level, industry or type of company. Say yes or no to each. If you're unsure, show the three to your client or the hiring manager before you read on. A quarter of an hour here beats reviewing 200 profiles against a picture of the role nobody else shares.

Then adjust the requirements to match your answers. If you said no to someone who meets every requirement, a requirement is missing. If you said yes to someone who misses one, it wasn't a requirement.

Fast, without getting sloppy

  • One decision per profile. Yes, no, or skip. Don't build a "maybe" pile that keeps growing.
  • Skip when you're unsure. Come back to them at the end. Most borderline cases settle themselves once you've seen the rest.
  • Compare with the requirements, not with the last candidate. A large study of real admission and hiring interviews found that evaluators measured each candidate against the one just before. After a strong candidate, the next one looks weaker than they are.
  • Don't think in quotas. Interviewers who had already recommended several people that day were less willing to recommend the next one. If ten are good, ten are good.
  • Work in blocks. Take 50 at a time, say, and break in between. The evidence for decision fatigue is disputed, but a break makes it easier to start again from the requirements.
  • Use the keyboard. The fewer clicks between two profiles, the more attention is left for the profile itself.

Write down why when you say no

A reason takes two seconds and pays off three times. It keeps you consistent, because you use the same few reasons every time: wrong level, wrong function, wrong industry, wrong location, a specific requirement missing. It makes it easy to explain the list to your client. And it lets you answer a candidate who asks why they didn't go further.

Many rejections for the same reason also tell you something about the search. If half are rejected on level, it's the search that needs fixing, not your judgment.

Fair: what not to look for

Denmark's Act on Prohibition of Discrimination in the Labour Market forbids employers to discriminate against applicants when hiring on grounds including race, colour, religion or belief, political opinion, sexual orientation, gender identity, age, disability, and national, social or ethnic origin, and the Equal Treatment Act does the same for gender. The ban also covers anyone who places people in jobs, so it covers agencies. Apply the same rules when you assess people who never applied.

Discrimination is rarely deliberate, which is why structure is the best defence. In a Danish field experiment with 800 applications to 400 real jobs, applicants with Middle Eastern-sounding names had to send 52% more applications than applicants with traditional Danish names to be invited to interview as often, with the same qualifications. Stick to the evidence per requirement, and watch for things that can stand in for a protected characteristic: a name, a photo, a graduation year, gaps in a career, and vague words like "chemistry" and "culture fit".

Keep the decision human

A tool can rank, but a person should decide. The GDPR gives people the right not to be subject to a decision based solely on automated processing that significantly affects them, and it names e-recruiting without human intervention as its example. The EU AI Act warns against relying too much on a system's output.

In practice: don't reject a block of candidates because their scores are low. Look at the ones just below the line, and bring back candidates the system filtered out when you can see they fit.

This is how the review in Navigent is built. Calibration shows three candidates at a time for up to three rounds, and your answers become the examples everyone else is scored against. The review card shows the band, Strong, Good or Possible, one evidence sentence per requirement citing the jobs it rests on, and "To clarify" for what search data can't answer. You decide from the keyboard: → or A for yes, ← or R for no, ↓ to skip and U to undo. Every yes and no is yours, and every decision is saved.

See what a review card looks like on How it works.

Sources

Get access when we open

Navigent opens soon. Sign up, and we'll write to you as soon as you can get started. The first 10 paying customers get 30% off for 12 months.

So we know who we're writing to.

We'll write here when Navigent opens.

The recruitment firm or company you recruit for.

We only use your details to write to you about Navigent opening. Read more in our privacy policy.