To rank resumes, define the requirements first, mark each as must-have, weighted or knockout, score every application against the same list with evidence, and sort by the result. A rank is a call order for a human, not a hiring decision: ties, missing information and borderline cases go to a person. This holds for spreadsheets and for AI tools alike. Status: November 2026.
Points or grades: pick one method and keep it
| Method | How it works | Good for | Weak spot |
|---|---|---|---|
| Weighted points | Each requirement gets a weight (sum 100); partial credit allowed | Many applications, one clear ranking | Looks more exact than it is; ties are common |
| Grades (A/B/C) | Rules such as "all must-haves met = A; one missing = B" | Small pools, hiring-manager hand-off | Coarse: many people share one grade |
| Pairwise comparison | Reviewers compare two candidates at a time | Top 5 to 10 only | Does not scale; hard to document |
For most teams the best setup is a hybrid: knockouts first, grades from must-have rules, points only to order people inside a grade. How to write the requirements themselves is covered in CV screening criteria.
A worked example, with ties and an open point
Assumptions (ours, for illustration): role "Customer Support Agent". Knockout: right to work in the country. Weighted requirements, 100 points: support experience of 2+ years (40), CRM tool (20), English B2 (20), weekend shifts (20). Partial credit is half the points.
| Candidate | Experience (40) | CRM (20) | English (20) | Shifts (20) | Total | Note |
|---|---|---|---|---|---|---|
| A | 40 | 20 | 10 (B1) | 20 | 90 | Highest confirmed score |
| B | 40 | 0 | 20 | 20 | 80 | Trainable on CRM |
| C | 20 (1.5 yrs) | 20 | 20 | 20 | 80 | Less experience |
| D | 40 | 20 | 20 | unclear | 80 + open | Ask about shifts |
| E | Knockout: no right to work (confirmed in the application) | Out | ||||
Three things this shows. Ties are normal: B, C and D all have 80, so the score cannot decide. Use a written tie-break such as "more must-haves met with evidence first". Unclear is not zero: D's shift answer is missing, so the 20 points are open, not lost; one question resolves it. And the rank orders the calls: ask D first, because a yes makes 100 and puts D ahead of A; then A, B and C. Nobody is rejected because of 80 versus 90.
How many go on the shortlist?
Size the shortlist by capacity, not by score: how many interviews can the team run this week? Take the top group plus everyone with an open point that one answer could resolve. Keep the rest ranked, not deleted, so a withdrawn candidate does not restart the search.
How AI ranking works, and what to ask the vendor
Most tools parse the CV, compare it with the job description or your criteria using language models, and output a score or a sorted list. Vendor documentation commonly describes criteria with weights from 1 to 10, matched text highlighted, and a note that a human should review. Independent evidence on accuracy and fairness is thin, so ask:
- What is scored? Your written requirements, or a hidden "fit" the vendor defines?
- Is every score explainable? Can you see the sentence in the CV behind each point?
- What happens with missing information? Counted against the person, or marked open?
- Who sets the grade? Your rules, or a number the model produces?
- Has it been tested for bias? Ask for the method: same CV, different names, same result?
- Is there an automatic rejection? There should not be.
- Where is the data processed and stored, and for how long?
For the wider picture see AI resume screening: what works and what fails and, for tools that only display a score, candidate matching software.
The legal limits of ranking
- GDPR Art. 22: a decision based solely on automated processing with legal or similarly significant effect is restricted (GDPR). In C-634/21 (SCHUFA, 7 December 2023) the Court of Justice held that even an automatically generated score can count as a decision when a third party relies on it heavily. If a recruiter just follows the ranking, you are close to that line. The human must have real authority to deviate.
- EU AI Act: AI that analyses and filters applications is high-risk under Annex III no. 4 (Regulation (EU) 2024/1689). Deployer duties include human oversight (Art. 14, Art. 26). The high-risk obligations apply from 2 December 2027 (Regulation (EU) 2026/1744); see EU AI Act in recruiting.
- AGG: if a rejected applicant shows indications of discrimination, § 22 AGG shifts the burden of proof to you. A ranking you can explain line by line is your evidence.
- Works council: in Germany, AI-based selection guidelines are subject to co-determination (§ 95 para. 2a BetrVG).
This is not legal advice; involve your data protection officer.
Ranking in Sprad
In Sprad's CV screening, requirements are marked must-have, nice-to-have or exclusion and weighted. Each is judged "certain" or "likely", assessed three times independently, and backed by the exact sentence from the CV or conversation; no quote means "unclear", and unclear never counts against the person. The grade (A, B, C, or up to five levels) follows rules your team sets, for example how many must-haves may be open. No model assigns it. Every rejection waits for confirmation by a person. Sprad is the provider of this page. CV review costs 2 credits per application (list price €0.10 per credit, status 29 October 2026; see pricing).
FAQ
How do I rank resumes without software?
Write the requirements, give each a weight or must-have status, score every CV in one spreadsheet with a note on the evidence, and sort. Use the same sheet for every candidate.
Should I rank with points or grades?
Use grades from must-have rules to group people and points only to order within a group. Pure points overstate precision.
What do I do when candidates tie?
Write the tie-break before you screen, for example "more must-haves met with evidence first", then fewer open points. Do not decide ties case by case.
Is an AI ranking a decision under GDPR Art. 22?
It can be, if the recruiter relies on it without real room to deviate (C-634/21). Keep a human with authority in the loop and document why they followed or overrode the ranking.
Can I let AI reject the bottom of the ranking automatically?
We advise against it. Automatic rejection raises GDPR Art. 22 and AGG risk; see can AI reject candidates automatically.
Jürgen Ulbrich
Jürgen Ulbrich has more than a decade of experience in developing and leading high-performing teams and companies. As an expert in employee referral programs as well as feedback and performance processes, Jürgen has helped over 100 organizations optimize their talent acquisition and development strategies.
