How AI CV Scoring Works: From Inbox to Ranked Shortlist

By Serhii Baklanov5 min read
  • AI CV scoring
  • AI resume screening
  • ATS
  • shortlisting

What Problem Does AI CV Scoring Solve?

A recruiter at a small agency might receive 60 to 200 applications for a single active role. Reading every CV end to end takes hours — and the quality of that reading degrades as the pile grows. The 40th CV gets a fraction of the attention the first one did.

AI CV scoring does not replace that judgment. What it does is compress the triage stage: instead of reading 60 CVs cold, you read the top 12 flagged by the system and make your own call from there.

How a CV Enters the System

In OctoRecruit, CVs reach the system in three ways:

  • Email inbox sync: when a Gmail or Outlook mailbox is connected, the system polls for new messages. Any message containing a CV attachment (PDF, DOCX, or RTF) is pulled in automatically. The candidate's name and email are extracted from the email headers.
  • Manual upload: a recruiter uploads a file directly from the candidate form.
  • Public apply link: each job has a unique apply URL. A candidate fills in their details, uploads their CV, and submits. The application lands directly in the pipeline.

In all three cases, the CV text is extracted and stored against the candidate record before scoring begins.

What the Model Actually Evaluates

The model does not read a CV the way a human recruiter does. It processes the extracted text as a structured prompt. That prompt includes:

  1. The candidate's CV text (plaintext extraction from PDF/DOCX).
  2. The job description — title, required skills, responsibilities, and any custom criteria the recruiter added.
  3. A scoring instruction that defines what 1–10 means for this specific role.

The model then returns a numeric score and a short plain-language reason — typically two or three sentences. For example: "Strong Python and SQL background with five years in fintech. Missing the required project management experience. Shortlistable if the team can provide PM oversight."

The score is stored on the candidate record alongside the reason, and both are immediately visible on the Kanban board card and in the candidate detail panel.

The Five-Axis Deep Analysis

For shortlisted candidates, recruiters can request a deeper analysis that evaluates the CV against five axes:

  • Skills match — how well the technical and soft skills align with the job requirements.
  • Experience relevance — industry background, seniority level, and role similarity.
  • Cultural indicators — communication style, career trajectory, and tenure patterns.
  • Growth potential — evidence of increasing responsibility or lateral skill development.
  • Red flags — unexplained gaps, unusually short tenures, or qualifications that cannot be verified.

Each axis gets its own score and commentary. This is the second tier of analysis — it runs only when a recruiter explicitly requests it, not automatically on every CV, to keep AI costs predictable.

Why a Score of 8.2 Beats 6.1 for a Given Role

The score is always relative to the job it was calculated against. A candidate with a strong data engineering background might score 8.2 for a data pipeline role and 5.4 for a sales manager role — the same person, entirely different scores.

This matters because OctoRecruit labels every score with the job it was computed for. On the global candidates list, a score always reads as "8.2 · Senior Data Engineer" rather than just "8.2". That pairing prevents a common mistake: promoting a high scorer to a new role they were never evaluated against.

Limitations to Be Honest About

AI scoring is a probabilistic tool, not a definitive gate. There are things it handles poorly:

  • Portfolio-heavy roles: a designer with a strong portfolio but a plain CV may score lower than a CV-polished candidate with weaker actual work.
  • Career changers: someone pivoting from a different industry carries transferable skills that are hard to parse from text alone.
  • Non-standard CV formats: densely formatted two-column PDFs, graphics-heavy files, and non-Latin character sets can degrade text extraction, which degrades scoring.

The practical guidance is to use the score for triage, not elimination. A candidate who scores 4.0 deserves a second look from a human before being filtered out.

How This Fits Into a Real Recruiting Workflow

The practical workflow in OctoRecruit looks like this:

  1. A new CV arrives via email sync or apply link.
  2. The system scores it against the active job automatically.
  3. The score and reason appear on the pipeline card within seconds.
  4. The recruiter reviews the top-scored candidates, reads the AI reason, then opens the actual CV for the ones worth a closer look.
  5. For shortlisted candidates, the recruiter triggers the deep five-axis analysis before scheduling a call.

The inbox-to-shortlist time — which in a manual workflow can be two to three days — typically compresses to a same-day review cycle.