Design Principle: Multi-Factor Scoring
A single semantic similarity score can be gamed (by padding resumes with job description language) and can miss important hiring signals like experience depth and role title trajectory. InnoHire.ai uses a weighted multi-factor model that blends five independent signals into one composite rank score out of 100.
The Five Scoring Factors
1. Semantic Match Score (Weight: 35%)
The cosine similarity score from the embedding stage (see Resume Matching Engine), normalised to a 0–100 scale. This is the heaviest single factor because it captures overall profile alignment.
2. Skills Coverage Score (Weight: 25%)
A ratio computed as: (matched required skills) / (total required skills). Required skills are extracted from the job description and weighted by their frequency and positioning (a skill mentioned in the first paragraph of a JD is weighted higher than one mentioned in a "nice-to-have" section).
A candidate matching 8 of 10 required skills scores 80 on this factor (before weighting). Bonus points are added for skills listed as "preferred" that the candidate also possesses.
3. Experience Depth Score (Weight: 20%)
This factor evaluates whether the candidate has enough years of relevant experience — not total career experience. It compares:
- The required experience stated in the job description (e.g. "5+ years in backend development")
- The candidate's computed relevant experience duration per domain (extracted and summed from work history)
Candidates meeting or exceeding the requirement score 100 on this factor. Candidates with under 50% of the required experience score 0–40.
4. Role Title Alignment Score (Weight: 12%)
The semantic similarity between the job title and each of the candidate's previous role titles. A candidate whose most recent title was "Backend Engineer" applying for a "Senior Backend Engineer" role scores near 95. A "Marketing Manager" applying for the same role scores near 10.
Title alignment is a lower-weight signal because title inflation and deflation are common — but it serves as a useful tiebreaker and red-flag detector.
5. Recency Score (Weight: 8%)
Skills and experience demonstrated more recently are weighted higher. A candidate who used a required technology 4 years ago and hasn't since scores lower on recency than one who used it in their last role. This penalises candidates with stale, outdated experience in fast-moving domains (e.g. ML frameworks, cloud platforms).
Composite Score Formula
CompositeScore =
(SemanticMatch × 0.35) +
(SkillsCoverage × 0.25) +
(ExperienceDepth × 0.20) +
(TitleAlignment × 0.12) +
(RecencyScore × 0.08)The composite score is then normalised on a 0–100 scale across the full candidate pool for a given job, so the highest-scoring candidate in a batch always appears as rank #1.
Rank Output
Candidates are returned sorted by composite score descending. The output for each candidate includes:
- Overall Rank — position within the evaluated pool (e.g. #1 of 47)
- Composite Score — the blended score (0–100)
- Per-Factor Breakdown — individual scores for each of the 5 factors
- Confidence Band — High / Medium / Low confidence label based on data completeness
Bias Mitigation
The ranking model explicitly excludes name, gender-coded language, educational institution prestige, and graduation year from all scoring factors. Only skills, experience, and demonstrated capability are scored.
Regular audits compare rank distributions across demographic signals to detect and correct proxy discrimination patterns.
Customising Weights
Enterprise plan customers can adjust factor weights per job role category via the InnoHire.ai dashboard. For example, a technical bootcamp role might increase the Skills Coverage weight to 40% while reducing Title Alignment to 5%. All customised weight sets must sum to 100%.