Trust Score Algorithm

The Trust Score is Sentry Analytics' proprietary algorithm that rates employer reputation on a scale of 0-100.

🎯 Overview

The Trust Score aggregates data from multiple sources and applies a 5-pillar weighted formula to produce a single, actionable metric.

Trust Score = Rating Quality (30%) + Sentiment (25%) + Volume (20%) + Consistency (15%) + Recency (10%)

📊 The 5 Pillars

1. Rating Quality (30%)

What it measures: Average star rating from review platforms

Calculation:

# Normalize 1-5 star rating to 0-100 scale
rating_score = ((avg_rating - 1) / 4) * 100

# Example: 3.8 stars
score = ((3.8 - 1) / 4) * 100 = 70

Data sources:

• Kununu (1-5 stars)
• Google Reviews (1-5 stars)
• Indeed ratings

2. Sentiment Analysis (25%)

What it measures: Percentage of positive reviews (AI-analyzed)

Calculation:

# Count reviews by sentiment
positive = count(reviews WHERE sentiment = 'POSITIVE')
total = count(reviews)

sentiment_score = (positive / total) * 100

# Example: 60% positive reviews
score = 60

AI Model: Gemini 2.5 Flash categorizes each review as:

• POSITIVE - Employee recommends the company
• NEGATIVE - Employee has significant concerns
• NEUTRAL - Mixed or factual without strong opinion

3. Review Volume (20%)

What it measures: Total number of reviews (more = more reliable)

Calculation:

# Logarithmic scale (diminishing returns)
if reviews >= 1000: score = 100
elif reviews >= 500: score = 90
elif reviews >= 200: score = 80
elif reviews >= 100: score = 70
elif reviews >= 50:  score = 60
elif reviews >= 20:  score = 50
else: score = 30  # Too few reviews

Rationale: A company with 500 reviews is more trustworthy than one with 10, but 5000 vs 500 makes less difference.

4. Consistency (15%)

What it measures: How stable are the ratings (low standard deviation = high consistency)

Calculation:

# Standard deviation of ratings
std_dev = calculate_std_dev(all_ratings)

# Lower std_dev = higher score
if std_dev < 0.5:   score = 100  # Very consistent
elif std_dev < 0.8: score = 80
elif std_dev < 1.0: score = 60
elif std_dev < 1.5: score = 40
else: score = 20  # Highly polarized

Rationale: A company with consistent 3.5★ reviews is more predictable than one with 50% 5★ and 50% 1★.

5. Recency (10%)

What it measures: How fresh is the review data

Calculation:

# Days since most recent review
days_old = (today - last_review_date).days

if days_old <= 7:    score = 100  # Very fresh
elif days_old <= 30: score = 80
elif days_old <= 90: score = 60
elif days_old <= 180: score = 40
else: score = 20  # Stale data

Rationale: Recent reviews reflect current company culture; old reviews may be outdated.

🚦 Risk Levels

The Trust Score maps to risk levels:

Score Range	Risk Level	Meaning
80-100	🟢 LOW	Excellent employer reputation
60-79	🟡 MEDIUM	Good reputation, minor concerns
40-59	🟠 ELEVATED	Mixed reviews, investigate further
20-39	🔴 HIGH	Significant concerns
0-19	⚫ CRITICAL	Serious red flags

📈 Example Calculation

Company: BMW Germany

Pillar	Raw Data	Score	Weight	Weighted
Rating Quality	3.97★ average	74	30%	22.2
Sentiment	50% positive	50	25%	12.5
Volume	5,621 reviews	100	20%	20.0
Consistency	0.9 std dev	60	15%	9.0
Recency	3 days old	100	10%	10.0
TOTAL				73.7

Final Trust Score: 74/100 (MEDIUM risk)

🔄 Automatic Updates

Trust Scores are recalculated when:

1. New reviews are scraped (triggers immediate recalculation)
2. Manual refresh via dashboard or API
3. Weekly maintenance job refreshes stale data

📊 Score Breakdown API

Get detailed breakdown via API:

curl "https://sentryanalytic.com/api/company/bmw/trust-score"

Response:

{
  "company_name": "BMW",
  "trust_score": 74,
  "risk_level": "MEDIUM",
  "breakdown": {
    "rating_quality": 22.2,
    "sentiment": 12.5,
    "volume": 20.0,
    "consistency": 9.0,
    "recency": 10.0
  },
  "data": {
    "avg_rating": 3.97,
    "positive_percent": 50,
    "review_count": 5621,
    "std_dev": 0.9,
    "last_review": "2025-12-25"
  }
}

🆚 Comparison with Competitors

Feature	Sentry Analytics	Glassdoor	Kununu
Multi-source data	✅ 4 sources	❌ Single	❌ Single
AI sentiment analysis	✅ Gemini 2.5	❌ None	❌ Basic
Transparency	✅ Full breakdown	❌ Hidden	❌ Hidden
Recency weighting	✅ Yes	❌ No	❌ No
Real-time updates	✅ On-demand	❌ Delayed	❌ Delayed

---

The Trust Score algorithm is continuously improved based on user feedback and data quality analysis.