Kindworth V2: People Intelligence Platform

Technical Architecture for Workforce Analytics from Recognition Data

1. Executive Summary

Kindworth V2 extends the original KQ (Kindness Quotient) scoring system with a comprehensive suite of workforce intelligence metrics. By analyzing recognition patterns, timestamps, and network dynamics, V2 provides predictive insights for burnout risk, leadership potential, team health, and collaboration efficiency.

Key Innovation

V2 introduces the Extra Work Index (EWI)—a novel metric that quantifies work done for others beyond core responsibilities by analyzing who receives recognition and when.

2. Data Pipeline Architecture

2.1 Input Format

Kindworth processes Slack-style conversation logs in the following format:

2024-12-09 08:14:11 #deals @sal189: Kudos @sal202 for the amazing presentation #recognition #creativity

Each message is parsed to extract:

Timestamp — Date, time, hour for temporal analysis
Channel — Context for message categorization
Sender — Recognition giver
Mentions — Recognition receiver(s)
Tags — Recognition categories (creativity, teamwork, etc.)
Recognition Flag — Boolean based on keyword matching

2.2 Recognition Detection

Messages are flagged as recognition events if they contain:

RECOGNITION_KEYWORDS = ["#recognition", "kudos", "thanks", "thank you", "shoutout"]

3. V2 Metrics Framework

3.1 Extra Work Index (EWI)

The Extra Work Index quantifies "invisible" work done for colleagues. When employee A is recognized by employee B, we infer that A performed work benefiting B.

EWI(u) = Σ recognitions_received(u) × ESTIMATED_HELP_MINUTES

Where ESTIMATED_HELP_MINUTES = 10 (configurable per organization)

This is converted to hours for reporting:

ewi_hours[u] = ewi_raw[u] / 60.0

3.2 Burnout Risk Score

A multi-factor score combining three weighted components:

EWI Factor (40%)

Normalized extra work hours relative to maximum in organization

After-Hours Factor (35%)

Recognition activity outside 9am-6pm or on weekends

Overwork Factor (25%)

Total recognitions received (indicator of being "go-to" person)

Burnout_Risk = (EWI_factor × 0.4 + After_Hours_factor × 0.35 + Overwork_factor × 0.25) × 100

Score capped at 100; values > 70 indicate high risk

3.3 Flight Risk Score

Compares recent activity (last 90 days) to earlier activity to detect engagement decline:

decline = ((early_activity - recent_activity) / early_activity) × 100
flight_risk = max(0, min(100, decline))

3.4 Leadership Score

Composite score predicting leadership readiness:

Component	Weight	Description
KQ Factor	35%	Base Kindness Quotient normalized to 0-1
Cross-Dept Factor	25%	Ratio of cross-department recognitions to total
Network Breadth	25%	Departments reached / total departments
Positive Delegation	15%	Net recognition balance (positive = delegates effectively)

3.5 Bridge Score

Measures cross-functional connectivity:

bridge_score = (depts_reached / (total_depts × 2)) × 50 + 
               (cross_interactions / max_cross_interactions) × 50

3.6 Reciprocity Score

Percentage of recognition relationships that are bidirectional:

reciprocity = (reciprocal_relationships / total_relationships) × 100

3.7 Giver/Taker Classification

Type	Criteria
Giver	sent/received ratio > 1.5
Taker	sent/received ratio < 0.67
Balanced	ratio between 0.67 and 1.5
Isolated	sent < 5 AND received < 5

4. Team Health Analytics

Per-department metrics aggregated from individual scores:

4.1 Cohesion Score

cohesion = (intra_team_recognition / total_dept_interactions) × 100

4.2 Distribution Equity

equity = 1 - (max_received - median_received) / max_received

High equity indicates recognition is spread evenly across team members.

4.3 Health Status Classification

Status	Criteria
Healthy	avg_burnout < 30 AND avg_kq > 40
Warning	avg_burnout 30-50 OR avg_kq 20-40
At Risk	avg_burnout > 50

5. Culture Index

Organization-wide health composite score:

Culture_Index = (Rec_Frequency/50 × 25) + (Equity/100 × 25) + (Cross_Team/100 × 25) + (Reciprocity/100 × 25)

Normalized to 0-100 scale

6. Original KQ Score (Preserved from V1)

The base KQ score uses eigenvector-weighted PDE time-decay:

6.1 Feature Extraction

T (Tasks) — Total recognition messages sent
D (Diversity) — Unique tags used
U (Unique) — Unique teammates recognized

6.2 Preprocessing Pipeline

Log Transform: T_log = log(1 + T)
Robust Z-Score: Using median and MAD
Winsorization: Clip at 2nd and 98th percentiles
Correlation Matrix: 3×3 matrix R
Shrinkage: K = (1-α)R + αI where α=0.1
Eigenvector: Principal component via power iteration

6.3 PDE Time-Decay

q_new(u) = ρ × q_old(u) + (1 - ρ) × s_today(u)
where ρ = exp(-γ × λ_max), γ = 0.02 (14-day half-life)

7. Privacy Architecture

Data Protection

Kindworth processes conversation content but retains only:

User IDs (anonymizable)
Computed metrics and scores
Aggregate statistics

Raw message content is never stored after processing.

8. Output Schema

{
  "version": "2.0",
  "summary": { "total_users": 350, ... },
  "users": [{
    "user": "eng001",
    "KQ": 78.5,
    "burnout_risk": 45.2,
    "flight_risk": 12.0,
    "leadership_score": 67.3,
    "extra_work_hours": 8.5,
    "user_type": "Giver",
    ...
  }],
  "team_health": { "Engineering": { "cohesion_score": 64.8, ... }},
  "watch_lists": { "burnout_risk": [...], "flight_risk": [...] },
  "culture_index": { "recognition_frequency": 47.2, ... }
}