AI-Assisted Feedback Generation

Calibration-Grade Review Engine Transform raw manager observations into calibration-ready performance reviews that align with your company's competency framework, eliminate cognitive bias, and produce actionable development plans. Observation Collection Manager inputs 5-10 raw sentences per competency area. AI accepts unstructured notes, bullet points, or even voice-to-text transcripts. Competency Mapping AI maps each observation to the company's defined competency framework (e.g., 'Communication', 'Technical Depth', 'Leadership') with a confidence

score. Bias Detection Sweep AI scans for 15+ cognitive biases: recency bias, halo effect, gendered language ('abrasive' vs. 'assertive'), and attribution errors. Calibration Normalization AI compares the review language against department-wide review data to flag if a manager is an unusually 'hard' or 'easy' grader. Development Plan Generation For each 'Meets Expectations' or lower rating, AI generates 2-3 specific, measurable development actions with suggested timelines. Use the SBI framework (Situation-Behavior-Impact) in your observations for maximum

AI accuracy. Run calibration analysis across all managers before the review cycle to normalize scoring. Include self-assessment data as a second input — AI can identify perception gaps between manager and employee. Don't let AI write the entire review — the manager's authentic voice must be preserved; AI structures and refines. Don't skip the bias check — studies show 74% of reviews contain at least one cognitive bias without intervention. Don't use absolute language ('always', 'never') — AI should flag these as accuracy risks. # Performance Review Bias

Detection Prompt Role: Industrial-Organizational Psychologist Task: Analyze the following manager observations for: 1. GENDERED LANGUAGE: Flag words that research shows are applied differently by gender (e.g., "aggressive" → suggest "assertive") 2. RECENCY BIAS: Identify if >60% of examples are from the last 30 days 3. HALO/HORN EFFECT: Check if one strong/weak area bleeds into unrelated competencies 4. ATTRIBUTION ERROR: Flag statements that attribute outcomes to personality rather than circumstances Input: {{raw_manager_observations}} Output: JSON with

{ flagged_phrases[], suggested_rewrites[], bias_type, severity }