Decoding Your Score: The Science of AI Explainability (SHAP) in Coaching 📊
Decoding Your Score: The Science of AI Explainability (SHAP) in Coaching 📊
You finish a mock interview. The screen flashes: "Score: 72/100."
On most platforms, that's where it ends. You're left wondering: Why 72? Why not 85? Was it my voice? My answers? My shirt?
At JobInterview.live, we don't believe in black boxes. We use a SHAP-inspired approach — based on Shapley values from cooperative game theory — to tell you exactly why you got that score. Enter SHAP (SHapley Additive exPlanations).
What is SHAP? (In Plain English)
Imagine you and 3 friends buy a pizza for $20.
- Alice ate 4 slices.
- Bob ate 3 slices.
- Charlie ate 1 slice.
- You ate 0 slices.
How do you split the bill fairly? You can't just split it equally ($5 each). You need to attribute the cost based on contribution.
SHAP does this for AI. It looks at your final score (the bill) and calculates exactly how much each "feature" (the slices) contributed to it.
How It Works in Your Interview Report
When you see a JobInterview.live score, you see the breakdown:
Overall Score: 72%
- Base Score: 50% (Average candidate)
- +12%: Strong use of "STAR" structure in Answer #2.
- +8%: Excellent vocal confidence (volume/pace).
- -5%: Excessive use of filler words ("um", "like").
- -3%: Failed to mention "Python" in the technical question.
- +10%: Good eye contact ratio.
72% Total.
Why This Changes Everything for Candidates
1. Actionable Feedback vs. Vague Advice
- Old Way: "You need to sound more confident." (Okay, but how?)
- SHAP Way: "Your low volume (-6%) dragged your confidence score down. Speak 15% louder."
2. Prioritizing Your Prep
If SHAP tells you that "Missing Keywords" cost you 20 points, but "Body Language" only cost you 2 points, stop practicing your smile and start studying the job description. Fix the biggest leaks first.
3. Trusting the AI 🤝
AI bias is a real concern. SHAP creates transparency. If an AI gave you a low score, you can check why. If the reason is "Background noise," you know it wasn't biased against your accent—it was just your microphone.
The 3 Dimensions of Our Analysis
- Content (What you said): Keywords, structure, sentiment, relevance.
- Delivery (How you said it): Pace, volume, pitch variance, filler words.
- Visual (What you showed): Eye contact, posture, facial expressions.
SHAP calculates the interaction between these. Example: A great answer (Content +10) delivered in a whisper (Delivery -10) might result in a neutral score.
Conclusion: Data is Confidence
Uncertainty breeds anxiety. Knowing exactly where you stand breeds confidence. By demystifying the scoring process, we turn interview prep from a guessing game into a science.
Don't just practice harder. Practice smarter.
Start Practicing and Get Your Analysis →
Published: February 2026 | Reading Time: 4 minutes
