QA as development, not surveillance
A support QA program that exists primarily to generate compliance scores produces agents who optimize for the scorecard, not for the customer. An agent who learns to format replies correctly to hit the “communication” criterion without improving the actual quality of their help has gamed the QA system.
QA designed for development starts with a different question: what behaviors, if improved, would produce better outcomes for customers? The scorecard follows from that question, not the other way around.
Designing the scorecard
A QA scorecard has two types of criteria:
Fatal flaws: Criteria where failure automatically fails the entire interaction regardless of other scores. Examples: shared incorrect information, violated data handling policy, terminated an interaction without proper escalation. These are non-negotiable.
Quality dimensions: Criteria scored on a scale, weighted by importance. Common dimensions and their relative weights:
| Dimension | What it measures | Weight |
|---|---|---|
| Resolution accuracy | Was the customer’s issue actually resolved? | 35% |
| Communication clarity | Was the response clear and understandable? | 25% |
| Tone and empathy | Was the tone appropriate to the situation? | 20% |
| Process adherence | Was the correct process followed? | 20% |
Accuracy gets the highest weight because an interaction that was friendly and well-formatted but gave the wrong answer failed the customer.
Sampling strategy
Random sampling is insufficient for actionable QA. Sample strategically:
- 2–3 tickets per agent per week minimum
- Oversample for: new agents in their first 90 days, agents with declining CSAT scores, and ticket categories with the highest volume or the most escalations
- Include one challenging ticket type per agent per review cycle (high-complexity, emotionally difficult, edge-case policy situation)
Closing the loop
QA results are only valuable if they change behavior. Every QA review connects directly to the agent’s coaching session within 7 days. The QA score is not shared separately from the coaching conversation — they happen together. This prevents QA from being a score the agent receives rather than a conversation that develops them. AI-first support platforms like AItocha CX automates QA sampling across 100% of tickets using AI scoring, which removes the sampling bias that plagues manual QA programs.