RevelirQA performs auto QA to the standard of a human quality analyst, calibrated with each customer until human QA reviewers validate that over 95% of its evaluations are correct on a sampled review. Soft skills like empathy, tone, and active listening get missed by AutoQA systems not because AI can't detect them, but because most scorecards are built around binary policy checks (did the agent verify identity, did the agent offer the refund script) that don't have an equivalent for "did this response sound human." The fix isn't to abandon policy checks or bolt soft skills on as an afterthought. It's to write soft skills into the QA scorecard as the same kind of observable, evidence-based criteria used for hard policy items, and to use coaching sessions, not the score itself, as the place where nuance gets discussed. Because RevelirQA evaluates 100% of conversations against a company's own QA scorecard, soft-skill criteria get applied with the same consistency as policy checks, on every ticket, not a hand-picked sample.

TL;DR

• RevelirQA performs auto QA to the standard of a human quality analyst, calibrated per customer until human QA reviewers validate that over 95% of its evaluations are correct on a sampled review. This human-level judgment is the differentiator; 100% coverage is table stakes most AutoQA tools already claim.
• Hard policy checks (compliance steps, disclosures, scripts) are easy to automate because they're binary. Soft skills like empathy and tone get skipped because most scorecards never translate them into observable, checkable criteria.
• Leading QA frameworks already treat soft skills as Non-Critical Errors, separate from policy violations, so a scorecard needs both categories scored independently, not folded into one score.
• Coaching on soft skills works best when it's built from specific flagged moments in a transcript, not a generic "be more empathetic" note.
• RevelirQA evaluates every conversation against a company's own SOPs, so both policy and soft-skill coaching programs are built on full visibility into an agent's actual behavior, not the 1-5% sample manual QA has historically relied on.

About the Author: This article draws on Revelir AI's work building and calibrating RevelirQA, an AI quality assurance platform running in production at high-volume customer service teams including Xendit and Tiket.com, scoring thousands of conversations a week across English, Indonesian-language, Thai, and Tagalog support operations.

Why Do AutoQA Scorecards Favor Hard Policy Checks Over Soft Skills?

AutoQA scorecards favor policy checks because policy items are structurally easier to define: either the agent verified the account or they didn't, either the refund disclosure was read or it wasn't. Soft skills resist that structure. "Empathy" and "tone" are abstractions, not actions, so when a QA team writes them into a scorecard without translating them into something observable, both human reviewers and AI scoring engines default to scoring on gut feel, which produces inconsistent results across reviewers [kaizo.com]. This is exactly why leading frameworks like the COPC Customer Experience Standard separate soft skills and professionalism into Non-Critical Errors, distinct from policy violations that make a transaction critically defective. The categories are tracked separately because they measure different things: one measures whether the agent followed the rules, the other measures whether the customer had a good experience while the rules were being followed.

What's the Risk of Scoring Only Hard Policy Checks?

The risk is a QA program that reports high scores while customer satisfaction quietly declines. Hard policy-based QA scoring often produces inflated internal scores that don't correlate with actual customer satisfaction, because these frameworks focus strictly on procedural adherence. An agent can follow every step of a script, verify identity, read every required disclosure, and close the ticket within SLA, while still leaving the customer frustrated because the agent sounded robotic, missed the customer's actual question, or never acknowledged that the customer had already contacted support twice about the same issue. A scorecard that only checks policy will mark that ticket as a pass. That's the gap soft-skill scoring exists to close, and it's also the gap that gets exploited when soft skills are left as a vague catch-all line item instead of a defined set of checks.

How Do You Turn Soft Skills Into Checkable QA Criteria?

You turn a soft skill into a checkable QA criterion by replacing the trait with an observable behavior. Instead of scoring "empathy" on a 1-to-5 scale, a quality scorecard should ask a yes/no question tied to something that actually happened in the transcript: did the agent name the customer's emotion within the first part of the exchange, did the agent acknowledge a repeat contact, did the agent avoid copy-pasted language when the customer expressed frustration [justcall.io]. One observable cue per row, scored yes or no, is more reliable than a 1-to-5 scale because it removes the guesswork that causes two reviewers to score the same conversation differently [justcall.io].

Soft skill	Vague scorecard item	Observable, checkable version
Empathy	"Showed empathy" (1-5)	Named the customer's emotion or frustration within the first reply (yes/no)
Active listening	"Listened well" (1-5)	Referenced a specific detail the customer gave, without asking the customer to repeat it (yes/no)
Tone	"Professional tone" (1-5)	Avoided scripted language after the customer expressed frustration twice (yes/no)
Conflict resolution	"Handled conflict" (1-5)	Offered a concrete next step or alternative when the first solution was rejected (yes/no)

These four categories, empathy, active listening, clear communication, and conflict resolution, are the standard soft skills categories recognized across customer service quality frameworks, and each one can be broken into two or three observable cues like the ones above rather than scored as a single abstract judgment.

How Should AutoQA Platforms Score Soft Skills Alongside Policy Checks?

Building on the scorecard structure above, the harder question is how an AutoQA engine actually applies it consistently across every conversation an agent handles, not just the ones a manager happens to review. Zendesk's 2026 Customer Service Quality Benchmark Report found that only 2 percent of customer service conversations are typically reviewed manually, which means soft-skill coaching built on manual sampling is coaching built on a 2 percent slice of an agent's real behavior. RevelirQA evaluates every conversation against a company's own QA scorecard, with soft-skill criteria applied with the same consistency as policy checks, on every ticket, not a hand-picked sample. Because the scorecard items are configurable as binary, multi-option, or scored criteria, a team can run both policy checks and soft-skill checks on the same ticket, scored independently, the way COPC-style frameworks recommend.

The part that matters more than coverage, though, is whether the AI's judgment on a soft-skill call actually matches what a trained human reviewer would conclude. Recognizing a policy miss is a lookup problem; recognizing whether an agent's tone de-escalated a frustrated customer requires contextual judgment closer to what a human QA analyst does. That's the calibration Revelir runs with each customer: the AI is tuned against that company's own scorecard and transcripts until human QA analysts, reviewing a sample of RevelirQA's evaluations, confirm that over 95% are judged correct. That threshold, independently validated per customer rather than claimed out of the box, is what the industry-standard for AI-powered QA systems is built around, and it's the bar that separates an auto QA tool that flags keywords from one that scores soft skills the way a quality analyst would.

How Do You Coach Agents on Soft Skills Once They're Flagged?

A related but distinct question from scoring is what coaching actually looks like once a soft-skill gap is flagged. Coaching works when it's anchored to a specific, observable moment in a real transcript, not a general trait. Call center coaching built around specific evidence, paired transcripts, and clear performance data is consistently cited as more effective than generic feedback sessions [zoom.com][salesken.ai]. A manager telling an agent "be more empathetic" gives the agent nothing to act on. A coaching note that says "in this ticket, the customer repeated their issue twice and you didn't acknowledge it until the third message, here's what an earlier acknowledgment could have looked like" gives the agent a concrete behavior to change next time [intryc.com]. This is also why letting agents see and challenge their own AI-generated scores matters: when agents understand why a score landed where it did, they re-engage with coaching instead of dismissing it as arbitrary [voxjar.com].

RevelirQA supports this by generating a reasoning trace behind every score, which documents the model considered, the policy or scorecard language it retrieved, and the specific reasoning that produced the score. For a soft-skill flag, that trace becomes the coaching material itself: the manager and agent can look at exactly which part of the transcript triggered the flag and why, rather than relitigating whether the score was fair.

Frequently Asked Questions

Can AI actually detect soft skills like empathy and tone in a transcript?
Yes, when the soft skill is translated into an observable criterion rather than left as an abstract trait. An AI scoring engine can reliably check whether an agent named a customer's emotion, referenced a specific detail the customer gave, or avoided scripted language after repeated frustration, because those are detectable patterns in the text, not subjective impressions.

Should soft skills and policy checks be combined into one overall QA score?
No. Leading frameworks like COPC score them separately, treating soft-skill and professionalism issues as Non-Critical Errors distinct from policy violations, so teams can see where an agent is failing policy versus where they're failing the customer experience, even when policy was followed correctly.

How many conversations does manual QA typically review for soft skills?
Industry benchmarking puts manual QA review at around 2 percent of total conversations, which severely limits how much soft-skill coaching data a manager has to work with.

What accuracy level should an AI QA system hit before a team trusts it on soft-skill scoring?
The industry-standard threshold is above 95 percent accuracy, validated by having the customer's own human QA analysts review a sample of the AI's evaluations and confirm agreement with the scores, rather than a vendor's generic benchmark.

Does scoring 100% of conversations replace the need for human QA analysts?
No. Full coverage changes what human analysts spend their time on, shifting them from manually sampling a small fraction of tickets to reviewing flagged patterns, calibrating the AI, and running coaching conversations with agents.

Can the same auto QA system score both human agents and AI chatbot interactions on soft skills?
Yes. As companies run AI chatbots alongside human reps, a single scoring QA scorecard applied to both gives CX leaders one consistent view of soft-skill and policy performance across the entire support operation, rather than separate standards for bots and humans.

About Revelir AI

Revelir AI builds RevelirQA, an AI AutoQA engine that scores conversations against a company's own policies and QA scorecard, calibrated with each customer until human QA analysts validate over 95% accuracy on a sample of evaluations. Founded in Singapore in 2025, Revelir runs in production at companies including Xendit and Tiket.com, scoring thousands of conversations a week across English, Indonesian-language, Thai, and Tagalog support operations. Every RevelirQA score comes with a full reasoning trace, the model, the retrieved policy documents, and the reasoning behind the score, giving QA and CX teams an auditable basis for coaching on both hard policy checks and soft skills. Because RevelirQA ingests a company's own SOPs via RAG rather than scoring against generic benchmarks, soft-skill criteria get evaluated against the standards that company actually trains its agents on.

If your team is trying to build soft skills into an auto QA program without losing consistency on policy checks, get in touch with Revelir AI to see how RevelirQA scores both on every conversation.