AI quality assurance for calls (auto QA, AI call scoring, or 智能质检 in Chinese) is a mature market in both the United States and China, but the two look very different.
In short: the US market is consolidating through acquisitions, and the selling point is moving from "100% automated scoring" to real-time agent assist and evidence-backed scores. In China, large cloud vendors sell speech recognition, the contact center and QA as one bundle, recording and QA rules in regulated industries such as insurance are a major driver, and since 2024 vendors have been replacing keyword rules with large models.
This article covers the representative vendors in each market, what sets each apart, and what Taiwanese companies can take from both when evaluating AI QA. It does not rank vendors or compare prices. Vendor details come from official pages and public reporting, linked in the text.
The US: consolidation, and a shift from post-call QA to real-time assist
US call QA vendors fall roughly into three groups: contact center suites that make QA one module of a larger system, AI-native startups whose core product is scoring with large language models, and revenue intelligence platforms focused on sales calls.
Representative US vendors
| Vendor | Type | What stands out in QA |
|---|---|---|
| CallMiner | Conversation analytics specialist | Scores 100% of interactions, with automatic redaction of personal and payment card data; acquired voice AI company VOCALLS in June 2025 |
| NICE (CXone Mpower) | Contact center suite | QA built into the full contact center platform, auto-scoring across voice, text and social; closed its acquisition of conversational AI company Cognigy in September 2025 |
| Verint (including Calabrio) | Contact center and workforce suite | Calabrio Auto QM became generally available in October 2024 with customizable generative AI scoring prompts; Verint was taken private in November 2025 and combined with Calabrio |
| Observe.AI | AI-native | Scores every interaction with contact-center language models and shows the evidence for QA teams to check |
| Level AI | AI-native | Each conversation score comes with evidence and reasoning, and reviewers can override the AI |
| Cresta | Real-time assist | Connects post-call analytics and QA with live guidance during the call |
| Balto | Real-time assist | Prompts agents during live calls, focused on regulated industries such as insurance, financial services, healthcare and collections |
| Invoca | Call tracking | Links QA to the caller's digital marketing journey, suited to inbound sales |
| Gong | Revenue intelligence | Analyzes B2B sales calls for coaching and forecasting rather than service compliance scorecards |
Sources: CallMiner acquires VOCALLS, NICE closes Cognigy acquisition, Calabrio Auto QM, Verint acquisition completed, Observe.AI Auto QA, Level AI, Cresta.
Four traits of the US market
1. Consolidation: suite vendors are buying AI capabilities. Besides NICE, CallMiner and Verint above, Zendesk acquired AI QA company Klaus in 2024 (TechTarget).
2. Suites vs standalone tools: companies already on NICE or Verint tend to switch on the built-in QA; AI-native vendors compete on scoring quality, evidence and speed to launch.
3. 100% scoring is the baseline; competition moves to evidence and real time: almost every vendor promises "no more sampling". The differences are whether each score points to evidence in the transcript, and whether the agent gets guidance during the call instead of a score afterwards.
4. Regulatory pressure: in February 2024 the FCC ruled that AI-generated voices count as "artificial" voices under the TCPA (Cooley summary); some states require all parties to consent to call recording; and calls involving card data carry PCI DSS redaction requirements. Together these make checking every call a risk-control need rather than a management preference.
China: cloud bundles, regulation-driven coverage, and large models rewriting the rules
Chinese QA vendors fall roughly into three groups: large cloud vendors that bundle speech recognition, cloud call centers and QA; speech and AI technology firms with deep roots in finance; and contact center SaaS and outbound calling vendors that include QA in their product lines.
Representative Chinese vendors
| Vendor | Type | What stands out in QA |
|---|---|---|
| Alibaba Cloud (Smart Conversation Analysis) | Cloud vendor | Transcribes recordings before checking them; rules combine keywords, semantic similarity, speaking rate and other conditions; recordings from its cloud call center flow in directly; a separate rule type is written for large models |
| Huawei Cloud (AICC) | Cloud vendor, contact center suite | Intelligent QA is one of the AICC capabilities enhanced by its Pangu large model |
| iFlytek | Speech technology | Built on its own speech recognition, serving bank customer service QA and QA of recorded financial sales sessions |
| Zhongguancun Kejin (Dezhu AI) | Finance-focused large models | Launched intelligent QA 2.0 in April 2025, covering text, voice, images and video, with QA rules described in one sentence |
| Ronglian Cloud (Rongxi) | Communications platform, contact center suite | Offers a QA agent among its large model applications, and mines calls for high-converting scripts |
| TI-Net | Cloud contact center SaaS | Positions full-coverage, fine-grained QA as a way to spot sales opportunities as well as risk |
| Zhuiyi Technology | NLP and large models | Full-coverage, multi-dimensional QA on its own domain model, emphasizing controllable compliance |
| BYAI | AI outbound calling | Its QA agent promises 100% conversation compliance monitoring in place of manual sampling, with custom compliance rules |
Sources: Alibaba Cloud intelligent QA, Alibaba Cloud large model QA rules, Huawei AICC, Dezhu intelligent QA 2.0, Ronglian Cloud, TI-Net, Zhuiyi, BYAI.
Four traits of the Chinese market
1. Cloud bundling: speech recognition, the cloud call center and QA come from one provider, so recordings never need to move. For companies already on that cloud, the barrier to adoption is low.
2. Regulation drives full coverage: the 2017 interim measures on traceable insurance sales require telephone sales to be recorded in full and archived, and require QA of closed policies during the cooling-off period, covering at least 30% and 100% for specified business types (gov.cn). In 2026 several local financial regulators issued more detailed traceability rules for life insurance sales (Xinhua). Note that these rules mainly target insurance sales; not every industry is required to check 100% of calls.
3. A fragmented market: IDC's July 2025 report says the top five vendors held about 35% of China's 2024 intelligent customer service market, and names intelligent QA as one of the scenarios where demand for large models is rising (IDC).
4. From keyword rules to large models and agents: early QA relied on keyword and regular-expression matching, which gets hard to maintain as rules pile up. Since 2024 vendors have launched large model versions that let you describe QA rules in natural language, or packaged QA as a "QA agent".
US vs China: same goal, different paths
| Item | United States | China |
|---|---|---|
| How it is sold | Built into suites, or standalone AI QA tools | Mostly bundled with cloud and contact center suites |
| Main drivers | Customer experience, coaching, TCPA and recording consent | Recording and QA rules in regulated industries such as insurance, service costs |
| Technical focus | Evidence-backed scores, real-time in-call assist | In-house speech recognition, large models rewriting QA rules |
| Market structure | Consolidating | Fragmented, top five at about 35% |
| Recent trend | Suites acquiring conversational AI companies | Large model QA and QA agents |
Three takeaways for Taiwanese companies
1. Full coverage is the starting point, not the finish line: both markets treat "score 100% of calls, replace manual sampling" as table stakes. The better questions are whether each judgment quotes the transcript and whether managers can review it quickly.
2. Rules should be written in your own words, easy to change and able to rescore: both markets are moving from keyword rules to natural-language rules. What matters is using your own scorecard and rescoring past calls when the rules change.
3. Verify language quality on your own recordings: Chinese vendors mainly serve the simplified Chinese market and US vendors mainly serve English. Taiwanese accents, vocabulary and code-switching between Mandarin and English need to be tested on your own recordings, and data location and retention should be cleared with compliance first.
The US and Chinese AI QA markets are heading in different directions: one is consolidating and moving toward real-time assist, the other is driven by cloud vendors and regulation and moving toward large-model rules. What they share is clear: full coverage is now the baseline, and what matters next is evidence behind judgments, flexible rules and performance in the local language.
Pathors Voxcal is AI call QA designed for Taiwanese service teams: it scores every recording against your own scorecard, quotes the line behind each judgment, versions your rules so past calls can be rescored, and works without changing your phone system.
Related reading:
Frequently Asked Questions
Which companies provide AI call QA in the US?
Representative vendors include conversation analytics specialist CallMiner, contact center suites NICE and Verint (now combined with Calabrio), AI-native Observe.AI and Level AI, real-time assist vendors Cresta and Balto, Gong for sales calls and Invoca for call tracking.
Which companies provide AI call QA in China?
Representative vendors include cloud providers Alibaba Cloud (Smart Conversation Analysis) and Huawei Cloud (AICC), speech technology firm iFlytek, finance-focused Zhongguancun Kejin (Dezhu AI), contact center vendors Ronglian Cloud and TI-Net, NLP company Zhuiyi, and AI outbound calling vendor BYAI.
What is the biggest difference between US and Chinese AI QA?
The US mixes contact center suites with AI-native tools, is consolidating through acquisitions, and focuses on evidence-backed scores and real-time assist. China is led by cloud vendors bundling speech recognition and call centers, regulation in industries such as insurance is a major driver, and vendors are pushing large model QA.
Does Chinese regulation require checking 100% of calls?
Not across the board. The 2017 interim measures on traceable insurance sales require insurance telephone sales to be recorded in full and require QA of closed policies during the cooling-off period, covering at least 30% and 100% for specified business types. These are insurance sales rules, not a requirement for every industry.
Can Taiwanese companies use US or Chinese AI QA tools?
They can evaluate them, but should check two things first: language quality, by testing transcription and judgment on their own recordings with Taiwanese accents, vocabulary and code-switching; and whether data location, retention and access controls meet company and compliance requirements.
