Bairong Launches the "Bairong Voice" Brand Across the Board: Unified Voice Foundation "VoxOne" Goes Live, with the AICC Business Ready to Bloom

"One Voice Moves a Thousand Industries · Sound Gives Rise to All Things": On October 27, 2026, in Shanghai — Bairong (06608.HK) held the "Bairong Voice" brand and product launch event, unveiling a Voice AI product family unified under the "Bairong Voice" brand and powered by "VoxOne" as the unified voice intelligence foundation. The lineup spans three product lines — enterprise-grade AI customer service (Baifu), intelligent marketing (Baiying), and an open voice interaction platform (Baiyu) — and, for the first time, fully discloses the business results, technology foundation, and business model of this business under AICC (AI Contact Center).

The message from the launch was highly focused: Bairong has moved Voice AI from "being able to converse" to "being accountable for outcomes."

What was launched were products; what was validated was the business. What truly determines the weight of this launch is the third question: once models are available to everyone, what still constitutes a moat? Competition in Voice AI has shifted from "can it be done" to "who can actually deliver the results."

01 Why Voice: AI's Hardest Problem — and the One Closest to the Business Frontline

The first set of figures presented at the launch comes from human communication itself: in waking-state communication, 45% is spent listening and 30% on spoken expression — 75% in total is voice-based; 79% of enterprise employees prefer voice to align quickly when facing issues that demand immediate action; and 76% of customers prefer voice communication when dealing with complex, high-stakes matters. Research from the University of Texas and the University of Chicago also shows that, compared with text-only exchanges, voice creates a stronger sense of social connection without adding awkwardness.

For AI, however, voice is precisely one of the hardest classes of problems. Bairong has built along the voice line for many years, and the difficulty lies in three places. Bairong describes the bar for enterprise-grade voice AI as three gates: understanding the need — from hearing the words to grasping intent; natural conversation — from question-and-answer to keeping up with follow-ups, interruptions, and corrections; and getting the business done — from conversational responses to task execution and delivery of results.

First, a single sentence carries two layers of information at once — understanding speech means capturing words, pauses, and emotion simultaneously; by the time a sentence is spoken, it already carries semantic and non-verbal cues: is it a rhetorical question, or a polite refusal?

Second, voice itself is a channel with limited information dimensions. Traditional telephony uses a narrowband standard of 8,000 samples per second, offering less acoustic information than broadband internet audio. Bairong has pre-trained on real-world voice environments and continues to refine narrowband speech recognition, noise suppression, user interruption handling, and turn-taking detection.

Third, voice is dynamic; text is not. Text is one question, one answer; voice is continuous — users fall silent, change their minds mid-sentence, express dissatisfaction through tone instead of facts, and ask for a human agent as emotions rise.

"Hearing clearly is not understanding; understanding is not answering correctly; answering correctly is not answering well; and answering well is not solving the problem."

These three points lead to the same conclusion: what separates the front and back layers is not model parameters, but an understanding of business processes, the ability to integrate with systems, and the mechanisms of accountability for results. In consumer scenarios, the cost of a failed interaction is a retry; in enterprise contact scenarios, it can be a complaint — this is exactly why Bairong built Voice AI as a foundation rather than a feature.

02 One Foundation, Three Faces

Bairong Voice VoxOne is Bairong's unified voice intelligence foundation, positioned as "production-ready Voice AI for the enterprise," delivering silicon-based employees that are accountable for results across different scenarios. Vox is Latin for "voice," and One stands for the singularity of the foundation.

Baifu (enterprise-grade AI customer service) comes in three forms: Baifu Prime is the flagship brand of Baifu AI customer service; Baifu Lite is delivered as an all-in-one appliance for small and medium enterprises that value data privacy and want fast, lightweight deployment — fully privatized local deployment, data never leaves the intranet, no system re-engineering required, go-live in as fast as one day, with industry-specific scripts and business processes pre-built and no tuning needed; Baifu Aura is built for field staff, consolidating all voice and social media communication into traceable records, augmented with AI customer insight and a closed follow-up loop.

Baiying: a marketing Agent for customer insight and precision outreach. It sits on the proactive side of "AI calls" — outbound marketing — turning outbound calling from "voice notifications" into an operating line of "screening, following up, and converting." The path is a five-step closed loop: high-potential identification → proactive outreach → breakpoint follow-up → human handover → conversion. Silicon-based employees precisely identify high-potential customers through data-based segmentation and handle the first four steps. Three scenarios have already gone live: WeCom (enterprise WeChat) contact-adding and account-opening breakpoint reactivation for brokerages, merchant self-service outbound calling on e-commerce platforms, and at-scale renewal outreach to existing customers for a leading property and casualty insurer.

Baiyu: an open voice interaction platform for diverse ecosystems, providing full-duplex real-time voice interaction — users can interrupt, add to, or revise their requests at any time — deployed in AI companion toys, smart wearables, and industry smart terminals, with support for voice APIs, chip and module adaptation, and device-cloud collaboration. Six atomic capabilities — noise-robust pickup, graceful interruption, human-voice discrimination, complex sound-field adaptation, and more — all point to one thing: on real-world call lines, every segment of sound is worth decomposing.

Products answer the first question — whether it can be done. But for enterprise decision-makers, the real question runs one level deeper.

03 Business Results Already Delivered: Data, Growth, and Cross-Industry Replication

Every figure of VoxOne Family business value disclosed at the launch can be probed. Interim 2026 results show AICC revenue of RMB 166 million, up 52% year over year, of which revenue from new scenarios such as logistics, brokerages, banking, telecom operators, and auto insurance reached RMB 11.26 million, up 195% year over year.

In other words, Bairong's growth structure is being redefined by AICC. But a set of good numbers does not by itself constitute a moat — numbers can be copied once.

Cross-industry replication is underway. The most closely watched case comes from logistics, where the pain point is specific: more than 60% of inquiries concentrate on standardized questions such as package tracking, expediting, and delivery timelines, consuming human agents with repetitive inquiries; during major promotions and holidays, inbound volumes surge and human answer rates drop noticeably.

Results on the ground: the AI customer service in this scenario now handles nearly 15,000 calls per day (fewer than 1,000 in the initial launch period), with the longest single call exceeding 20 minutes and resolving the issue autonomously; it has cumulatively served more than one million users who dialed in. Inbound calls are handled independently by AI, with complex matters transferred to human agents together with service records and continuously optimized through monitoring and quality inspection — and the solution has been deployed across all scenarios at a leading logistics company.

In the brokerage industry, replication is happening as well. As of August 2026, 13 brokerage clients had been signed or were in the pipeline; at one leading brokerage, the number of AI customer-service employees grew from about 30 at the start of the year to about 210, and the monthly fee per AI customer-service employee is lower than the fully loaded cost of a comparable human seat — clients expanding their usage after validating results is the hardest signal to fake in enterprise AI.

With Baifu Prime, silicon-based employees have entered logistics, airlines, brokerages, banking, and telecom operators.

Bairong also offered an industry-side estimate: over the medium term (3–5 years), the labor cost that could be released through AI transformation in high-labor-cost industries is substantial (methodology: estimated headcount × AI transformation rate × annual salary; the company's own estimate).

What can be replicated is the method; what is hard to replicate is whatever produced these results in the first place.

04 Paying for Results: Business Model and Technology Foundation

All products launched this time adopt RaaS (Results as a Service) pricing: fees based on successfully completed tickets or effective calls, on the number of AI employees actually deployed, or as revenue sharing on the business volume facilitated. Take a logistics customer as an example: the fee per successful AI-handled call is lower than the fully loaded cost of a comparable human seat, and customers also save on recruiting and training expenses; as processing volumes rise, revenue grows in step with successful tickets and effective calls.

Bairong is not selling AI tools to customers; it is taking over the specific tasks and business processes previously carried by human agents, freeing those agents to focus on work that requires deep communication and emotional care — this is the single most consequential adjustment to the business structure.

What makes this model possible is a matrix of proprietary models. The launch showcased BR-Voice, an end-to-end voice multimodal large model, and BR-LLM-Proactive, a proactive-guidance dialogue large model: end-to-end response latency within 500 milliseconds; a character error rate of 2.30% on the public AISHELL2 benchmark; an F1 score of 97.03% for industry hot-word retrieval; an F1 score of 96.35% for dialogue turn detection; and a TTS human-likeness score of 95.00%. Bairong pre-trains dedicated models for real-world voice environments, with parameter sizes compressed into the 0.4B–30B range, directly benefiting inference cost and deployment speed; its proprietary inference engine re-architects speech recognition, large-model understanding, and speech synthesis into a parallel streaming design, lifting overall response speed by more than 3x and cutting inference cost by 30%.

On external leaderboards, BR-Voice ranked third globally and first among China's industry players with an official accuracy of 94.84% in Task 2, "Multilingual Conversational Speech Understanding," of the MLC-SLM 2026 Challenge at INTERSPEECH 2026 (more than 110 participating teams, 14 languages, about 2,100 hours of real two-person long conversations); on the public VoiceBench benchmark, BR-Voice-Reasoner scored 90.46, ranking second on the current public leaderboard; and at the IEEE SLT 2026 SmartGlasses Challenge, BR-Voice placed in the global top five on all four tasks, including second place globally on both spoken language understanding tasks.

Of greater long-term value is data. Bairong has accumulated 2.6 billion real conversation records — data that cannot be synthesized and is not publicly available, a moat that vertical models cannot easily replicate in the short term.

Leaderboards can prove model capability; they cannot prove delivery capability. Between the two lies the most labor-intensive part of the story.

05 From Model to Delivery: The First 99 Kilometers and AI Roll-up

Bairong stresses repeatedly: a stronger model that can operate a computer does not mean it can be dropped straight into an enterprise. The real barrier lies in the "first 99 kilometers" — the stability, controllability, and security of processes and systems, integration with ERP/CRM and internal systems, and business process design; only when those 99 kilometers are paved does the execution capability of the last mile become meaningful.

To cover these 99 kilometers, the company has built a team of FDEs (Forward Deployed Engineers) who work on-site with customers to map business requirements, operating rules, and system environments, converting manual workflows into AI workflows — closing the loop from user interaction and task handling to system write-backs and delivery of results, and iterating continuously based on actual operating performance after go-live.

The first 99 kilometers are paved by people, but every pass adds real-world feedback to the model. Only then does the third question truly surface: where can this game go next?

In North America, AI Roll-up — using expensive labor markets as the entry point — has become a trend. The path breaks into three steps: capital consolidation, acquiring client relationships, service capacity, and real workflows; AI transformation, embedding AI into industry processes to improve service delivery; and operating results, judged by cost, quality, and customer value. U.S. employee compensation was about US$16 trillion in 2025 (FRED annual data), while total U.S. software spending in 2022 was about US$313 billion — software spending is only 2% of labor spending (source: a16z, "Vertical SaaS: Now with AI Inside"). What generates profit is not the software itself, but the results that follow from putting AI into real operating processes.

Bairong is now translating this path into industrial collaboration, with two disclosed practices so far: with Zhixing Heli, advancing human-machine collaborative services in the AI + BPO direction — standardized tasks handled by AI, complex matters handled by humans; and an industrial fund established in partnership with CICC Capital — the company announced on October 15 that, together with CICC Capital and others, it is setting up an industrial fund with an initial size of RMB 500 million, of which Shuke Youxiang subscribed RMB 195 million, or 39%, with CICC Capital serving as general partner and fund manager, investing in AI, computing power, advanced manufacturing, and smart hardware.

A company whose core business is AICC anchored its industrial fund on AICC as well — what it wants is not financial return, but upstream assets that can plug into its delivery chain.

The story does not end — it simply extends from the company level to the industry level.

06 Three Perspectives, Converging at One Launch

For enterprise customers, Bairong will be clear about how AI delivery acceptance criteria are defined, who triggers the handover to human agents, how billing corresponds to results, and how much existing systems need to be modified — from direct integration with existing contact centers to a turnkey, fully privatized local deployment where data never leaves the intranet, with clear boundaries given for each form.

For investors and industry researchers, Bairong will lay out the speed and cost of cross-industry capability replication and the latest progress of its delivery system — from on-site FDE transformation to at-scale deployment in logistics and brokerage — as well as the upgrade path that AI Roll-up offers the contact center industry.

For the media, Bairong will provide the statistical definitions behind each business metric and verifiable materials, including public evaluation leaderboards, company announcements, and independently verifiable third-party sources.

Bairong founder and CEO Zhang Shaofeng has previously described the path as "validate first, then replicate, then scale up"; this launch is its full presentation in Voice AI.

"AI Roll-up offers a new path for industrial collaboration and the intelligent upgrading of the contact center." Bairong said at the close of the launch that it will keep deepening its work in customer service scenarios and work with industry partners to advance human-machine collaborative upgrading.

Extending intelligence across boundaries, going far together with fellow travelers — making every connection more valuable.