Skip to main content

Bairong's Silicon Employees Go to Work: Enterprise Agents Paid by Results

As large models grow more powerful, what exactly are enterprises willing to pay for? Bairong's latest interim report offers a clear answer: model capability alone is no longer enough to justify enterprise adoption. The focus is shifting from the models themselves to how AI can produce measurable, accountable output. From "selling models" to "getting paid by results," Bairong's silicon employees are going to work at scale.

AICC Revenue Grows Against the Trend: New-Scenario Income Nearly Triples​

In the first half of 2026, Bairong's total revenue fluctuated due to external factors. But its AICC (AI Contact Center) business grew 52% year over year, with revenue from new scenarios in more industries up 195%. AICC's share of Bairong's revenue rose from about 6.8% in H1 2025 to about 18.1% in the same period this year.

AICC simply means plugging AI into enterprise contact centers, letting agents directly handle queries, marketing, service, and operations. Bairong says it has matured its AICC and enterprise agent capabilities in financial scenarios and is now replicating them across logistics, securities, aviation, telecom, and insurance.

From Finance to Logistics and Securities: Real Results from Silicon Customer Service​

Bairong calls the enterprise agents that enter real service workflows "silicon customer service." In July, its agents began handling complex inbound tasks in logistics — shipment inquiries, delivery exceptions, and complaint feedback. Daily call volume grew from under 1,000 when launched to over 15,000 by the end of August.

The securities scenario is also progressing: as of July, 13 institutions were signed or in contract, and one top brokerage expanded its deployment from about 30 seats at the start of the year to about 210. These results show that AICC capabilities built mainly around finance are now entering new production environments.

Why Did Enterprise Agents Land on AICC First?​

Enterprise agents are entering more and more repetitive workflows, but scaling from technical feasibility to production requires solving one thing first: how to clearly measure business value. Contact centers handle large volumes of queries, marketing, and service work with well-defined boundaries, processes, and evaluation criteria. Once AI is plugged in, call volume, handoff rates, and unit service costs quickly show up in operating results.

CEO Zhang Shaofeng revealed that Bairong's customer-service and marketing silicon employee projects were first set up in 2017, and have been iterated for seven or eight years. The finance industry's high demands on accuracy, stability, and compliance shaped capabilities that overlap closely with what AICC needs: voice interaction, intelligent contact, and complex workflow execution.

RaaS: A Business Model Where Agents Get Paid by Results​

Bairong's RaaS (Results as a Service) model is a concrete form of this shift. According to the company, RaaS can charge per successful call or ticket completed without handoff to a human, per deployed agent seat, or in some scenarios, as a share of the business volume generated.

In the logistics business, for example, charging per effective call that solves a problem has already cut per-call processing costs by more than 50% versus traditional human labor. The service provider's revenue is now directly tied to work actually completed. This is the biggest difference from traditional software pricing: after software delivery, the buyer bears ongoing operating costs, whereas result-based pricing pushes the tech company deep into the business process — model performance, stability, cost, and task completion all enter the same operating logic.

Vertical Models + Data Flywheel: The Tech Foundation for Production-Grade Agents​

Bairong has built a fairly complete technical system around real production environments. On the model side, it has BR-Voice (speech multimodal), BR-LLM-Proactive (proactive dialogue), and BR-Vision-Doc (document understanding), with parameter sizes mainly between 0.4B and 30B. The goal is a balance between accuracy, stability, response latency, and inference cost that suits production.

In the INTERSPEECH 2026 MLC-SLM Challenge for multilingual conversational speech understanding, Bairong achieved 94.84% accuracy — third globally and first among Chinese industry players. Bairong also mines positive and negative feedback from real deployments, converts it into reinforcement learning signals, and continuously adjusts models, service strategies, and dialogue approaches, forming a data flywheel of "business runs, feedback, model iteration, back to business."

Bairong has also built a communications engineering system around telephone networks, adapting to 8kHz narrowband recognition, noise reduction, user interruption, and turn-taking. It deploys FDE (Field Deployment Engineer) teams to business sites to sort out service standards, knowledge bases, and system interfaces, reorganizing the parts suitable for AI into agent workflows.

AI Roll-up: From Technology Product to Factor of Production​

Bairong describes this path as "technology-driven AI Roll-up" — first reshape workflows with AI, then expand industry coverage. Once a benchmark scenario works, the same capabilities and memory can replicate across more roles and industries. As deployment scales, the large volume of interaction feedback from real operations feeds back into model training and strategy optimization, sustaining the flywheel.

Looking at the industry trend, AI is being pushed from the periphery of enterprise software toward the core of production operations. Software digitized business processes, large models added understanding and reasoning, and now agents are entering the processes themselves to take on concrete tasks. Bairong's case shows that machine intelligence is moving from a technology product to a factor of production in economic activity — and "getting paid by results" is the direct reflection of this shift in business models.

Sources: QbitAI "Enterprise-grade Agent deployment benchmark! Bairong silicon employees go to work, paid by results", Bairong interim report, 36Kr "Defining the era of flexible silicon employment"