On the afternoon of January 16, Iflytek Co.,Ltd. unveiled its "Bidding and Procurement Agent Platform," announcing the integration of large language model decision-making and planning with Robotic Process Automation (RPA) execution capabilities. Utilizing "Xingchen RPA" as the agent's "automated hands," the platform enables stable cross-software operations. It also allows companies to upload their own data to fine-tune models, achieving optimization for a specific scenario model in an average of 40 minutes. Companies can build their own AI procurement expert with zero code in just 5 minutes.
Currently, bidding and procurement, as a core operational activity for enterprises, face multiple challenges including difficulties in improving efficiency, ensuring compliance, and optimizing costs. Zhang Yongliang, CTO of Iflytek's Enterprise Digitalization Business Group, introduced at the launch event that through a series of scenario-specific agents—such as "Intelligent Compilation and Review of Tender Documents," "Abnormal Behavior Detection," and "Assisted Bid Clearing and Evaluation"—the platform can reduce the average tender document preparation time from 5-7 working days to just 30 minutes. The accuracy rate for detecting bid-rigging and collusion reaches 96%, and overall bid evaluation efficiency is improved by over 70%.
As a key co-development partner, the National Energy Group's "Intelligent Unmanned Review System" has stably processed over 180,000 procurement projects with a review accuracy rate exceeding 97%. The system has achieved a breakthrough in review efficiency—transitioning from the traditional model of "reviewing 1-2 projects per day" to "outputting intelligent review results in 10 minutes." More notably, through intelligent, full-process control, the system has effectively helped increase the procurement cost-saving rate, providing a quantifiable technological pathway for large enterprises to optimize procurement costs.
Differing from previous project-based deliveries, the "Bidding and Procurement Agent Platform" launched by Iflytek represents a significant shift in business model. Based on the company's self-developed "Xingchen Agent" technological foundation, enterprises can flexibly assemble AI capability components using low-code or even zero-code methods to quickly build custom agents that closely fit their specific business processes.
"We are not providing a single solution, but rather the infrastructure supporting the intelligent transformation of the industry," stated Yu Jidong, Senior Vice President of Iflytek and President of the Consumer Business Group. He emphasized that this platform-based model significantly lowers the application threshold and trial-and-error costs for enterprises, creating conditions for the large-scale adoption of intelligent bidding and procurement technologies.
In discussions, Zhang Yongliang pointed out that during the abnormal bidding behavior detection phase, the platform faces challenges such as large tender document volumes, complex mixed text and image layouts, and increasingly concealed abnormal behaviors. To address this, the platform employs a multi-model collaborative mechanism—including text semantic similarity analysis, image similarity detection, and quotation pattern analysis—to conduct deep identification of risks like identical quotations, abnormal qualifications, and hidden affiliations. Currently, the comprehensive recall rate for text and image elements is 96%, and the comprehensive detection rate for abnormal behaviors leads the industry. The platform can also perform cross-verification using multi-dimensional clues such as IP addresses, hardware fingerprints, and equity relationships, providing traceable and auditable compliance assurance for the procurement process.
In the intelligent assisted bid clearing and evaluation phase, the platform uses an intelligent evaluation agent to provide assistance with the highly repetitive and rule-intensive tasks inherent in the review process. Data shows that for scenarios involving single-project tender documents exceeding 1,000 pages and more than 100 bidding units, the expert scoring deviation rate in traditional manual reviews is approximately 32%, with about 25% of projects requiring re-inspection. Through a human-machine collaboration mechanism, the platform achieves 97% accuracy in evaluating objective items and a 90% consistency rate between human and machine scores for subjective items. Overall review efficiency is improved by over 70%, significantly alleviating review pressure while ensuring fairness.
Comments