ESI 20th Anniversary: Mao Daqun Discusses Data Governance in the AI Era and the Ruizhi OPT Agent Architecture

Recently, the 20th anniversary and customer conference of Esensoft was held in Beijing. At the conference, Mao Daqun, general manager of Esensoft, delivered a keynote speech titled "Born of Data, Co-creating the Intelligent Era," in which he reviewed the company's 20-year development journey and shared three core judgments on the development trends and competitive logic of the data industry in the AI era.

In his speech, Mao Daqun said that over the past two decades, enterprise growth was mainly driven by scale and investment, while in the next decade growth will increasingly rely on data and intelligence. Data has become the fifth factor of production, and intelligent assets are becoming the core assets newly accumulated by enterprises. Against this backdrop, how to unlock the value of data and make intelligence truly become productivity is the core question the entire industry needs to answer.

20 years focused on the data track, serving more than 13,000 customers

Mao Daqun divided Esensoft's 20 years into two stages: in the first decade, the company focused on two core products, i@Report and BI, covering data collection and data analysis; in the second decade, it extended from data governance and data management all the way to the AI field.

"In these 20 years, we haven't done anything else—we've just revolved around data and never left it," Mao Daqun said, noting that Esensoft is the only company in China that has never left the main data track and still takes data as its main business to this day.

According to the introduction, Esensoft has achieved industry firsts in multiple fields: in 2006 it took the lead in proposing the BI concept, in 2017 it released the conversational BI "Zhiwen 1.0," in 2018 it was the first to advocate All in One BI, and in 2024 it published the book "Written on the Eve of Data Asset Capitalization," the earliest book on data asset capitalization. In 2014 it released a full-lifecycle data product family, and in 2016 it completed the large-scale replacement of the international brand Cognos across the national tax system.

At present, Esensoft has served more than 13,000 customers and has more than 1,800 ecosystem partners, covering 390 cities and 32 provinces across the country. It holds an 85% share of the financial leasing market and serves 90% of the provincial and municipal health commissions and traditional Chinese medicine hospitals nationwide. IDC has ranked it first in China's data governance solutions market for five consecutive years (2021–2025); Gartner's lists of representative vendors for data assets, data governance, data fabric, and data analytics platforms all include Esensoft; in the DAMA 2025 Data Governance Best Practice Awards, three projects served by Esensoft—China Electronics, Anhui Electric Power, and Sinopharm Digital Technology—won awards.

In 2025, Esensoft was the first to propose the "AI and data double helix" view at the DAMA conference, and in 2026 this view became the theme of the DAMA conference. Mao Daqun explained that, on the one hand, high-quality data is the core foundation of AI applications (Data for AI); on the other hand, large model technology injects new momentum into data organization and management (AI for Data). The two influence each other and amplify each other level by level, forming a spiraling upward trend.

Judgment 2: What determines victory or defeat is not large models, but private domain data, localized algorithms, and human-machine collaborative learning capabilities.

Mao Daqun believes that for enterprises, large models are infrastructure, like water, electricity, and gas. What truly constitutes core competitiveness is three elements: the private domain data accumulated in an enterprise's own business, localized algorithms that integrate industry knowledge and business rules, and the ability to turn human-machine collaborative learning into organizational behavior.

"As long as you don't learn, or learn slowly, you don't need to learn—because the technology has iterated again." Mao Daqun pointed out that not paying attention to AI industry progress for three months will produce widespread anxiety, and the way to cope is to elevate human-machine collaborative learning capability from individual behavior to organizational behavior.

Mao Daqun proposed that AI can replace most execution work, but five capabilities are irreplaceable: framing problems (asking valuable questions), assigning meaning (establishing a deep connection between outcomes and value judgments), judgment (making choices at key nodes), accountability (bearing the consequences), and integration (integrating scattered outcomes into actions with a consistent direction).

For data practitioners, he further proposed five core qualities: taste (judging what data is worth investing in), topic selection (finding the right problems), verification (testing the reliability of AI output), connection (bridging the knowledge gap between business and IT), and hands-on action (making data land on business results).

Release of the Ruizhi OPT agent architecture, greatly improving data governance efficiency

In his speech, Mao Daqun officially introduced the Ruizhi OPT agent work architecture proposed by Esensoft: one person, plus a group of silicon-based experts, plus N sets of reusable skill packages (Skills).

Among them, the "group of silicon-based experts" covers the entire data governance chain, including roles such as requirements research experts, model design architects, data compliance experts, data development engineers, data quality experts, data standards experts, and data asset experts, which are called on demand and work collaboratively. The "N sets of Skills" cover plug-and-play skill packages such as data research questionnaires, data strategy blueprint preparation, data architecture design, ODS extraction, quality report generation, data standard writing, and data dashboard generation.

Mao Daqun said that in the past, data governance was the most time-consuming and costly part of data management work, usually requiring a team half a year to a year to complete. Under the OPT architecture, one person can complete the same workload, and governance teams move from "labor-intensive" to "brainpower-liberated." He also judged that large model governance and data governance will merge in the future, and there will no longer be a distinction between "AI governance" and "traditional data governance."

Mao Daqun revealed that Esensoft has already applied AI on a large scale in its own R&D and data engineering practices. Over the past two years, the R&D team's programming efficiency has increased by more than 50% each year in a row without a net increase in headcount. Overall data engineering efficiency has improved at a rate of 30% to 50% per year. At the project level, the implementation cycle for master data projects has been reduced by 50%, and manual work has dropped by 70%; the implementation cycle for data governance projects has been reduced by 50%, and the accuracy of AI-assisted data mapping and quality checks is better than manual work; AI programming has improved overall efficiency by about 3 times.

According to the introduction, Esensoft took about half a year to complete the construction of its AI product panorama. The full-lifecycle data products of the past have all been layered with AI capabilities this year, including EsXCode (an AI-native software R&D platform), EsXWork (an enterprise-level AI office platform), and fully agentized data products—Ruizhi OPT, Ruima OPT, ABI OPT, Zhiwen Agent, and others.

"Twenty years ago, we were born of data. Today, standing at the crossroads of AI and data integration, the most important thing is to make good use of the capabilities of large models," Mao Daqun said at the end of his speech. "Let there be no difficult-to-use data in the world—this is the shared vision of all data professionals."