AI News

Baiwang CEO Wang Qiangyu: How to Build an AI-Native Product System

LeiphoneUpdated Nov 16 21:00

AI summary

Wang Qiangyu takes over as CEO of Baiwang, driving the transformation of accumulated tax and finance scenario experience into specialized models and agent products, while enabling R&D, delivery, and organizational capabilities to jointly support commercialization. Baiwang has launched an AI tax and finance expert team, introducing natural language interaction and standardized task execution in invoicing, bookkeeping, and filing, while retaining rule validation and human confirmation. The article also uses Salesforce and NVIDIA's CRM reasoning model Koa as a reference to discuss how specialized models, data governance, and FDE delivery organizations can work together.

Body · AI translation

After companies integrate large models into their products, new business questions are only just beginning to unfold. When a customer proposes a task, what information the system can access, who approves the results, and how exceptions are handled all affect whether AI can be put into continuous use. The capabilities shown in product demos need to go through a complete service process before they can be converted into customer value.

Wang Qiangyu's appointment as Chief Executive Officer (CEO) of Baiwang Co., Ltd. provides a concrete angle for observing this change: how Wang Qiangyu will transform Baiwang's accumulated fiscal and tax scenario experience into professional models and agent products, and enable R&D, delivery, and organizational capabilities to jointly support commercialization.

The delivery unit of products is beginning to change

Take transaction reconciliation as an example. The system may need to read invoices, contracts, and payment records within an authorized scope, identify discrepancies, list the supporting evidence, and then hand the items requiring confirmation to the relevant positions. For customers, value depends on whether problems are accurately identified, whether subsequent handling is smooth, and whether the time invested is reduced.

This changes the starting point of product design. Teams need to clarify which types of tasks they serve, under what conditions execution can proceed, and which situations should be stopped or handed over to humans. Models, data, tools, and review mechanisms need to be combined around the same task.

Baiwang has launched an AI fiscal and tax expert team, introducing natural language interaction and standardized task execution into services such as invoicing, bookkeeping, and tax filing, while retaining rule validation and human confirmation. As this direction goes deeper into more complex fiscal and tax operations, the ability to reliably complete tasks will become even more important.

This is also why Wang Qiangyu's appointment as CEO deserves attention. As the scope of product services expands, the trade-offs among R&D investment, customer commitments, and delivery assurance often need to be coordinated at the company's operating level.

Bringing data and professional judgment into the same task chain

For the same transaction, different systems may use different customer names, contract numbers, and record standards. Data needs to be checked and linked, rules need to specify applicable conditions, and permissions need to follow task execution. Otherwise, even if a model can explain fiscal and tax concepts, it may still be unable to judge whether the materials at hand are sufficient.

Vertical models in professional domains undertake professional understanding and reasoning tasks here, knowledge systems provide continuously updated materials, and agents connect tools and workflows. For steps requiring certainty, such as tax amount calculation and rule validation, corresponding programs and professional mechanisms are also needed. How each layer cooperates should be determined by task requirements and evaluation results.

Koa, the CRM reasoning model launched by Salesforce and NVIDIA, provides a reference for an enterprise software company: post-training on top of open models, using enterprise workflow knowledge for professional tasks. According to its announcement, Koa was trained using synthetic scenarios and did not use customer data. Professional capabilities can be formed in a variety of ways; the key is whether the business is understood and whether the results can be validated.

For Baiwang, model building and data governance need to advance in parallel. Authorized business information, confirmed professional knowledge, and quality records from tasks should be managed separately and invoked as needed. Before business feedback enters model or product improvements, professional review and purpose confirmation are also needed.

Can the customer site improve the next delivery?

The FDE, or forward-deployed engineer, that Wang Qiangyu previously discussed is precisely an organizational approach that connects model capabilities with the customer site. Engineers and industry experts work together to clarify problems, determine the required data, rules, and tools, and let solutions be tested in real environments.

The business value of this work also depends on whether on-site experience can be reused. For example, if a team improves a method for matching invoices and contracts in one project, or identifies a type of exception requiring manual review, it needs to determine whether it can become a general component, rule template, or evaluation case.

Only when subsequent customers therefore reduce repeated development and trial and error does it mean that the company's product capabilities accumulate as projects grow. Accordingly, delivery cycles, reuse of standard components, and per-project service investment can all serve as bases for observing this accumulation.

On the deployment side, private deployment or integrated computing power appliances can address some customers' needs for data control, system access, and operating environments. Specific plans still need to be evaluated in light of task volume, hardware, and operations and maintenance investment. For Baiwang, deployment choices should also be considered together with delivery and cost management.

Only when organizations collaborate around tasks can there be a common evaluation standard

These teams need to form a common definition of the same task. What counts as completion, what counts as failure, how insufficient materials should be reported, and who is responsible for professional review should all be made clear as the project progresses. Otherwise, even if each department completes its own work, customers may still fail to receive complete service.

The evaluation approach must also be adjusted accordingly. In addition to technical evaluation, it is also necessary to observe task success rates, manual review and rework, and customers' continued use. Whether to expand investment in a product can be judged together with these results and service costs.

Coordination at the CEO level helps provide a consistent basis for cross-departmental trade-offs: which scenarios to prioritize, which requirements can be standardized, which projects need controlled commitments, and which experience should return to R&D. Wang Qiangyu's move from head of AI business to CEO lies in his potential role in driving these business decisions to connect with the technology roadmap.

These are the organizational tasks Baiwang needs to face in advancing its transformation; the specific results still need to be verified by subsequent public progress and customer outcomes.

Explaining the value of a technology company through business results

The State Council executive meeting held on October 16 emphasized using implemented applications to drive technological innovation. For enterprise software companies, responding to this direction requires enabling AI to continue playing a role in real business.

Its value path can be translated into several specific changes: professional AI expands the services customers can purchase, standardized capabilities improve delivery efficiency, and continued use brings renewals or demand for more tasks. Revenue quality, customer retention, and service costs can help the market judge whether this path holds.

Therefore, the market generally believes that the significance of Wang Qiangyu's appointment as CEO is more importantly reflected in whether the company can organize models, products, and customer service into a continuously operating business process. Each time a valuable task is completed, it not only benefits the customer but also leaves confirmed experience for the next service; only then will Baiwang's transformation into an AI-native enterprise gain a sustainable foundation.

Source

Leiphone