[Edaily Reporter Han Kwangbeom ] AI, which used to search corporate internal data to provide answers, is now evolving into a “digital employee” that summarizes meeting follow-ups and analyzes financial performance. Chinese AI company DiPEAK unveiled technology and implementation cases that connect complex corporate data and business systems, enabling AI to perform actual tasks.

Zhang Shilong, Chief Operating Officer (COO) of DiPEAK, explained the future direction of enterprise AI at the “Edaily Global AI Forum (GAIF 2026)” held on the 12th at the JW Marriott Dongdaemun in Seoul, stating, “It shouldn’t just stop at answering questions; the action must be completed.”

He emphasized that what companies want from AI is not merely the provision of information. Rather, AI must be able to assess business conditions in real time, make decisions based on sufficient evidence, execute actual tasks, and measure the results in terms of cost savings or return on investment (ROI).

◇Connecting Scattered Corporate Data… AI Handles Everything from Analysis to Execution

The problem D-Peak aims to solve is that data is scattered across multiple systems within each company. For example, if a CEO asks about revenue, the Customer Relationship Management (CRM) system shows the contract amount, the Enterprise Resource Planning (ERP) system shows the order amount, and the financial system shows accounting revenue. Although all these figures relate to revenue, the definitions and criteria differ, so the answers may vary.

COO Jang pointed out, “The problem is that while the model itself is smart, it doesn’t understand the actual business.” To solve this, Dipik developed “X-Engine,” which links the meaning of data, metric definitions, and access permissions to AI. The company explains that this allows it to integrate multiple systems without moving existing data to find the figures that match the query.

They combined this with the enterprise inference model “ShanHe Model,” “RUI Source” (which manages AI agents’ tasks and approval procedures), and the “Ask” series of business applications. The structure works such that after the AI understands a question, it uses the necessary tools to verify the data, validates the results, and processes the task after obtaining human approval where required. Access permissions are applied to all tasks, and execution logs are maintained.

D-Peak stated that the accuracy of enterprise data Q&A using the ShanHe Model exceeds 99%, while token usage has been reduced by more than 60%. Tokens are the units used by AI to process information, so reducing their usage offers potential for cost savings. However, the specific conditions used to measure accuracy and the benchmarks for token reduction were not disclosed in the presentation materials.

“AskRUI,” which COO Jang repeatedly highlighted, is a data analysis tool designed to make this technology easily accessible to corporate employees. When users ask questions in natural language, it finds and analyzes relevant data and even explains the underlying causes. According to the presentation materials, it enables line employees to handle 70% of analytical tasks directly. For example, when profit margins decline, users can track which products and regions experienced rising production costs.

◇China Mobile Reduces Meeting Work by 90%…Expanding to Steel and Finance

Actual client case studies were also presented. China Mobile reported that by applying AI to the tasks of recording meeting minutes and generating action items, it reduced post-meeting processing time by more than 90%. China Telecom explained that by utilizing AI to prepare bid documents, it cut preparation time by 60% and manual work by 50%.

China Baowu Steel Group, a global steel company, integrated AI with the financial data of more than 1,000 subsidiaries. According to Dipik, the accuracy of financial Q&A reached 99%, and the efficiency of financial inquiries and analysis improved by 85%. In Singapore’s public service sector, a case study was presented showing that by enabling citizens to submit complaints via voice and verify supporting documents, manual work was reduced by 40% and the application success rate increased by 60%.

Dipeak, founded in 2022, stated that it has secured more than 300 corporate clients to date. Its contract value increased from 43.5 million yuan in 2024 to 240 million yuan in 2025, and the company expects it to reach 600 million yuan this year. The client performance metrics and contract projections provided by the company require separate verification.

The “digital employee” touted by Dipik is an AI that understands corporate data and handles repetitive tasks. To this end, it is designed to handle the entire workflow—from setting work objectives to searching for data, verifying results, and executing tasks—while allowing human intervention for critical decision-making. COO Jang emphasized, “Enterprise AI shouldn’t stop at answering questions; it must be able to execute actual tasks and quantitatively measure the results.”