SK Telecom data work increases AI agent precision to 80%

SK Telecom, a telecom company in South Korea, says that its AI agents give correct results for more than 80% of questions about network data. A tool with the name Data Desk writes the meaning of each data field, and the precision increased from about 40% to more than 80%. The company did tests on more than 9,000 network databases with about 300,000 fields.

How it works

Persons know the meaning of a data field from their work. AI agents do not, the company says. Data Desk writes the meaning and a description for each field.

A second tool, Z.E.Z, keeps private data in an isolated zone. It gives the AI only the data that the work needs, for example metadata and calculated values.

The numbers

In the tests on network databases, precision increased from about 40% to more than 80%. Kim In-soo of the AI Board team gave the numbers at a briefing.

In-house agents became around 7,000. The company cut them to 1,500.

What is not known

The company gave no external test, and the numbers are for network databases only. It has not shown cost savings or new revenue at this time. Kim says that the company can make more agents or cut some, as the results show.

Insight: Data meaning matters as much as the model

DataHub is a company with data software. It says that agents with no company context are right about 50% of the time. SK Telecom shows about 40%, then more than 80% with its tool. The numbers come from the companies that use the tools. This reads as a move to control the layer that gives meaning to company data. That layer is possibly more important than the model.