Alibaba Cloud Log Service SLS can now automatically extract business semantics from existing queries, dashboards, and alert rules, enabling agents to generate analysis statements that more closely match real business definitions. We believe this comes closer to the core challenge in enterprise data analysis than simply adding a natural-language query interface: the issue is not whether users can retrieve data, but whether the numbers returned actually answer the original question.
What this is
SLS Data Semantics can be understood as adding a reusable layer of business documentation to logs, SQL, and dashboards. The system identifies the business objects, states, and units associated with fields; restores filter conditions, metric formulas, and statistical scopes; and organizes query methods such as multi-table joins and grouping order into context agents can use.
Consider the “share of anonymous successful visits.” Beyond being a percentage, it must specify what counts as a success, where filtering occurs, and what the numerator and denominator represent. Two SQL queries may both execute successfully while answering different questions. SLS's value lies in capturing these definitions—often embedded in colleagues' experience—and checking formulas, constants, null values, and data ranges item by item after generating a query.
Industry view
Notably, this capability builds on assets teams already have rather than requiring business users to create a new knowledge base or set of prompts. It also uses read-only queries to resolve ambiguity, checking whether two IDs refer to the same object, whether joined records are duplicated, and whether conflicting sources require teams to return to the original material for correction. This reduces one of the most common agent errors: syntactically valid queries that carry the wrong business meaning.
The risks, however, are clear. Automatic extraction remains inherently probabilistic. The system may misunderstand fields, inherit incorrect definitions, or mistake temporarily missing details for evidence that no target exists. In the article's example, only 7 of 8 calls found the relevant object details, while 1 still required further investigation. Conflating “prefix not matched” with “no object details found” would directly distort the conclusion. Moreover, the logs, object counts, and rules in the SLS example are demonstrations. Permission constraints, cross-system joins, historical dirty data, and conflicting definitions across teams in real production environments do not disappear simply because semantics have been extracted.
Impact on regular people
For enterprise IT: Log analysis is beginning to shift from a tool only people who know how to write queries can use toward a natural-language interface. Deployment still depends on standardized fields, metrics, and access-control rules. In the short term, IT teams are more likely to gain a tool that reduces repetitive troubleshooting time than fully unattended operations.
For individual careers: Business professionals should be able to complete fault diagnosis and explain metrics faster, but validating results and defining boundaries will remain central to the job. People who can articulate data definitions and identify agent errors will be more valuable than those who merely submit questions.
For the consumer market: This capability may gradually enter cloud customer service, e-commerce operations, and user-feedback analysis, reducing the cost of manual querying and organization. Its reliability, however, depends on the quality of enterprise data. Consumers should not assume that statistical conclusions are equally reliable just because the interface feels more natural.