SQL
Structured Query Language
What is SQL?
- Query, store, and update data in a database
- The foundation of working with data
- Useful for non-developers too
When even one person on a team can work with SQL, what changes is not the technology but the distance between a question and its answer. Checking 'How many repeat customers did we have last quarter?' used to mean filing a ticket with the dev team and waiting three days. Now it takes three minutes, done yourself. The bigger change is that people ask more questions. Once each question costs less, hypotheses that used to be waved through on gut feeling get tested against real numbers.
Commands fall into three groups: DML (SELECT, INSERT, UPDATE, DELETE), which reads and changes data; DDL (CREATE, ALTER, DROP), which creates and modifies table structure; and DCL (GRANT, REVOKE), which grants and revokes permissions. Business users realistically need just SELECT, so most companies issue only read-only accounts. Dialects differ by product too: Oracle, MySQL, PostgreSQL, and BigQuery each handle date functions and string operations differently, so before choosing a textbook, first check which database your company actually uses. The boundary with Excel is clear, too: Excel struggles past a million rows, while SQL filters tens of millions of rows by condition in seconds.
The most common mishap in practice is connecting directly to the production database. Run a full-table aggregation query against the server that handles live orders, and the checkout screen freezes. Analysis belongs on a replica or a data warehouse. Cases where a single UPDATE missing its WHERE clause wiped out every customer's tier are not rare. And learning SQL does not mean knowing your data. The hard part is not the syntax but how your company defines its data, such as whether status value 7 in the orders table means a cancellation or a return. A wrong join condition duplicates rows and doubles reported revenue, and if that number goes straight into a report, no one catches it. Even now, with AI writing queries for you, judging whether the result is right is still a human job.
Seolyeong ParkGenerative AI and AX training expert with 17 years of hands-on experience and 200+ lectures, who gets non-specialists using AI right away
John KimJohn Kim, CEO of Nanum Consulting and author of “B2B Marketing Design”Developed at IBM in the 1970s for relational databases, it has since become the international standard for querying data.
Consider a hypothetical case: Company A, a household goods manufacturer and distributor with 320 employees and KRW 90 billion (≈ US$67,335,000) in annual revenue. Its six-person marketing team was sending the dev team more than 40 data extraction requests a month, and each took an average of 5 business days to get a reply. Again and again, the customer list arrived only after the campaign planning was already done.
- The data team lead sorted the last six months of extraction requests, 240 in all, and found that 183 were 'the same query with a different date range or filter'. In the meeting, someone said, 'It would be faster for the business teams to pull this themselves.'
- To keep the production database untouched, they set up a separate analytics replica that syncs overnight, and issued read-only accounts to 12 people in marketing and planning, limited to running queries.
- Instead of grammar lectures, the training had everyone bring three questions from their own work. Over four weeks, at two hours a week, each person wrote queries such as 'average order value of last month's first-time buyers' and had the results verified.
- After two reports went up early on that counted cancellations as revenue, the team posted a one-page data dictionary on the internal wiki, laying out the seven order status values and the distinction between returns and exchanges.
- Three months later, extraction requests to the dev team dropped from 40 a month to 9, and segment checks went from 5 business days to same-day. Repeat-purchase campaigns now run weekly instead of monthly.