Data Warehouse · Lake
Data Warehouse / Data Lake
What is a data warehouse?
- Warehouse = an organized storehouse for analysis
- Lake = a reservoir that holds data as-is
- Choose by purpose
Choosing a storage system is really a decision about when you will organize your data. A warehouse requires you to fit the table structure before loading; a lake takes everything in first and interprets it at read time. That is what decides your response speed when a business unit says, 'Could you add just one more metric?' A warehouse takes 2–3 weeks to adjust the model and reload, but the numbers stay stable; a lake can be opened today, but you end up asking again each time who vouches for those numbers.
Three practical criteria separate the two. The first is form: structured data with fixed rows and columns, such as orders and accounting, suits a warehouse, while varied data such as app logs, call recordings, and images suits a lake. The second is cost: object storage that simply holds raw data and a warehouse that includes an analytics engine differ by an order of magnitude in storage cost per TB. The third is sequence: whether you clean data before loading it (ETL) or load it first and clean it afterward (ELT). A lakehouse narrows this boundary by layering table formats and transactions on top of a lake’s storage. A data mart is a subset carved out of the warehouse for a specific department, so it is not a separate option.
The point where teams most often go wrong is the belief that “if we store everything now, we’ll use it someday.” Pile up data for three years with no catalog and no owners, and out of 800 tables no one is left who can explain what the columns mean. The lake becomes a swamp no one can use. The opposite misconception is also common: assuming that having a lake means no warehouse is needed, teams connect the executive dashboard directly to the raw data. Every executive query then scans several TB, and the monthly cloud bill soars. What you must decide before loading is not the tool but who owns this data.
Jaryong KooMarketing PhD and authority in AI, data analytics, and personal branding
John KimAuthor of B2B Marketing Design and CEO of Nanum Consulting| Category | Data warehouse | Data Lake |
|---|---|---|
| Storage format | Organized, structured data | Raw, as-is (including unstructured) |
| Use case | Instant analysis and reports | Store first, use later |
| Analogy | A tidy warehouse | A lake that holds every kind of raw material |
The data warehouse was established in the 1990s by Bill Inmon and others; the data lake emerged in the big data era (around 2010) to hold raw, unstructured data as well.
Consider a hypothetical case. Company A, a household goods distributor with 410 employees and annual revenue of KRW 210 (≈ US$1) billion, kept order and member data neatly in a warehouse but left app logs and call recordings sitting on a file server. The churn prediction model the marketing team requested fell through twice because “there was no raw data to train on,” and adding a single new metric took 18 days on average.
- Kim, who leads the data platform team, moved 4.2 TB of scattered app logs from the past three months into object storage as is, and attached a catalog listing just three fields: load time, source, and owner.
- When the CFO asked how much storage costs would grow, Kim, the team lead, changed the setup to keep raw data in the lake and only refined data in the warehouse, cutting warehouse load to one-third of its previous level.
- Lee, a marketing analyst, trains the churn prediction model by joining order history to the lake’s logs, but fixed the rule that validation labels come only from the warehouse’s confirmed-revenue table.
- As requests from business teams piled up, each table got a data owner: Sales Management owns the order definitions and Development owns the logs. The rule became: no table without an owner goes into the catalog.
- Six months later, the time to add a new metric dropped from 18 days to 4. The churn prediction model now identifies 61% of customers who cancel within three months before they leave, and the team uses it to choose who receives repurchase coupons.