Data warehouse

Each system knows only its own slice. The ad platform knows the click, the CRM knows the deal and the billing system knows how much the customer paid and for how long. A warehouse combines them, so the return by channel becomes something you calculate rather than guess.

It also lets you look back. Operational tools often show only the current state of a record, while a warehouse can keep history, so you can compare cohorts of customers from the same month and see how they behave over time.

It creates accountability as well. A team can see which campaigns and segments actually perform, and everyone reads the same numbers. The price is set-up, definitions and upkeep, so it pays when decisions need data from more than one system.

• Start with the systems that hold your key decisions: the CRM, the marketing tool and billing.
• Load raw data first and reshape it inside the warehouse, so remodelling later never means extracting again.
• Agree what a customer, a sale and revenue mean before building reports. A managed warehouse can be set up in an afternoon, but those definitions take longer, and that is where an analytics engineer helps most.
• Build a handful of shared reports instead of letting every team query it its own way.
• Add a new source only once the current ones are trusted and used.

Business tools such as a CRM, a billing system and an ad platform are built to run daily operations, one record at a time. A data warehouse is built for the opposite job: reading large amounts of data to answer questions that cut across systems. Well-known examples are Google BigQuery, Snowflake and Amazon Redshift.

Data reaches the warehouse through a data pipeline, and it stays separate from the live tools, so a heavy report never slows down the CRM. The warehouse holds a copy. The original system stays the place where records are created.

Take the question "which channel brings customers who stay for more than a year?" The ad platform knows the click, the CRM knows the deal and the billing system knows how long the customer paid. In a warehouse that question becomes one query. Without it, someone matches names across three exports by hand.

• Buying a warehouse when a spreadsheet would still do. A few thousand rows from two tools rarely need one.
• Treating it as a dumping ground with no definitions, so the same word means three things.