What Is a Dataset in CRM Analytics?

Creating datasets is the first real step in any CRM Analytics (Tableau CRM) project. Before you can build a single dashboard, tell a story, or train an Einstein Discovery model, you have to get data into the platform and shape it into something you can query. That "something" is a dataset, and this whole section is about building them well.

After seven years of building on this platform, here is the honest version: nine times out of ten, your work starts by creating a dataset. There are exceptions — direct queries and live connections — but those solve narrow problems inside a specific dashboard. The bread and butter is the dataset.

So what actually is a dataset?

A dataset is a storage format. You take objects or tables that relate to each other, combine them, and store the result in a shape that is purpose-built for one analytics need.

Say you want to analyze opportunities. You would pull in the Opportunity object, but you'd also bring the Account each opportunity belongs to, the User who owns it, and maybe — down the road — related activities. All of that gets combined into one wide, self-contained dataset about opportunities. That is the exercise we'll do by hand later in this section.

Why it's so fast

You don't need this on a daily basis, but it helps to know what's under the hood. A dataset uses a flattened, inverted-index, key-value storage system — a fancy way of saying it's a columnar file store with a proprietary compression and indexing algorithm.

The payoff is the whole point: you can store large volumes of data in a dataset and still query it fast enough to power an interactive dashboard. That combination — big data and sub-second response — is why the platform bothers with a dedicated storage format instead of just querying Salesforce directly.

When not to build a dataset

Datasets are the default, but they aren't the only option. CRM Analytics also supports:

• Direct queries — query a small set of Salesforce data on the fly, without materializing a dataset.
• Live connections — read directly from an external source (like Snowflake) at query time.

Both are real and useful, but they fit a specific dashboard component or requirement, not the general case. If you're just starting out, assume you're building a dataset and treat live/direct as the exception you reach for deliberately.

What's ahead in this section

We'll cover the three practical ways to create datasets, from simplest to most powerful:

By the end you'll have built the same opportunities dataset two different ways and then combined it with external CSV data — exactly the muscle memory you need on real projects. Next up: the data landscape that shows where datasets sit in the bigger picture.