Quick answer
A data model describes what a dataset contains and how it connects: the entities you keep records about, the attributes of each, and the relationships between them. In analytics, the core of the model is how events, users and their properties link together.
What is a data model?
A data model is a written description of what a dataset contains and how the pieces connect: which entities exist (things you keep records about), which attributes each one has, and which relationships link them. It is the blueprint that comes before tables, dashboards or queries.
A shop's data model, in one sentence: "A customer places many orders; an order contains many products; each product has a price and a category." Three entities, three relationships, a handful of attributes. Anyone who reads it knows what questions the data can answer, and which it cannot.
For analytics, the most important part of the model is how three things connect: events (what happened), users (who did it) and properties (the details of each). Get that right and funnels, retention and segments all follow; get it wrong and numbers disagree between reports.
What are the three levels of a data model?
Models are drawn at three levels of detail, each for a different reader.
| Level | Describes | Written for | Example (shop) |
|---|---|---|---|
| Conceptual | The main entities and how they relate, in business words | Business owners, product managers | Customer — places — Order |
| Logical | Every entity's attributes, keys and relationship rules, independent of any database | Analysts, data designers | Order has order_id (key), customer_id, placed_at, total |
| Physical | The actual tables, columns, data types and indexes in a specific system | Engineers | orders(order_id BIGINT PRIMARY KEY, customer_id BIGINT, placed_at TIMESTAMP, total DECIMAL(10,2)) |
Work top-down: agree the conceptual picture first, because arguments about column types before agreeing on what an "order" is waste time. Codd's relational model (1970) and Chen's entity–relationship diagrams (1976) are the roots of this vocabulary.
What are the common types of data model?
| Type | Shape | Good for | Watch out for |
|---|---|---|---|
| Relational | Tables joined by keys | Transactions, consistent business records | Joins get slow at very large event volumes |
| Dimensional (star schema) | A central fact table of events or sales surrounded by descriptive dimension tables | Reporting and BI | Needs planning of which dimensions you query by |
| Document | Nested records (JSON) | Flexible, changing attributes | Harder to enforce the same fields everywhere |
| Graph | Nodes and edges | Networks of relationships | Less natural for plain aggregates |
| Event-based | An append-only stream of "who did what, when, with what details" | Behavioral analytics | Needs strict naming, or the same action gets recorded under many names |
Most product-analytics tools use an event-based model on top of a dimensional layout: one big fact table of events, plus a profile for each user.
What does an event model look like?
Sample data: a three-row slice of an event model for a web shop.
| event | time | visitor | user (after login) | properties |
|---|---|---|---|---|
| page_view | 10:02:11 | v_8f3 | — | page = /pricing, source = google |
| signup | 10:04:40 | v_8f3 | u_204 | plan = free |
| purchase | 10:19:05 | v_8f3 | u_204 | order_total = 49.00, currency = USD |
And the user profile, stored once:
| user | email domain | country | created |
|---|---|---|---|
| u_204 | example.com | DE | 10:04:40 |
Read the model off the example:
- An event has a name, a time and a person. The same name always means the same action.
- Event properties describe that one occurrence (the page, the order total). They are different for every row.
- User properties describe the person and live on the profile (country, plan). They change slowly.
- Identity is the join: visitor v_8f3 became user u_204 at signup, so the earlier page view can be attributed to the same person.
Where should a property live: on the event or on the user?
Ask: is this a fact about this moment, or about the person?
- "Order total", "page path", "button label" → event property.
- "Country", "signup date", "plan today" → user property.
- "Plan at the time of purchase" → copy it onto the event, because the user property will change later and history should not.
The cost of getting it wrong is real. Sample data: 10,000 users with 120 events each is 1,200,000 event rows. Copying six user attributes onto every row stores 1,200,000 × 6 = 7,200,000 values; keeping them on the profile once stores 10,000 × 6 = 60,000. Duplicates also drift: if one row says "plan = free" and another says "pro" for the same person, which is right? Store what must stay constant on the event, and keep the rest in one place.
What are the design principles for an analytics data model?
- One name per action.
signupeverywhere, neversign_up,Signupandregistered. Names are the key to every report. - Few events, many properties. Twenty well-named events with properties beat five hundred one-off events. Put "which button" in a property rather than creating a button_a_click event.
- Consistent types. A property is a number everywhere or a string everywhere. A price stored as "49.00" in some rows and 49 in others breaks sums.
- Define keys on purpose. Decide which id means a visitor, a device, a person and an account, and how they join.
- Write it down. A model nobody can read is not shared. A data dictionary lists each field's meaning, and an event schema pins down the structure of each event.
For how several systems are merged into one model, see unified data model.
What mistakes do teams make with data models?
- Modeling the screen, not the behavior. Events named after UI elements break every redesign. Name the action (
add_to_cart), not the widget. - Free-text values where a list belongs. "Plan = Pro", "pro", "PRO" are three values to a computer.
- No owner. If nobody decides whether a new property is allowed, the model grows without bound.
- Changing meaning silently. If
purchaseused to include trials and no longer does, old and new data are no longer comparable. Record the change date. - Collecting everything "just in case". Every property you do not use still has to be maintained, and some may be personal data you should not keep.
How does TapCub model your data?
TapCub's model is event-based with user and account layers, and it is built to be written down before it is built. Following the web-shop example:
- Plan the events first. In All-platform analytics › Data › Tracking plan, each row records an event you intend to collect: name, display name, category, platforms, trigger, properties (name, type, required or not, example), owner and status. Planned events can be listed before any code is written.
- Check what arrives. Turn on live verification (24 h) to see, per event, how many hits arrived and how complete the required properties are. New events nobody planned land in All-platform analytics › Data › Event inbox to be claimed.
- Control the properties. All-platform analytics › Data › Property dictionary has separate User, Event and Account tabs. Each key has a type (String, Number, Boolean, Datetime or List), a sensitivity label and optional validation rules such as allowed values.
- Connect people. An anonymous visitor has a visitor id; after your site calls
tapcub.login, the person appears in All-platform analytics › Users › People, and earlier anonymous activity on that device is attached to them. Login needs identified mode, which depends on your plan. Counting by visitor, device, person or account is a choice you make per analysis. - Add the account layer if you sell to companies. After
tapcub.group('company', 'acme', {...}), people roll up to accounts in All-platform analytics › Users › Accounts.
TapCub does not design the model for you or enforce a particular warehouse schema; the plan, dictionary and validation rules are the tools for keeping your own model consistent.
Sources
- Codd, E. F. (1970). A relational model of data for large shared data banks. Communications of the ACM, 13(6), 377–387.
- Chen, P. P.-S. (1976). The entity-relationship model: toward a unified view of data. ACM Transactions on Database Systems, 1(1), 9–36.
- Microsoft Learn, Understand star schema and the importance for Power BI.
Was this explanation clear?
We use your answers to fix and expand this page.
Couldn’t send. Please try again in a minute.
Thanks for the feedback. We’ll use it to improve this page.You’ve already sent feedback on this page. Thanks.

Data Analyst
Explores traffic sources, conversion funnels, and retention, using clear metric definitions and concrete examples to explain what the numbers mean.