Tableau interview questions on LODs, filters and extracts

Field names assume a hypothetical orders data source with Order ID, Order Date, Customer ID, Region, Category, Sub-Category, Sales and Profit.

Fundamentals the rest depends on

1. What is the difference between discrete and continuous fields?

Discrete fields (blue pills) create headers and separate panes, one per value. Continuous fields (green pills) create an axis. Dimension or measure is a separate idea: it describes whether the field is aggregated. YEAR(Order Date) as discrete gives a column per year; as continuous it gives a single time axis. Red flag: "blue is dimension, green is measure." Colour follows discrete/continuous, not dimension/measure.

2. In what order does Tableau apply filters?

Extract filters, then data source filters, then context filters, then dimension filters, then measure filters, then table calculation filters. The order matters because a FIXED LOD is computed after context filters but before ordinary dimension filters. Red flag: not knowing that a Top 10 filter and a Region filter can produce different results depending on whether Region is added to context.

3. A manager wants the top 10 customers within the selected region, but the view shows the top 10 overall filtered down to that region. Fix it.

The Top N condition is a dimension filter and is evaluated at the same stage as the Region filter, so it ranks customers across all regions. Right-click the Region filter and choose Add to Context. Context filters run first, so the Top 10 is then computed within the selected region only.

Level of detail expressions

4. Calculate each customer's first order date and show sales by acquisition year.

First Order Date = { FIXED [Customer ID] : MIN([Order Date]) }
Acquisition Year = YEAR([First Order Date])

Put Acquisition Year on columns and SUM(Sales) on rows to see revenue by customer cohort. FIXED computes at the customer level regardless of what is in the view.

5. FIXED, INCLUDE and EXCLUDE — when do you use each?

Red flag: saying FIXED "ignores all filters." It ignores dimension filters but respects context, data source and extract filters.

6. LOD or table calculation for percent of total?

A table calculation (SUM([Sales]) / TOTAL(SUM([Sales])), or the quick calculation Percent of Total) works on what is in the view, is quick to build and changes if you filter the view — which may be what the user wants. An LOD computes in the data source and stays stable regardless of view layout. Pick based on whether filtering a category out should change everyone else's share. Ask the stakeholder rather than choosing silently.

Data connections and performance

7. Relationships, joins or blending?

Relationships (the default logical layer since version 2020.2) keep tables separate and join them at query time at the right level, which avoids duplication when tables have different grains, such as orders and monthly targets. Physical joins merge tables into one before aggregation and can inflate sums if the grain differs. Blending combines data from separate data sources on a linking field, aggregated per source, and is mostly used when the sources cannot be combined any other way. Red flag: joining monthly targets to order lines and summing the target.

8. Extract or live connection?

An extract (.hyper) is a snapshot, fast to query, can be filtered and aggregated to reduce size, and is refreshed on a schedule. Live connections always show current data but put load on the source database and are only as fast as it is. Choose live for near real-time operational views on a fast warehouse, and extracts for most management reporting. Mention that extract filters permanently remove data from the extract, so they should be documented.

9. A dashboard takes 20 seconds to load. How do you investigate?

Use Performance Recording (Help, Settings and Performance) to see whether time goes to queries, layout or rendering. Common causes: too many sheets on one dashboard, quick filters showing "only relevant values" on high-cardinality fields, marks in the hundreds of thousands, complex row-level string calculations, and live connections to slow sources. Fixes include an aggregated extract, fewer filters, action filters instead of quick filters, and moving heavy calculations upstream into SQL.

Design judgement

10. The business wants a dashboard with 14 KPIs on one screen. What do you do?

Ask what decision each person makes with it and how often. Usually that splits into three to five headline KPIs with a clear comparison (against target or last period), then drill-down views reached by action filters. Keep colour for meaning, such as below target, not decoration. Show the data refresh time on the dashboard so nobody makes a decision on yesterday's numbers without knowing it.

How to practise this bank

Build a small workbook from any public retail sample data and deliberately recreate questions 3, 4 and 6. Then change one filter at a time and write down, before you look, whether each number should change. Being able to predict the effect of a filter is the skill every question here is probing.

If you are deciding between Tableau and Power BI for your first reporting tool, compare these questions with the Power BI and DAX interview bank. Dashboards in the Data Analyst programme are assessed on whether the stated question is answered, not on how many charts they contain.