KajoData / Career & business
A Tableau sales dashboard: parameters, filters and layout
I connect sales views into one report while keeping their amounts, filters and explanations consistent on desktop and phone.

A sales dashboard should help explain what is happening to sales and where to investigate. Yet four individually correct charts can start describing four different populations after a single click. When that happens, the report has a logic problem even if every margin is perfectly aligned.
I start with measure definitions and interaction scope, then arrange the containers, colours and spacing. This guide builds a small Tableau example containing a KPI, a trend, a selectable breakdown and a detail table. The data is synthetic and the amounts are small enough to reconcile by hand. The instructions use this explicit example and the Tableau documentation checked on 22 November 2026, rather than relying on access to a private course workbook.
Establish the audience and questions
Assume the reader wants to inspect the overall sales amount, its distribution over time and its breakdown by category or shipping mode. They also need to trace a chart back to a few records. That is enough to design a useful first screen. I avoid immediately adding a second dimension, an employee ranking and a collection of metrics whose meanings have not yet been agreed.
The KPI answers “how much sales is in the selected scope?”. The trend distributes that same amount over time. The bars split it by a selected dimension, and the table explains which records contribute. These elements should reconcile. If one panel deliberately retains a broader scope, I explain that beside the panel so that the reader can distinguish a design choice from a broken filter.
I also record what the report does not measure. This example has no returns, costs, taxes or multiple currencies. Sales is an illustrative amount in a common unit. I would not relabel it as margin or net revenue. In a real project, those definitions need agreement with the person responsible for the process before a large number is placed at the top of a dashboard.
Start with a small reference table
Each row represents one order. OrderID is unique, and the dates cover two months. I retain English field names so that calculations are identical in both language versions of this guide. Display labels can be translated later without changing the definitions of the underlying fields.
| OrderID | Order Date | Category | Shipping Mode | Sales |
|---|---|---|---|---|
| 1 | 2026-01-05 | A | Standard | 100 |
| 2 | 2026-01-20 | B | Express | 50 |
| 3 | 2026-02-03 | A | Express | 80 |
| 4 | 2026-02-12 | B | Standard | 70 |
The overall amount is 300. January contributes 150 and February also contributes 150. Category A contributes 180 and B contributes 120. Standard shipping contributes 170 and Express contributes 130. These are different partitions of the same total. Switching the grouping dimension should change the distribution of bars while leaving the overall KPI unchanged.
Before building views, I check that the date and amount have the appropriate types. An order identifier identifies a record; it is not another measure to sum. If Sales arrives as text, I correct how the source is read. Adding a currency format to a label cannot repair a field that was imported with an unsuitable data type.
Build four worksheets independently
The first worksheet displays SUM([Sales]) as text. The second places the date on the horizontal axis and total sales on the vertical axis. The third compares categories with bars. The fourth shows OrderID, date, category and Sales in a detail table. I check each sheet separately before combining them into a dashboard.
With four one-row orders, row count and order count are the same. A real export might contain order lines, in which case COUNT([OrderID]) counts occurrences rather than unique orders. That simplified assumption should not move from the exercise into an unfamiliar dataset. Establishing the grain first prevents a technically valid aggregation from receiving an incorrect business name.
I give worksheets names that explain their purpose, such as KPI Sales, Trend Sales, Breakdown Sales and Details. Duplicating a sheet is convenient, but I inspect inherited filters, fields on the Marks card and sorting after making a copy. A copy can carry logic that is not immediately visible in the resulting chart. That deserves attention before matching the colour of every heading.
Use a parameter to choose date granularity
A Tableau parameter is a workbook variable. Creating it and showing its control does not change the chart on its own. I must use it in an element. I create a string parameter called Date granularity with Year, Quarter, Month and Week as allowed values, then define a calculated field named Trend date.
CASE [Date granularity]
WHEN "Year" THEN DATETRUNC("year", [Order Date])
WHEN "Quarter" THEN DATETRUNC("quarter", [Order Date])
WHEN "Month" THEN DATETRUNC("month", [Order Date])
WHEN "Week" THEN DATETRUNC("week", [Order Date], "monday")
END
DATETRUNC returns a date truncated to the requested level. For a month, that is the beginning of the month. The week branch explicitly uses Monday as its start, avoiding dependence on a default week setting. ISO and fiscal calendars need their own agreed definitions; different week conventions are not interchangeable.
In the trend sheet I replace the original date with Trend date and use continuous Exact Date. Sales remains a sum. Month produces two points of 150 each, while Year produces one point of 300. The parameter changes grouping granularity rather than filtering the dataset. This gives a small, concrete check that the control is connected to the intended calculation.
I avoid formatting the axis with month names alone if several years can appear. January in two different years must remain distinguishable. A date’s value and its label are separate concerns. Here the calculation defines the groups, while view formatting controls their presentation. The trend title should show the currently selected granularity so that the reader can interpret each point.
Use a second parameter for the bar dimension
I create a string parameter named Breakdown, allowing Category, Shipping Mode and Total. The calculated field Breakdown label returns text in every branch, so it can act as one dimension. I do not mix a text category and a numeric measure in this calculation; switching measures would require a separately defined approach to aggregation.
CASE [Breakdown]
WHEN "Category" THEN [Category]
WHEN "Shipping Mode" THEN [Shipping Mode]
WHEN "Total" THEN "Total"
END
I place this field on Rows and SUM([Sales]) on Columns. Category gives 180 and 120, Shipping Mode gives 170 and 130, and Total gives one bar of 300. The parameter’s allowed values must match the strings used in CASE. A restricted list prevents arbitrary input from leaving the calculation without a matching branch.
For the first version I keep one selectable dimension. A second independent choice would increase the combinations that need checking and might complicate reading. If a real question later requires category within shipping mode, the design can expand. The reader should not need to understand a control panel more complicated than the decision the report is meant to support.
Every filter has a worksheet scope
I add a Category filter and an Order Date range. According to the Tableau filter documentation, a filter applies to its current worksheet by default. Placing its control on a dashboard therefore does not ensure that every panel uses the same selection. I open Apply to Worksheets and deliberately choose the targets.
In this single-source example, both filters should affect all four worksheets. I can select them explicitly with Selected Worksheets or use All Using This Data Source. The former provides a visible list, while the latter also covers appropriate new worksheets. The choice depends on the workbook’s intended structure, so I consider sheets on other dashboards as well as the current screen.
Selecting A should produce a KPI of 180, table rows for orders 1 and 3, and trend amounts of 100 in January and 80 in February. Switching the bars to Shipping Mode should then produce Standard 100 and Express 80. These values follow from the two remaining records. A small reconciliation helps distinguish a grouping change from a filter that has stopped reaching a sheet.
Multiple sources need a model decision
Tableau supports filters across worksheets using related sources when the required relationships and configuration exist. The documentation also explains the role of the primary source when blending. I therefore avoid the blanket instruction to “never filter all sources”. First establish which worksheets should respond and whether the data model gives that filter a meaningful relationship to their records.
A cost table at another grain needs particular care. Monthly costs and product-level sales do not automatically give a category filter a defensible meaning for costs. Keeping the cost panel outside that filter may be appropriate if it is explained clearly. Allocating a monthly cost to products requires an additional business rule, which should not be invented merely to make every panel respond alike.
Relationships and physical joins behave differently. A join between tables with different multiplicities can duplicate an amount before aggregation. Relationships keep logical tables separate and build queries according to the view, but they still require correct matching fields. If a total grows after adding a dimension, I return to the model and grain before trying a different aggregation to hide the symptom.

Clicking a bar adds another interaction
A visible filter control differs from a filter action. An action sends values from a selected mark to target worksheets. I specify the source, targets and trigger. For the detail table I choose selection, because accidentally moving the pointer should not change the records while someone is reading them.
I also configure what happens when the selection is cleared. Should the details return to all records, remain restricted or disappear? In this exercise I choose to restore the full scope allowed by the other filters. The reader can then return to the starting point without refreshing the page. I check this setting in Actions rather than assuming its default will match the design.
With a dynamic dimension I use Breakdown label in the action and inspect every parameter choice. Clicking A should limit details to A; clicking Standard should limit them to that shipping mode. An action previously configured only on Category will not automatically understand the new role of the bars. When a view’s logic changes, its interactions need another review as well.
Design the layout for a known width
The Tableau layout documentation describes fixed, range and automatic sizing. I begin by establishing where the report will be viewed, then arrange the panels. Changing the dashboard size at the end can require substantial rearrangement, particularly when objects have precise positions.
I use tiled containers as the base: a vertical container for the main flow and horizontal containers for panels that belong side by side. Containers help preserve relationships among objects. Floating objects can be useful for deliberate overlays or exact positioning, but their behaviour needs checking at the selected sizes. Neither mode guarantees a well-designed report by itself.
I inspect inner and outer padding, title alignment and space for controls. I do not remove every tooltip automatically. A tooltip explaining a measure or showing an exact amount can be useful when written clearly. I remove technical field names and unnecessary repetitions instead. Presentation should help the reader understand the chart rather than requiring them to infer missing context.
A phone needs its own hierarchy
Device layouts support a phone version. Any element needed there must first exist in the Default layout. On a narrow screen I place the KPI, filters, trend and breakdown vertically. The detail table can follow them, since it normally requires more space and a more deliberate reading effort.
I check that parameters can be selected comfortably and long category names remain readable. I also inspect the state with one bar remaining and the state with no results. An empty panel should be understandable. A plain blank rectangle can look like a loading failure even when the filters have correctly removed every record from the view.
A separately edited phone layout needs maintenance as the workbook changes. After adding a new filter to Default, I return to the phone layout and inspect its presence and position. Automatically generated and manually edited layouts have different update behaviour, which the documentation explains. I treat the phone as part of the delivered report, rather than a shrunken desktop screenshot.
Save an understandable starting state
I save this example with Month selected for the date, Category for the breakdown and the full data range visible. That shows the total of 300 immediately. Saving after an accidental selection could make a small subset look like the entire population. Before publishing, restore the agreed starting state and provide a clear route back from a selection.
A short instruction beside the controls explains that filters restrict records, date parameters change grouping and clicking a bar opens details. These actions may not be obvious to a first-time reader. I also distinguish a filter with no matching orders from an existing order with an unknown Sales amount. They represent different situations and should not automatically become the same zero.
As a final usability exercise, find A’s January amount and return to the overall total outside edit mode. If that requires remembering hidden author settings, simplify the interface. The reader should be able to concentrate on the sales question and predict which panels will change scope.
Reconcile the final dashboard with four orders
I inspect several states: no filters, only A, only January, A within January and an empty range. I compare the KPI, bar totals and detail table. A within January should lead every applicable view to order 1 for 100. Clearing selections restores 300. Changing date granularity should leave that amount unchanged.
Separately, I click a bar, clear its selection and switch the breakdown parameter. I read the titles and labels to confirm that they describe the current state. Where needed, I include the reporting period and data refresh time. Appearance alone does not establish numerical quality. Clear formatting makes a report easier to read; measure definitions and scope make its values explainable.
For an exercise, add a fifth order and calculate the expected changes on paper first. Then change filters and parameters while predicting which panels should respond. If you also work in Python, compare this preparation with the guide to GroupBy and charts. Different tools still require the same care about data grain, scope and the meaning of a single number.
Key takeaways
Does a Tableau parameter filter data automatically?
No. Use it in a calculation, filter, reference line or another supported element. Showing a parameter control alone does not change a result.
Why do the KPI and chart show different amounts?
They may use different filters, sources or measure definitions. Compare those settings and the grain of the data before changing the formatting.
Does a tiled layout guarantee a good phone dashboard?
No. Tiled containers help organize objects, but the dashboard size and device layout still need to be designed and inspected separately.
AI was used in an automated way while working on this article. If you want to learn how to do it too, check out this project.
