Tableau

What Tableau is and the problem it addresses

Tableau is a Salesforce family of visual analytics and business intelligence software. It turns structured data into views that people can explore: comparing categories, following a time series, locating items on a map or investigating relationships between variables. Inputs can come from spreadsheets, relational databases, data warehouses and connected services.

Its distinctive approach is to build a view progressively from fields. An analyst places dimensions and measures, changes the representation and receives visual feedback. VizQL connects these operations to queries and graphical encoding. This reduces the need to write a new query for every chart, while still requiring decisions about what to measure, how to group it and what the result means.

The product grew out of Stanford research into multidimensional databases and visualisation, with Polaris as a predecessor. Tableau combines that analytical heritage with enterprise publishing, data preparation and AI features. The brand now covers several products; buying one does not automatically provide all of them.

Reference documentation: Tableau — What is Tableau? · Tableau — History of innovation · Stanford — Polaris research · Salesforce — Acquisition completed.

History, ownership and corporate development

Christian Chabot, Chris Stolte and Pat Hanrahan founded Tableau in 2003. Its initial operations were in Mountain View before moving to Seattle. Venture funding and a 2004 OEM agreement with Hyperion helped turn research into a business. The official chronology dates Tableau 1.0 to April 2005.

• 2010: Tableau Public introduced a free place to publish openly accessible visualisations, particularly useful in data journalism and education.
• 2013: the company listed in New York under the DATA symbol. Its public-company period supported expansion; the 2018 annual report recorded approximately $1.155 billion in revenue, a historical figure that does not describe its current business within Salesforce.
• 2015–2018: Vizable explored touch interaction on iPad; Tableau subsequently expanded preparation and automated analysis. The acquisition of Empirical Systems, announced in June 2018, brought expertise in automated statistics.
• 2019: Salesforce completed the acquisition on 1 August. The closing date is distinct from the deal’s announcement in June.
• Subsequent development: the dashboard ecosystem added metrics through Pulse, AI assistance and Tableau Next, connected to the Salesforce platform.

Leadership also changed: Adam Selipsky succeeded Chabot in 2016; Mark Nelson led the company from March 2021 to December 2022; Ryan Aytay took over in 2023 and departed in 2026. An interview published in May 2026 identified Mark Recher as executive vice president and general manager of Tableau. Older references to its leaders or independent-company status therefore need a date.

Reference documentation: Tableau — History of innovation · Salesforce — Acquisition completed · Tableau — New Public policy, 2011 · Tableau 2018 Form 10-K · Tableau — Empirical Systems acquisition · Tableau — Mark Nelson biography · TechTarget — Interview with new Tableau leader · Tableau press release — Vizable launch.

Visual analysis supported by a data model

Tableau lets users build analytical views and bring them together in a dashboard. A dashboard combines perspectives on data; its value depends on each visualisation answering an understandable question. More charts do not necessarily improve interpretation. Identify the analytical unit, reporting period and decision each view can support before designing it.

The model distinguishes a logical layer, where tables are related, from a physical layer, where joins and unions are performed. This matters when tables have different levels of detail. A site table, an inspection table and an observation table do not represent the same kind of row. Preserve that distinction so an inspection is not counted repeatedly because it has several observations. Document the relationship before relying on any resulting totals.

Reference documentation: Tableau Help — The Tableau Data Model · Tableau Help — Dashboards.

Products, platforms and ways to use them

• Desktop: the Windows and macOS authoring application for analytical models, worksheets, dashboards and publishing.
• Cloud and Server: environments for hosting, administering, exploring and sharing analytics. Cloud is hosted; Server runs on infrastructure managed by the organisation or its provider. Both include web authoring, with differences from Desktop.
• Prep Builder and Prep Conductor: Builder creates cleaning and transformation flows; Conductor schedules published flows, depending on the purchased edition.
• Public and Reader: Public distributes open visualisations; Reader consumes Desktop workbooks. Neither should be mistaken for a private enterprise governance platform.
• Mobile: an iOS and Android application for consuming Cloud or Server content. Vizable was a separate historical iPad data-exploration product, not another name for Mobile.

A common workflow uses Prep to harmonise sources, Desktop to build the analysis, and Cloud or Server to distribute it. Assign ownership to each stage: changing a source field can affect the flow, calculations and final dashboard. Check compatibility before upgrading Desktop when Server remains on an older release.

Reference documentation: Tableau — Products and features · Tableau — Desktop and Prep installation · Tableau press release — Vizable launch · Tableau Desktop · Tableau Cloud · Tableau Server · Tableau Prep.

Fields, calculations and visual grammar

A dimension organises categories such as site or activity; a measure usually expresses a quantity such as hours or record count. Discrete versus continuous is a separate distinction: it determines whether a view creates individual headers or a continuous axis. A numeric site identifier should not be summed simply because it contains digits.

Marks encode data through position, colour, size, shape or text. Worksheets can contain bars, lines, scatter plots or maps; dashboards combine worksheets and support interaction. A filter changes the included data, whereas a parameter supplies a value for a calculation or action. Designing interaction means explaining what changes when someone selects a mark.

Calculations can operate on individual rows, aggregates or level-of-detail expressions, known as LODs. Averaging individual percentages is not necessarily equivalent to dividing their total numerators by their total denominators. Filters also have an execution order: a FIXED expression and a table calculation may respond differently to the same visible filter. A useful check is to predict a small sample’s result and reconcile it after each change.

Reference documentation: Tableau — Dimensions and measures · Tableau — Level of detail expressions · Tableau — Order of operations · Tableau Help — Dashboards.

Sources, live connections and extracts

Tableau provides connectors for files, relational databases, cubes and data services. The available set depends on the product, drivers and release. Check types, names, nulls and keys when connecting: a date interpreted as text or an identifier converted into a number can distort the whole analysis.

A live connection queries the source; an extract retains a representation of data for analysis and needs a refresh strategy. Modern extracts use Hyper format. Extracts can reduce load on an operational system, but require a visible cut-off date. Live queries do not guarantee completeness either if the source has not yet received all records.

A Tableau data-source definition is different from an extract: the former describes connection details and metadata; the latter contains data. Review what travels inside packaged workbooks before distributing them. Bridge can connect Tableau Cloud to private-network sources. Virtual connections and their data policies centralise access and row-level security; these are different controls from permission to open or download a workbook.

Reference documentation: Tableau Help — The Tableau Data Model · Tableau — Extract your data · Tableau — Virtual connections and data policies.

Maps, geocoding and geographical analysis

Tableau can interpret supported locations through geographic roles or plot latitude and longitude coordinates. It also connects geometries in Shapefile, KML, GeoJSON, TopoJSON, MapInfo and Esri file geodatabases. Place names and geometries serve different purposes: a postal code needs a geographical match; a polygon already describes a boundary.

Analysis can use points, areas, geographical hierarchies or custom groupings. In OSH, a site map might show inspection coverage or outstanding actions with access to supporting detail. A map of absolute counts tends to emphasise areas with more activity. For comparisons, add exposure, coverage and sample size instead of colouring a region as though colour demonstrated its risk level.

Review unknown locations, duplicate place names, reversed coordinates and changed administrative boundaries. Excessively precise locations can reveal information the audience does not need; a site or broader area may be sufficient.

Reference documentation: Tableau — Spatial files · Tableau — Mapping tutorial.

Licensing, governance and artificial intelligence

The established commercial roles are Creator, Explorer and Viewer: authoring, exploring authorised content and consuming it are different needs. Effective permissions also depend on site roles and access rules, not just the licence. Enterprise and Tableau+ bundle additional capabilities; list authors, readers, scheduled flows, private sources and administrative requirements before choosing.

Pulse brings metrics and explanations into the working environment; Tableau Agent assists with creating or understanding analytics. Catalog supports lineage, impact analysis and data-quality warnings. Tableau Exchange extensions and connectors broaden the possibilities while adding dependencies to maintain.

Tableau Next is a distinct platform, built within Salesforce and integrated with Data 360 and Tableau Semantics. It is not simply an upgraded Desktop workbook. When assessing AI, establish which product runs the feature, which data it accesses, how the metric is defined and which licence or consumption entitlement it needs. Generated explanations should be traceable to an indicator and its population; they do not turn correlation into causation.

Reference documentation: Tableau — Products and features · Tableau — Virtual connections and data policies · Tableau Pulse · Tableau Agent · Tableau — Licence types.

Open publication and a historical controversy

In December 2010, Tableau removed WikiLeaks visualisations from Tableau Public. The company said it acted after a public request from US Senator Joe Lieberman and referred to its terms of use. The episode prompted debate about platforms’ power to remove information.

On 21 February 2011, Tableau announced a revised data policy: a more formal complaints process, closer alignment with online free-expression principles and an advisory board. These are historical events concerning the governance of a publishing service, not evidence of a current security incident.

The practical implication for a visualisation author is to identify the publication destination from the outset. Tableau Public is intended for public content. An anonymised chart is insufficient if its workbook or downloadable data still exposes identifiable information. Review the entire published package, not only the visible image.

Reference documentation: Tableau — WikiLeaks removal explanation · Tableau — New Public policy, 2011 · The Guardian — WikiLeaks visualisation removal · Tableau — Products and features.

Uses that require exploring differences and context

• Work observations: examine their distribution by activity, shift or site and identify gaps. Record volume needs interpretation alongside the observation activity undertaken.
• Time-based review: compare indicator changes with changes in recording practice, scope or the included population.
• Action review: combine age, status and record-level detail to prepare specific questions for action owners.

These scenarios proposed by Sabentis use Tableau for exploration. A visual difference between sites may justify investigation, but does not establish that one site is safer without considering exposure, coverage and data quality. Treat the chart as a way to focus a question, rather than as an explanation of the answer.

Worked example: avoiding duplicate inspection counts

Hypothetical example. A team holds one inspection register and another listing the observations associated with each inspection. It wants to compare activity and findings without mixing those two counts.

1. Define site and inspection identifiers. Check that each inspection appears once in its own table and every observation has a valid reference.
2. Model the relationship while preserving the units: one row per inspection and one per observation. Isolate unmatched observations for review.
3. Build a view of completed inspections and another of observations by category. Label the two counts unambiguously.
4. Create a dashboard with period and site selection, displaying active filters. Include detail allowing a total to be reconstructed.
5. Test an inspection without observations and one with several. Reconcile with source records and confirm that the inspection total does not increase with the number of findings.

Acceptance depends on that reconciliation and on recipients interpreting the output correctly. A site with more observations may have conducted more inspections or used a different recording criterion. Retain those explanations beside the figures where they affect comparison.

Proposed numerical check: with 12 inspections, four having no observations and eight having two each, the expected result is 12 inspections and 16 observations. An inner join that removes inspections without findings would leave only eight identifiers and 16 rows. The dashboard should retain all 12 inspections, show the four without observations and explain whether 16/12 represents observations per inspection. It must not be presented as the percentage of exposed workers or an accident frequency rate.

Reference documentation: Tableau Help — The Tableau Data Model · Tableau Help — Dashboards.

What visualisation can conceal

Colour, scale and aggregation can exaggerate or hide differences. Avoid inconsistent scales across views intended for comparison and explain categories containing few records. An average may conceal extreme cases; detail and context help determine whether further investigation is needed.

Table relationships do not correct invalid identifiers or changing categories. A successful technical refresh does not establish population completeness either. Review connections, reporting date, exclusions and publication permissions. Before sharing, check whether recipients can access underlying detail and whether that detail is authorised for them. Verify the publication environment and licensing for the actual deployment. Avoid presenting a demonstration environment as proof of production access controls.

Criteria for choosing Tableau

Tableau can add value where visual exploration, interaction between views and presenting data to different audiences are regular activities. It requires people able to model, interpret and maintain reports; a visual tool does not remove those skills.

Compare Power BI against sources, infrastructure, distribution needs and team knowledge. Excel may remain useful for preparing or reconciling a small dataset. Keep a common indicator definition even where several tools display it, with a named owner responsible for changes.

Frequently asked questions about Tableau in OSH

Is a relationship between tables the same as a join?

No. Tableau distinguishes logical-layer relationships from physical-layer joins. The choice affects how data at different levels of detail is combined.

Can a colour map classify risks?

It represents the rule defined for it. Classification still requires an appropriate method and data; colour provides no independent validation.

What should be checked before presenting a dashboard?

Totals, filters, periods, units, excluded records, refresh status and detail access. Also confirm that readers understand the question each chart answers.

Reference documentation: Tableau Help — The Tableau Data Model.