Microsoft Excel

What Microsoft Excel is and what kind of tool it provides

Microsoft Excel is a spreadsheet application developed by Microsoft. It combines a grid of data, a calculation engine, visualisation tools and mechanisms for importing information and automating work. It is used for budgeting, reconciliation, planning, analysis and many other tasks that involve relating information through explicit rules. It belongs to the Microsoft 365 ecosystem and is also supplied in one-time-purchase editions of Office.

Its working unit is the workbook, which contains worksheets. A cell can hold text, a number or a formula; formatting controls how it appears. This distinction matters: displaying a value to two decimal places does not necessarily round the number used in calculations. Similarly, turning a range into a table does not automatically turn the file into a database with integrity controls.

In occupational safety and health (OSH), Excel is a general supporting tool. It can organise records and produce understandable analyses, but it does not automatically supply preventive judgement, reliable observations or approval of a decision.

Reference documentation: Microsoft: Microsoft Excel · Microsoft Support: Overview of formulas in Excel · Microsoft Support: Difference between Microsoft 365 and Office 2024.

Origins, development and major stages

Microsoft released Excel for Macintosh in September 1985. The Windows version followed in 1987, and an OS/2 edition also existed in 1989. It therefore did not begin as a Windows-only application. Its development encompasses changes in interface, capacity, programming and connections to other sources.

• 1993: VBA. Microsoft dates the introduction of Visual Basic for Applications in Excel to that year. Automation gained a language for building procedures and solutions around the application, distinct from the XLM macros of earlier generations.
• Excel 2007: interface, formats and scale. The ribbon reorganised commands. The grid expanded from Excel 97–2003’s 65,536 rows and 256 columns to 1,048,576 rows and 16,384 columns. Modern XLSX and XLSM formats separated workbooks without VBA code from macro-enabled ones. Saving to an older format could lose data or features.
• Excel 2010: broader analysis. Microsoft documented in-cell sparklines and PowerPivot, associated with the SQL Server 2008 R2 generation, in 2010. This stage expanded the possibilities for representing series and working with analytical models beyond a flat table.
• 2015: Power Query integrated into Excel 2016. The September 2015 announcement described the move from an add-in for Excel 2010/2013 to the native Get & Transform feature. Importing and preparing data became an integrated experience on the Data tab.
• 2018: dynamic arrays announced. In September, Microsoft announced the new way for one formula to return multiple values. The change also introduced compatibility considerations for formulas and workbooks created before that capability.
• 2023: AI and Python. Microsoft announced Microsoft 365 Copilot, including Excel, in March, initially in limited testing. It introduced the public preview of Python in Excel in August. These are announcement milestones, not dates of universal availability for every account.
• Excel 2024 and Microsoft 365. Excel 2024 brings together text and array functions, LAMBDA, and improvements to images, charts and accessibility. The one-time-purchase edition coexists with Microsoft 365’s evolving subscription feature set.

Version comparisons should focus on the capabilities a workbook needs. Product names, internal version numbers and update channels are not interchangeable. Nor should Windows, Mac, web and mobile be assumed to receive every addition simultaneously.

Reference documentation: Microsoft Learn: The History of Microsoft, 1985 · Microsoft Learn: The History of Microsoft, 1987 · Microsoft Learn: The History of Microsoft, 1989 · Microsoft Source (1996): Availability of Visual Basic, Applications Edition · Microsoft Support: Working with Excel 4.0 macros · Microsoft Support: Worksheet compatibility issues · Microsoft MSDN Magazine (2007): Using RibbonX from Visual Basic · Microsoft Support: File formats that are supported in Excel · Microsoft Source: Excel 2010 product archive · Microsoft Blog (2015): Integrating Power Query technology in Excel 2016 · Microsoft Support: Excel functions that return ranges or arrays · Microsoft Source (2023): Introducing Microsoft 365 Copilot · Microsoft 365 Blog (2023): Introducing Python in Excel · Microsoft Support: What is new in Excel 2024 for Windows and Mac · Microsoft Support: Difference between Microsoft 365 and Office 2024.

How it works: data, formulas and workbook structure

A formula combines references, operators, constants and functions. Relative references move when copied; an absolute reference such as $B$1 keeps its position. Before extending a calculation across hundreds of rows, check which parts should change. Visible syntax, function names and separators can depend on language and regional settings: a formula should be tested in the environment where it will be used.

Excel tables provide headers, filters and structured references. A useful design separates original inputs, transformations, parameters and results. For example, an action table contains one row per action and a stable identifier, while the report’s cutoff date is stored as a visible parameter. This makes it possible to understand what each row represents and reconstruct the calculation.

Recalculating formulas and refreshing external data are different processes. The former evaluates dependencies within the workbook; the latter queries a source again. A result can be correctly recalculated from outdated data. To avoid ambiguity, the report should show the date of the underlying data, its refresh date and the period being analysed.

Reference documentation: Microsoft Support: Overview of formulas in Excel · Microsoft Support: Overview of Excel tables · Microsoft Support: Change formula recalculation, iteration, or precision in Excel · Microsoft Support: About Power Query in Excel.

Functions, PivotTables, charts and data models

Functions can add, count, look up values, transform text and work with dates. Dynamic arrays can return several results from one formula, occupying neighbouring cells. If that area is blocked, the result cannot spill into it. Spilled formulas sit outside tables, although they can use table columns as inputs. LAMBDA can define reusable functions within a workbook in compatible versions.

A PivotTable groups records and calculates summaries without changing the source. Selecting sum, count or average changes the question being answered. If a numeric column contains text, the summary may count records when the intention was to add amounts. Refresh the PivotTable when its input changes and inspect the detail behind each total.

The Data Model relates tables; Power Pivot adds tools for working with that model and its calculations. This differs from copying every column into one worksheet. An action table, for example, can relate to a site directory if its identifiers are consistent. Charts turn results into visual representations; scale, filters and units remain the author’s decisions. A truncated axis or a sum of percentages can create a misleading impression even when the chart renders correctly.

Scenarios, Data Tables and Goal Seek support exploration of assumptions. Solver asks a different question: which variable values optimise a target cell while satisfying constraints? The Analysis ToolPak adds statistical procedures. These features do not select a valid model by themselves: an omitted constraint can make a mathematical solution infeasible. In preventive planning, an estimated saving does not justify removing a necessary measure.

Chart choice follows the question: columns compare categories, lines show change, and scatter plots relate two numeric variables. Histograms show distributions across intervals. Before interpreting a chart, establish what each point represents, how missing values were handled and whether the units are comparable.

LAMBDA can also express recursive functions. Microsoft Research describes the extended formula language as Turing-complete: a theoretical property of what it can express, not a promise of unlimited resources, speed or correctness in every workbook.

Reference documentation: Microsoft Support: Dynamic array formulas and spilled array behavior · Microsoft Support: What is new in Excel 2024 for Windows and Mac · Microsoft Support: Create a PivotTable to analyze worksheet data · Microsoft Support: Create a Data Model in Excel · Microsoft Support: Introduction to What-If Analysis · Microsoft Support: Use the Analysis ToolPak · Microsoft Support: Define and solve a problem by using Solver · Microsoft Support: Available chart types in Office · Microsoft Research: LAMBDA, the ultimate Excel worksheet function.

Power Query, external sources and data exchange

Power Query arranges data import and preparation as a sequence of steps that can run again. It works with files and other sources, transforming columns and combining datasets. Connectors and their capabilities vary by edition and platform; a workbook prepared on one computer should be checked in the environment responsible for refreshing it.

A query’s input should retain its provenance. When combining monthly files, include the filename, period and a record key. Before joining two tables, establish the expected relationship: multiple matches can multiply rows and inflate a total without any additional real transactions. Validation means reconciling records, keys and quantities before and after each significant transformation.

A connection does not guarantee permanent access, automatic refresh or a stable schema. A change to a column name, credentials or data type can affect the process. An exceptions table is therefore useful alongside the output: rejected data, unknown identifiers or dates that could not be interpreted.

Reference documentation: Microsoft Support: About Power Query in Excel · Microsoft Support: Create a Data Model in Excel · Microsoft Learn: What is Power Query?.

File formats, export and preservation

• XLSX: a modern XML-based format that does not store VBA code.
• XLSM: a macro-enabled workbook; executable content needs specific review.
• XLSB: a binary workbook format, distinct from the older XLS.
• XLS, XML Spreadsheet, template and add-in formats: serve different generations and purposes; their extensions are not interchangeable.
• CSV and ODS: support information exchange, with possible losses or differences from the original workbook.

A CSV represents delimited data, not an entire workbook with all its sheets, formulas, permissions and presentation. Its delimiter, encoding and date handling need agreement with the receiving system. Exporting the visible result is not equivalent to delivering a reproducible model. To preserve an analysis, retain the workbook, an identified copy of the inputs and an explanation of the parameters used.

An interoperability check means reopening the exported file in the destination application and comparing a sample of identifiers, dates, decimals and totals. Review linked-file paths and the names used by automations as well. Moving a folder can leave an apparently correct workbook connected to a source that is no longer refreshed.

Saving limits matter too: Microsoft documents problems when the complete path and filename exceed 218 characters in the Excel cases described. The limit is not simply the visible filename length. When distributing a workbook, check that it opens and saves successfully in its destination location.

Reference documentation: Microsoft Support: File formats that are supported in Excel · Microsoft Learn: Troubleshoot errors when saving Excel workbooks.

Automation: VBA, Office Scripts, Python and Copilot

VBA automates desktop Excel tasks and works with the application’s object model. Office Scripts supports automation in compatible environments and can integrate with Power Automate. Permissions, execution and feature coverage differ: uploading a file to the cloud does not make a desktop macro run in a browser. Add-ins extend Excel, but web add-ins and desktop components also have different scopes.

Python in Excel runs code in an isolated Microsoft Cloud container and can return results to the workbook. It is not equivalent to a local Python environment with unrestricted access to the computer or network. This architecture matters when deciding which data to send, which libraries are required and what dependencies the analysis will have.

Copilot adds AI assistance for creation and analysis, subject to available access. A suggested formula or generated summary is an output that still requires checking. Acceptance criteria can be specific: compare against known examples, examine excluded rows and explain the rules before applying them to real data. Automation reduces repetition; it does not determine which preventive decision is correct.

Legacy files may also contain Excel 4.0 or XLM macros; Microsoft recommends migrating them to VBA. Desktop automation can interact with other applications through technologies such as COM and OLE. Inventory these dependencies before moving a workbook to Mac or a browser, because retaining the file extension is not sufficient.

The Macro Recorder turns a sequence of actions into code that can be reviewed in the Visual Basic Editor. It provides a starting point for automating formatting or data preparation; it does not interpret the user’s intention. Before replaying it, check ranges, references and effects on a test copy. A custom function and a procedure that changes cells serve different purposes.

Reference documentation: Microsoft Learn: Differences between Office Scripts and VBA macros · Microsoft Support: Data security and Python in Excel · Microsoft: Microsoft Excel · Microsoft Support: Working with Excel 4.0 macros · Microsoft Support: Get started with Copilot in Excel · Microsoft Support: Automate tasks with the Macro Recorder.

Editions, platforms, licensing and collaboration

Excel is offered on Windows, macOS, the web, iOS and Android. Microsoft 365 uses subscriptions; Office 2024 provides a one-time purchase and does not automatically include new features from subsequent editions. The web edition supports free use with a Microsoft account, while advanced and enterprise capabilities depend on the product purchased. Excel is commercial Microsoft software, not an open-source project.

The desktop application can work with local files without a permanent connection, although activation, updates, external sources and connected services have their respective requirements. The old Excel Viewer has been retired. Reviewing a file calls for a current application that supports the features it contains.

Collaborating in a shared workbook can avoid multiple emailed copies, but it requires editing and access rules. Before adopting a workflow across Windows, Mac, web and mobile, test the actual workbook: formulas, queries, macros, add-ins, charts and permissions. Successfully opening a file does not prove that every platform can refresh it or execute its content in the same way.

Reference documentation: Microsoft Support: Difference between Microsoft 365 and Office 2024 · Microsoft: Microsoft Excel · Microsoft Support: Supported versions of the Office viewers · Microsoft Learn: Differences between Office Scripts and VBA macros · Microsoft Support: Collaborate on Excel workbooks at the same time · Microsoft Support: Differences between using a workbook in the browser and in Excel.

Size limits, precision and statistical reliability

The current grid supports 1,048,576 rows and 16,384 columns per worksheet; a cell can contain up to 32,767 characters. Practical capacity also depends on memory, formulas, connections and complexity. These maximums are not design recommendations. A workbook that refreshes too slowly or is understood only by its creator may become unsuitable well before reaching a physical limit.

Excel works with numeric precision of 15 significant digits. Long identifiers should remain text when every digit matters, since they are not quantities to calculate. Manual calculation mode can leave results pending. The precision-as-displayed setting changes stored values and should not be confused with applying a display format.

Research published by McCullough and Heiser in 2008 identified accuracy problems in statistical procedures in Excel 2007. This is evidence about that version and those tests, not an assessment of current Excel. For sensitive analysis, verify the procedure using reference datasets, explicit assumptions and independent results. A long string of decimal places does not establish accuracy, an appropriate method or a good sample.

A separate historical case was Excel 2007’s display bug: certain calculations near 65,535 could appear as 100,000. Chris Lomont’s original analysis and the October 2007 KB 943075 fix distinguish display from calculation. This is a documented, corrected historical incident, not a defect attributed to the current version.

Some limits are function-specific. Microsoft documents VLOOKUP problems with lookup values longer than 255 characters. An old MOD note described errors with certain large quotients; it does not establish their persistence across all current versions. Test the actual range of inputs, alongside small examples that produce correct results.

Reference documentation: Microsoft Support: Excel specifications and limits · Microsoft Support: Change formula recalculation, iteration, or precision in Excel · McCullough y Heiser (2008): On the accuracy of statistical procedures in Microsoft Excel 2007 · Chris Lomont: Original analysis of the Excel 2007 formatting bug · Microsoft KB 943075, conserved in the unofficial PKI Solutions archive · Microsoft KB 119083, conserved in the unofficial PKI Solutions archive · Microsoft Support: Correct a VALUE error in VLOOKUP.

Dates, automatic conversions and identifiers

Excel uses date systems based on 1900 or 1904, separated by 1,462 days. When reusing historical workbooks, check both the source and destination systems. For historical compatibility, the 1900 system also retains a nonexistent leap day in February of that year. This affects certain historical calculations; it does not mean that all leap years are calculated incorrectly.

Another risk is an identifier being converted into a date, number or scientific notation. Microsoft documents controls for several automatic conversions, but their effect should be checked in the actual import process. The 2016 study by Ziemann, Eren and El-Osta documented gene-name conversion errors in scientific material. It illustrates an error mechanism rather than measuring the quality of all Excel files.

In OSH, the same control principle is useful for site codes, equipment identifiers and document references: define them as text, retain leading zeros and compare a sample with the source. For dates, specify the format and time zone where relevant; a string such as 03/04 may be interpreted differently under different regional settings.

Dates earlier than the supported system’s starting point require an explicit representation and method; typing an old date as text does not make it a native date suitable for subtracting days. This matters when reconstructing historical series or importing files from other systems.

Reference documentation: Microsoft Support: Date systems in Excel · Microsoft Learn: Excel incorrectly assumes that 1900 is a leap year · Microsoft Support: Data import and analysis options in Excel · Ziemann, Eren y El-Osta (2016): Gene name errors are widespread in the scientific literature · Microsoft Support: Excel specifications and limits.

Protection, privacy and change control

Microsoft distinguishes worksheet protection from file security: locking cells prevents certain changes but does not replace access control. A worksheet password, read permission and file encryption address different problems. Macros and add-ins need particular attention because they can execute actions with a different scope from an ordinary formula.

In a preventive record, separate personal detail from the aggregated report. Before sharing, review hidden sheets, comments, connections and data retained inside the workbook; hiding a column does not delete it. Recipients should receive only the level of detail their work requires.

Data validation can restrict direct entries, but copying or filling cells may bypass some warnings. A dropdown list therefore needs checks on the received dataset as well. Identify the model owner, approved version, data origins and significant changes. These proposed controls help make the work reviewable; they do not turn the file into a compliance guarantee.

Reference documentation: Microsoft Support: Protect a worksheet · Microsoft Learn: Differences between Office Scripts and VBA macros · Microsoft Support: More on data validation.

General uses, community and critical assessment

Excel lends itself to budgeting, transaction reconciliation, scenario comparison and data exploration because it brings input, calculation and presentation into one environment. That proximity supports rapid iteration; it can also hide an implicit decision inside a formula. A good spreadsheet explains its assumptions as carefully as it presents its results.

The community also uses it for competitions, games, art and other creative work. Microsoft documents the Excel World Championship in its 2025 retrospective. These uses demonstrate flexibility, although a corporate history is not an independent assessment of accuracy, security or superiority.

Research on data conversions and statistical accuracy offers another perspective: a widely used tool still needs verification. Distinguish defects in a particular version, limitations inherent in a format and errors in a user-built model. Each calls for a different response.

Reference documentation: Microsoft Signal: Microsoft Excel celebrates 40 years · Ziemann, Eren y El-Osta (2016): Gene name errors are widespread in the scientific literature · McCullough y Heiser (2008): On the accuracy of statistical procedures in Microsoft Excel 2007.

Reasoned applications in occupational safety and health

The following uses are proposed workflows from Sabentis, not automatic prevention features or certified templates. Their value depends on record definitions, interpretation criteria and professional review.

• Action tracking: bringing together an identifier, site, owner, deadline, status and evidence makes delays visible. Closure should depend on an agreed check; changing a cell’s colour does not demonstrate that a measure has been implemented or is effective.
• Observation analysis: grouping findings by activity and context helps formulate questions. An increase may reflect a participation campaign, a classification change or greater exposure, rather than necessarily indicating deterioration.
• Training planning: comparing task-specific needs with training records can reveal gaps. Recorded attendance does not itself demonstrate competence or authorisation to perform a task.
• Indicator preparation: organising numerators, denominators, periods and sources supports reproducible comparison. People, hours, tasks and events must be distinguished; adding them without a shared definition produces ambiguous figures.

A simple example shows why the denominator matters: if two sites record four events each, their counts match, but interpretation changes if one has twice as many hours of activity. Even an adjusted rate does not explain severity, context or underreporting by itself. A report should present the figure, its limitations and the questions requiring investigation.

To make the work transferable, the workbook can include a field dictionary, a parameters sheet and an exceptions list. Another person can then distinguish a confirmed zero from a missing value and see which records were excluded from the analysis.

Worked case: outstanding actions across three sites

Hypothetical example, with no client data or measured outcomes. Three sites send files containing preventive actions. The objective is a report of outstanding work at a specified cutoff date, with detail that makes its totals checkable.

1. Define the input: one row per action, a unique key incorporating the site and a shared status catalogue. Preserve received files unchanged. The process owner agrees what evidence allows an action to be considered closed.
2. Prepare the data: import through Power Query, preserve the source filename and set identifier and date types. Separate records with a missing owner or deadline, or an unknown status. Do not fill these gaps through assumptions.
3. Resolve duplicates: if an action appears twice, use a documented rule to determine the valid version. Do not delete rows merely because their descriptions look similar. Compare record counts before and after every join.
4. Calculate: an action is overdue when its deadline precedes the cutoff date and it does not meet the closure criterion. Rows without deadlines remain exceptions. Open status and overdue status are different variables.
5. Test: manually check a past action, a future one, one closed with evidence, one with no date and a duplicate. Also check what happens on the due date itself, according to the agreed rule.
6. Deliver: provide the detailed list, site summary and exceptions, with the refresh date and person responsible for review. Retain the parameters needed to reproduce the report.

For an acceptance check, suppose 30 rows are received and two confirmed duplicates are removed: 28 unique actions should remain. If 20 are open and eight closed, the groups must total 28. If seven open actions are overdue and three have no date, those three must not be counted as on schedule. These are test figures for the example, not performance attributed to Excel or Sabentis.

The useful output is an explainable list for action and review, accompanied by its exceptions. Validation is complete when any row can be traced to its source and every total justified.

Reference documentation: Microsoft Support: About Power Query in Excel.

When to choose Excel and when to complement it

Excel suits bounded analyses, reconciliations and flexible models with an identified owner. Consider volume, refresh frequency, data sensitivity and the number of editors. A file with many diverging copies or unmaintained rules calls for a process review even if its formulas still work.

Power BI can complement the periodic distribution of indicators through a shared model. Google Sheets merits comparison when work centres on Google Workspace and shared editing. Evaluate a representative example, checking calculation, collaboration, permissions and file exchange.

Records that require persistent relationships and controls may be better held in a database or management application, with Excel retained for analysis. The deciding questions include who controls the data and how a modification is checked, alongside who knows the interface.

Frequently asked questions about Excel and OSH

Can an Excel workbook be a risk assessment?

It can be a supporting record. The method, data, observation of work and professional judgement determine assessment quality; the file extension does not validate it.

Does refreshing a query guarantee a correct report?

No. It repeats the configured steps. Review structural changes, rejected records and consistency with the source, and check that the summaries have also refreshed.

Is protecting formula cells enough?

It helps prevent accidental changes, but it does not replace file permissions, access reviews or result checks.

What should accompany an OSH indicator?

Its definition, period, numerator, denominator, source, refresh date, exclusions and interpretive limitations. This is the minimum information needed to reproduce it and discuss its meaning.

Are macros or AI necessary for good analysis?

No. A clear table and checked rules may be sufficient. Automation makes sense when it addresses a specific repetition and its inputs, changes and outputs can be verified.

Reference documentation: Microsoft Support: Protect a worksheet.