Excel Power Query Performance Tips That Actually Help

How-to basics · Updated November 21, 2026

To improve Excel Power Query performance, filter rows and remove unneeded columns as early as practical, let supported data sources do work through query folding, and avoid costly transformations until they are necessary. Then load only the results you need and measure refresh times to find the actual bottleneck rather than guessing.

Understand what is making a query slow

A slow refresh can come from different places: reading a large source, downloading data over a network, applying transformations, merging tables, loading results into a worksheet, or recalculating formulas that depend on the output. A query that takes a long time to preview may also behave differently during a full refresh because the preview and final load can process different amounts of data.

Start by noting which query is slow and whether the delay occurs before data appears, during transformation, or while loading. In Excel for Microsoft 365 on Windows, open Data > Queries & Connections to see the queries and their refresh status. Refresh one query at a time when investigating, rather than using Refresh All, so you can distinguish its cost from other queries and workbook calculations.

Keep a simple baseline: record the approximate refresh time, source size, and output row count before making changes. Change one thing at a time and refresh again. A faster preview alone does not prove that the final refresh or workbook is faster.

Filter rows close to the source

Filtering early is often the most useful improvement because every later step may otherwise process rows that will eventually be discarded. If a report only needs the current year, filter out older records near the start of the query. If you need only active customers, apply that condition before an expensive merge or grouping step.

Where the connector supports query folding, Power Query can translate compatible steps into a request that the source system executes. For example, a SQL database may return only the requested date range instead of sending the entire table to Excel. Excel workbooks and ordinary text files generally cannot perform this kind of server-side work, so filtering still reduces later processing but may not reduce the initial file read.

  • Prefer source filters that can be expressed simply, such as a date range or a status equals a specific value.
  • For database sources, check whether the connector and transformation steps preserve folding. The available indicators and options vary by connector and Excel build.
  • Do not assume that a missing View Native Query option proves that no work is folding; that option is not available in every situation.

When the source is a database, ask whether a database view or a source-side query can deliver the same filtered and shaped data. That can be more efficient than repeatedly importing a large raw table, provided the view is maintained appropriately.

Keep only the columns and data you need

Remove unused columns early, especially wide text fields, binary content, or columns with complex values. Fewer columns mean less data to transfer, hold in memory, transform, and load. This is particularly helpful when combining many files, where an unnecessary column is repeated for every record.

Use a defined Excel table or a specific worksheet range as the source when possible, instead of importing an entire workbook area that includes unrelated data. In a CSV or text file, confirm that the delimiter, encoding, and header row are interpreted correctly. A malformed source can produce extra columns or text values that require extra cleanup and make later steps harder to reason about.

Assign appropriate data types deliberately. Correct types are essential for reliable date filters, numeric calculations, and joins. But converting every column repeatedly is unnecessary; remove irrelevant fields first and avoid redundant type-conversion steps. A date stored as text may need conversion before a date comparison works as intended, while a column used only as a label may not need elaborate parsing.

Make merges and groupings less expensive

Merges can become slow or produce unexpectedly large outputs when the input tables are much larger than needed or the join keys are inconsistent. Before merging, filter each input to the required rows and remove columns that will not be used. Check that corresponding keys have compatible data types and that values are normalized—for example, that one source does not contain trailing spaces or a mixture of numbers and text.

Check the uniqueness of lookup keys before joining. If the lookup table has several rows for the same key, a merge can multiply matching records. That may be correct for a one-to-many relationship, but it can also inflate the result and make a refresh appear slow. Validate row counts before and after the merge so that an unintended increase is visible.

For grouping, aggregate only the columns and measures required for the report. If you need a total by month and region, reduce the data to those dimensions and the value being summed; do not carry along detailed fields that the output will not use. When a database connector can fold the merge or aggregation, keeping the steps compatible with that connector may let the source do more of the work.

Be selective with expensive transformations

Operations such as sorting a large table, splitting or parsing complex text, adding custom calculations, and repeatedly invoking a custom function can consume substantial time. Apply them only when the result requires them. For example, sorting is usually unnecessary if the final consumer does not require a particular order. If it is required, sort as late as practical, after reducing rows and columns.

Some transformations can prevent query folding for a particular source. A non-folding step does not automatically make a query bad, but it can mean that Power Query must retrieve more data and perform the work locally. If performance changes sharply after a particular step, inspect that step and test whether a simpler equivalent operation performs better. Keep the logic readable; a compact but opaque query is harder to maintain and diagnose.

A common misconception is that Table.Buffer is a general-purpose speed switch. Buffering can force a table to be materialized in memory, which may help in specific cases where the same small result is reused, but it can also increase memory use and stop folding. Do not add it by default. Consider it only after measurement and with a clear reason.

Choose an appropriate load destination

Loading a query to a worksheet adds work and has practical limits: a worksheet can contain at most 1,048,576 rows. If the result is too large or is used only as an intermediate input for another query, avoid loading it as a visible sheet table. In the Load To options, choose a connection-only result for staging queries, or load the final result to the Data Model when that is suitable for analysis in PivotTables and related features.

Connection-only staging can keep a workbook uncluttered, but it does not guarantee that a source will be read only once. Power Query may evaluate referenced queries separately during refresh, depending on the query structure and source. Avoid creating many near-identical imports of the same large source without checking whether they are necessary.

Excel for Windows offers the broadest Power Query authoring and connector support. Excel for Mac has Power Query capabilities, but available connectors and features can differ by version and update channel; a Windows menu path or connector may not be present on Mac. Excel for the web also has a different set of authoring and refresh capabilities. If a workflow depends on a particular connector or transformation, verify support in the version where the workbook will be maintained.

Diagnose refreshes instead of guessing

Use a repeatable test to locate expensive steps. In Power Query Editor, refresh the preview after selecting a step and compare how long it takes to reach that point. Preview results are useful clues, not a perfect measure of a full refresh, so confirm improvements by refreshing the query and observing the completed load in Excel.

Where available in the Excel Power Query Editor, Query Diagnostics can capture information about evaluations and operations. The exact commands and diagnostic detail can vary with Excel version and connector. Diagnostics may generate additional queries or data, so use them for investigation and remove diagnostic artifacts when finished. For a database connection, source-side monitoring or a database administrator may be able to identify whether the wait is in the query, network, or server.

  • Compare a refresh with and without a suspected merge, sort, or custom transformation.
  • Check whether output row counts are unexpectedly high or whether duplicate keys expand a join.
  • Separate source time from load time where possible, and check whether workbook formulas or other queries are also refreshing.
  • Test with a representative sample, but verify the final result against the full dataset before relying on it.

Common performance traps

Repeatedly expanding all columns from a nested table is a frequent source of oversized results. Expand only the fields needed, and check whether the nested records contain duplicates that change the row count. Similarly, combining files from a folder can be slow if the folder contains old exports, temporary files, or unrelated formats. Narrow the folder or filter the file list before combining, and make sure the sample-file transformation matches the actual files.

Another trap is refreshing everything when only one output needs updating. Use the relevant query’s refresh command while troubleshooting, and review connection properties if background refresh makes it difficult to tell when a refresh has completed. Background refresh can be convenient, but it is not inherently faster; it changes when Excel returns control and may make timings harder to interpret.

Finally, privacy settings are not merely a performance toggle. Power Query uses privacy levels to help prevent unintended data from one source being sent to another. Changing or ignoring privacy behavior to make a refresh proceed faster can weaken that protection. Set privacy levels appropriately for the sources and organizational rules, and consult an administrator when the privacy firewall blocks a legitimate combination.

Frequently Asked Questions

Does filtering early always make Power Query faster?
It usually reduces the work required by later steps. It may also reduce data transferred when query folding is supported, but filtering a local Excel or CSV source may not avoid reading the source file in the first place.

Why is my Power Query preview fast but the full refresh slow?
A preview may process only a limited view of the data, while a full refresh must evaluate and load the complete result. Loading to a worksheet, refreshing dependent queries, and recalculating formulas can also add time.

Should I use Table.Buffer to speed up a query?
Not automatically. Buffering can use more memory and may prevent folding. Test it only when you have a specific reuse or evaluation problem and can compare full-refresh results before and after.

Does connection-only loading make a query refresh faster?
It avoids loading that query’s result to a worksheet or Data Model, so it can reduce load work. The query still has to be evaluated when needed, and referenced queries are not guaranteed to share a single cached evaluation.

Can Power Query use multiple processor cores?
Some operations and connectors may use parallel processing, but behavior depends on the operation, source, and Excel environment. More cores do not fix slow network access, non-folding steps, excessive output, or an inefficient join.

Is Power Query faster in 64-bit Excel?
64-bit Excel can address more memory than 32-bit Excel, which can help with large workbooks that are constrained by memory. It is not a guaranteed speed increase, and source design, query steps, and available system resources still matter.

Related

Try it: the Date Difference Calculator: days, months, years and weekdays between two dates, with the DATEDIF formulas.

More in How-to basics

Sources: Function behaviour and menu paths: Microsoft Support documentation for Excel. Method & sources

Spotted an error in this guide? Let us know.