Deletion, formulas, one-time scrubbers, and local pseudonymization on the same criteria
A spreadsheet is ready to share, except for the personal data still in it. Most teams handle that with one of four methods: deleting columns and running find & replace, building formulas or Power Query steps, running the file through a one-time PII scrubber, or using browser-local pseudonymization with a persistent mapping. Each of the four is the right tool in some situations and a poor fit in others, so this guide holds all of them to the same criteria rather than ranking them.
The comparison criteria
Nine criteria recur in every section below:
- Cost — license or subscription fees, plus your time.
- Setup time — the work required before the first clean file comes out.
- Fit for recurring work — what the second, fifth, and twentieth file cost you.
- Free-text detection — whether values inside notes and message columns are found. No tool finds all of them; the detection benchmark publishes what one pattern layer actually scores, format by format.
- Consistent aliases — whether the same person receives the same replacement everywhere.
- Restorability — whether a replaced value can be traced back to the original later.
- Structure preservation — whether rows, columns, and repeat patterns survive the cleaning.
- Whether the file is transmitted — whether the content leaves your machine during processing.
- Need for manual review — how much checking remains after the method has run.
Method 1: Delete columns and find & replace
Delete the name and phone columns outright, or swap known values with Ctrl+H. Cost is zero, setup takes minutes, and for a one-off file where you can point at every sensitive column, this is the right choice. Work on a copy so the original survives a slip of the keyboard.
The criteria show where it strains. Free-text detection: a phone number inside a notes column is only replaced if you search for that exact value, so anything you did not anticipate stays in. Alias consistency depends on your own discipline; replace the same customer with different placeholders in different rows and the fact that one person contacted you three times disappears from the data. Deleted values are also unrestorable, which is why the working copy matters. In exchange, nothing is transmitted anywhere, and the structure you keep is exactly the structure you chose to keep. Manual review carries most of the weight here, since your eyes are the detection engine.
Method 2: Formulas and Power Query
SUBSTITUTE and LEFT&REPT formulas, or a set of Power Query transformation steps, encode your cleaning rules once so they can run again. The setup cost lands up front: expect hours of building and testing rather than minutes. That investment pays off when files of the same shape arrive on a schedule — a weekly export with fixed columns is the situation where this method is a genuinely good choice, and teams already fluent in Excel often need nothing else.
Consistent aliases are achievable with a lookup sheet that maps each original value to its code, and the same sheet gives you restorability for as long as you keep it. It is also, itself, a file of personal data, so it needs the same protection as the export it de-identifies. Free-text detection is the harder gap: a formula replaces the patterns you anticipated, and a phone number written with unusual spacing inside a memo column passes through unless a rule covers that shape. Renamed columns and layout changes break the rules, so budget for maintenance. Nothing is transmitted if you work in a desktop spreadsheet, and structure is fully preserved.
Method 3: One-time PII scrubbers
Search for "remove PII from CSV" and you will find a long list of tools, many free. Tools in this category detect and replace sensitive values for a single processing run. Detection quality across the category has improved measurably — by April 2026, OpenAI had released Privacy Filter, an open-source (Apache-2.0) PII masking model that runs in a browser, with OpenAI itself noting it is an aid, not a safety guarantee. For a one-off file with messy text columns, where no mapping needs to outlive the session, a scrubber can be the fastest adequate answer: no setup, no formula maintenance, and pattern detection that reaches inside free text.
Before using any one-time scrubber, confirm where processing happens — some run in the browser, others on a server. How browser-local file processing works—and how to verify it describes an offline test and a Network-tab check that answer this for whichever tool you are evaluating. If a tool does process files on its server, the original you set out to clean travels to a third party before any protection is applied, and depending on your jurisdiction, handing personal data to a third-party processor can carry legal obligations of its own.
Many one-time scrubbers keep no mapping between sessions, so next week's export starts from zero and an AI answer about PERSON_001 cannot be traced back. Whether that matters depends on whether your files recur; When one-time PII masking breaks a recurring workflow works through that case.
Method 4: Browser-local pseudonymization with persistent mappings
Tools in this category run detection and transformation inside the browser tab, so the file's content stays on your machine while it is processed, and they keep a mapping between originals and aliases that can persist beyond a single run. The typical workflow combines pattern detection in free text, aliases that stay consistent across the file, a review screen before export, and a mapping that can be exported or stored for cross-file consistency and for restoring originals later. Data Alias is one implementation; if you evaluate it, or any tool in the category, try the workflow with sample data first and confirm that repeat values, column structure, and export warnings behave as expected before processing a real customer file.
The trade-offs sit elsewhere. Purpose-built tools are often paid, where Method 1 is free and Method 2 uses software you already own. Local pseudonymization keeps the file on your machine, and it still inherits the limits of automatic detection: some values are missed, and the mapping becomes a file you are responsible for. See What automatic PII detection misses in spreadsheets before treating any output as clean. Custody of that mapping — where it lives, who can open it, when it is deleted — is the same ongoing responsibility a Method 2 lookup sheet carries.
Side by side
The cells are deliberately terse; the sections above carry the qualifications.
| Criterion | Delete & replace | Formulas / Power Query | One-time PII scrubbers | Local pseudonymization (persistent mapping) |
|---|---|---|---|---|
| Cost | Free | Free (own software) | Often free | Often paid |
| Setup time | Minutes | Hours | Minutes | Minutes |
| Fit for recurring work | Poor (redone by hand) | Good (rules rerun) | Poor (single run) | Good (with vault or exported mapping) |
| Free-text detection | No (manual search only) | Partial (known patterns) | Usually yes | Varies by tool |
| Consistent aliases | Manual discipline | Yes (lookup sheet) | Varies by tool | Yes per file; cross-file only with a persisted mapping |
| Restorability | No | Yes (if sheet kept) | Often no (check) | Usually, if mappings can be exported |
| Structure preservation | Partial (columns lost) | Yes | Varies by tool | Usually; verify the exported file |
| File transmitted | No (desktop app) | No (desktop app) | Varies (verify) | No (verify per tool) |
| Need for manual review | Yes (it is the method) | Yes | Yes | Yes |
Choosing by situation
- One-off file, sensitive values in columns you can point at → Method 1, on a copy.
- Recurring files with an identical shape, a team fluent in Excel → Method 2; protect the lookup sheet like the data it decodes.
- One-off file with messy free-text columns, no need to restore → Method 3, after confirming where it processes.
- Recurring exports feeding AI analysis whose answers come back into real work → Method 2 or Method 4, weighing rule maintenance against tool cost.
When the bullets do not settle it, three questions usually do. Where does the original file go while it is being cleaned? Does the output still support the analysis it was cleaned for? Will the same work recur next month? A method that answers all three for your situation is the right one, whichever number it carries in this guide.