Data Validation: Techniques, Tools, and Where the Check Belongs

Data validation is checking that a value meets a defined rule before it is accepted: the right type, the right format, within range, and drawn from the permitted set. The techniques are well understood and largely settled.

The question that is not settled is where the check runs. Nearly every treatment of validation places it in the pipeline, after the data has been created and is on its way somewhere. For data that people type, that placement is the difference between preventing a problem and documenting one.

What is data validation?

Checking a value against a defined rule before accepting it.

Two words in that sentence carry the weight. Rule: validation requires a prior decision about what is acceptable, and you cannot validate against a standard that does not exist. Before: a check that runs after acceptance is a report, not a validation.

The discipline is treated in the data-management literature as a quality control applied to data entering a system (IBM, “What Is Data Validation?”, accessed 2026-09-11). That framing is correct and carries an assumption worth surfacing: that there is a system boundary the data crosses, and that crossing it is the moment to check. Sometimes there is. For campaign metadata, the value was created by a person long before it crossed anything.

The four types

Type, format, range and consistency.

This is the standard taxonomy and it is stable across sources (Wikipedia, “Data validation”, accessed 2026-09-11).

Worth noticing what none of the four catches: a value that is the right type, correctly formatted, in range, internally consistent, and simply not the value that was meant. channel = display on a paid social placement passes every check in the table.

The techniques

Type, format, range, consistency, uniqueness and lookup against a reference list.

The last row is the one that matters for marketing data, and it is the only technique in the set that depends on a decision rather than on logic. Type, format, range and consistency can all be derived from the data’s shape. Set membership requires someone to have decided the set. Which is why validation and data standards are the same project viewed from two ends.

Testing vs enforcing

A test tells you how many values failed; enforcement means they could not have been created.

Both are worth having and they are not substitutes. Testing tells you whether your enforcement is working and catches what enters by routes you do not control. But a program that only tests has bought measurement, not quality.

The distinction shows up in how each is reported. A testing program reports a pass rate, and a pass rate that improves is genuinely good news — but it improves by the data team fixing things, which means the improvement stops the moment they stop. An enforcement program reports a rejection count, and a falling rejection count is the good news, because it means people have stopped attempting the wrong value.

Those are opposite-signed metrics measuring the same underlying health, and a team running both will occasionally find them pointing in different directions. When that happens the enforcement number is the more trustworthy one: it counts attempts, not corrections.

Data quality in detail: Read: data quality — the dimensions validation is measuring against.

Tooling categories

Spreadsheet features, pipeline frameworks, quality platforms, and entry-point enforcement.

A word about Excel, since that is what most searches for this phrase mean. Excel’s Data Validation feature restricts what a cell will accept, and Microsoft’s support page documents it fully (Microsoft Support, “Apply data validation to cells”, accessed 2026-09-11). It is genuinely the same idea at cell scale, and it is the right tool for a spreadsheet. It stops being sufficient the moment the values leave the sheet, which for campaign data is immediately.

Where the check belongs

At the point of entry, in the workflow of whoever types the value.

The cost of a failed value is not fixed. It rises steeply with how far the value has travelled.

At entry it costs a correction: the form offers the right options, the person picks one, nothing downstream ever knows. In the pipeline it costs a quarantine and a decision about what the value should have been, made by someone who was not there. At the warehouse it costs a mapping table that somebody maintains permanently. In the report it costs a decision made on a wrong number, and the correction never fully lands because the quarter has closed.

Validating data and enforcing compliance before campaign activation is one of the most common jobs enterprise teams bring us, across 39 accounts, and it is consistently framed pre-launch rather than post-hoc for exactly this reason. Manual entry and copy-paste workflow causing errors is raised across 66 accounts, the same problem stated from the other side.

The standards layer: Read: what are data standards? — the permitted sets a lookup check validates against.

Validating data you did not create

When an agency builds the campaign, the check has to live in their form.

This is the case that decides the architecture, and it is the one the published treatments never reach.

A meaningful share of campaign metadata is created outside your organization, in systems you do not administer, by people who have never seen your standards. Every validation category above except the last one runs inside your perimeter. An agency’s placement is validated, if at all, weeks after it is built, by a job that can flag the value but cannot ask what was meant.

Bristol Myers Squibb’s digital media operations team described the state before the check moved.

“There was no quality control, security, automation, or scalability, coupled with a wide margin of error,” — Tim Scales, Digital Media Operations Consultant, Bristol Myers Squibb

The four things named there are not separate problems. They are one problem: values created without a check, at a volume no one can review by hand. Moving the check to the point of creation addresses all four at once, and it is the only placement that reaches somebody who does not work for you.

Which makes the prerequisite unavoidable. A check at the point of entry needs a permitted set to check against, owned by someone who can approve an addition — the data dictionary and the standard behind it. Validation is the enforcement half of that work, not a substitute for it. Where a value is preserved unchanged but was wrong to begin with, that is a data integrity question with the same root.

Frequently asked questions

Sources

Outbound citations, named and dated:

• IBM, “What Is Data Validation?” (accessed 2026-09-11) — the data-management definition and its system-boundary framing.
• Wikipedia, “Data validation” (accessed 2026-09-11) — the four-type taxonomy.
• Microsoft Support, “Apply data validation to cells” (accessed 2026-09-11) — the spreadsheet feature, named and set aside.