Data Integrity: What It Means, and Where Marketing Data Loses It

Data integrity is the assurance that data stays complete, accurate and consistent across its whole lifetime: from creation, through storage and transfer, to use. It is a property of the data’s history, not of any single value.

That is why it is not the same as data quality. Quality asks whether the value is fit for the decision; integrity asks whether it is still the value that was recorded. A campaign name that was wrong the moment someone typed it can have flawless integrity: unaltered, auditable, faithfully preserved, and wrong in every report it reaches.

What is data integrity?

The word doing the work in that definition is “stays.”

The standard framing covers three commitments: that data is not altered without authorization, that it remains complete as it moves between systems, and that it stays internally consistent, meaning a record’s parts still agree with one another (IBM, “What is data integrity?”, accessed 2026-09-11).

Notice what all three are about: preservation. Integrity is a promise about what happens to a value after it exists. That promise is worth a great deal, and it is not the promise most people think they are buying.

Integrity vs quality

Integrity asks whether the value survived; quality asks whether it was any good.

The last row is the one that matters. Integrity and quality can diverge completely, and when they do, integrity is the one that reports clean.

Harvard Business School Online frames integrity as the foundation analysis rests on (Harvard Business School Online, “What Is Data Integrity and Why Does It Matter?”, accessed 2026-09-11). That is right as far as it goes. A foundation that faithfully preserves a wrong measurement is still a foundation, and the building is still crooked.

Data quality in detail: Read: data quality — the dimensions a value is measured on, and how to measure them.

The five principles

Attributable, legible, contemporaneous, original, accurate.

These five come from regulated environments, where proving a record’s history is a compliance obligation rather than a preference.

• Attributable. You can say who created or changed the record, and when.
• Legible. The record is readable and permanent, including its corrections.
• Contemporaneous. It was recorded when the event happened, not reconstructed afterward.
• Original. The first capture is preserved, not only a transcription of it.
• Accurate. It reflects what actually occurred.

Four of the five are about provenance. Only the last is about correctness, and it is the one no control in the list can enforce on its own.

Three senses of the term

Security, regulated life sciences, and business data management.

“Data integrity” means three different things depending on who says it, and conversations run aground when two of the senses are in the room.

The security sense is the most commonly published one — Fortinet, for example, frames integrity primarily as protection against unauthorized alteration (Fortinet, “What Is Data Integrity? Why Is It Important?”, accessed 2026-09-11). For a marketing team the third sense is usually the operative one, and the first is what their search results will mostly return.

What protects integrity

Constraints, checksums, access control, audit trails and referential rules.

These are well understood and, in most enterprises, genuinely well implemented. Qlik’s treatment of the control set is representative of the standard advice (Qlik, “What is Data Integrity? Why You Need It & Best Practices”, accessed 2026-09-11).

Look at the “Acts” column, though. Every control fires at write or later. Not one of them evaluates whether the value being written was the right value.

What integrity software actually buys you. Tools sold in this space cluster around those same controls: schema enforcement and constraint management, transfer verification, immutable audit logging, access governance, and backup with point-in-time recovery. They are worth having, and for the threats they address they are close to solved problems.

What none of them claims to do is judge correctness. A constraint engine will reject a date in the wrong format and accept a date that is simply not the launch date. An audit log will record who set the channel to Emial and preserve that record impeccably. The software is doing its job; the job is narrower than the phrase “data integrity” suggests to someone outside the data team.

How integrity is measured. Because the controls are binary, so are most integrity metrics: constraint-violation counts, failed-transfer counts, unauthorized-modification events, time-to-recovery. A healthy dashboard here shows zeros. Those zeros are real and worth having — and they say nothing about whether the values underneath are the right ones, which is why an organization can hold a clean integrity posture and an unusable channel report at the same time.

Where marketing data loses integrity

Almost never in transit; almost always at entry, and then faithfully preserved.

This is the case the published references do not cover. All four of the sources above assume the recorded value was right and ask only whether it survived. For campaign data the assumption usually fails, because the value is typed by a person at the moment a campaign is set live.

The controls then work exactly as designed. The wrong channel name is written with a valid type, preserved with a correct checksum, transferred without loss, attributed to a known user, and reported. Integrity is intact end to end. The number is wrong.

Creative trafficking and campaign taxonomy errors causing execution failures is one of the most common problems raised with us, across 40 enterprise accounts, and it almost never presents as an integrity incident, because nothing was corrupted. Approval workflows, audit trails and access control come up separately, across 39 accounts, usually after a team discovers that a complete audit trail tells you who entered the wrong value but not that it was wrong.

The standards layer: Read: what are data standards? — the agreed values a record is checked against at entry.

How to protect it

Protect the pipeline, then move the real control to the point of creation.

1. Keep the standard controls. Constraints, referential rules, checksums, access control, audit trails, backups. Nothing below replaces them; corruption and unauthorized change are real threats.
2. Establish what a correct value is. For each field that feeds reporting, decide the permitted set or the format. A control cannot check against a standard that does not exist.
3. Apply the check where the value is created. In the form, the workflow or the campaign builder, before the record is saved. This is the only point at which a wrong value costs nothing to fix.
4. Extend it past your own perimeter. A large share of campaign data is created by agencies and partners in systems you do not administer. A control that reaches only internal users covers a minority of the records.
5. Reconcile across systems on a schedule. Referential integrity protects joins inside a database. Nothing protects a join between two platforms that never shared a key — that is data validation and cross-system reconciliation, run on a cadence.
6. Audit for correctness, not only for change. An audit that confirms nothing was altered has confirmed integrity. Ask separately whether the values are right.

Step 3 is the one most programs skip, because it sits with the campaign team rather than the data team, and integrity is filed as a data-team concern. That filing is how the gap survives.

The sequencing matters as much as the list. Teams that start at step 3 without doing step 2 end up enforcing a convention nobody ratified, which produces the same arguments as having no standard at all, only louder and with a system to blame. Teams that stop after step 1 have a well-protected record of whatever was typed. The order is deliberate: decide what correct means, then enforce it where it is cheapest to enforce, then verify across the systems that were never designed to agree.

Integrity beyond the database

Brand integrity depends on the same thing: consistent values applied consistently.

The word does duty outside data management too, and the mechanism turns out to be similar. Brand integrity is the promise that the brand means the same thing everywhere it appears — same name, same claims, same standards across markets and agencies.

That promise fails the same way data integrity fails. Not because anyone corrupted anything, but because the value applied at creation was not the agreed one, and everything downstream faithfully preserved it. A brand name entered four ways across three regions has perfect data integrity and no brand integrity at all.

The overlap is more than rhetorical. The same fields carry both: a campaign record’s brand, market, product and channel values are simultaneously the data that reporting groups by and the brand taxonomy that determines how the work is described. Governing them once serves both, and governing them in two places guarantees they diverge.

Which is why the two problems tend to arrive together. A team that cannot reconcile its channel report usually cannot produce a clean list of how its brand was named across last year’s campaigns either, and for exactly the same reason.

Carhartt’s analytics team described what shifts when the underlying data stops being the thing everyone argues about.

“We have shifted the mindset from ‘data is the problem’ to now, ‘data is the solution’ and we are recognized as strategic partners who help drive the business forward with deep insights, solutions, and new ideas,” — Andrew Laycock, Analytics Manager – Direct to Consumer, Carhartt

That shift is the practical test of whether integrity work landed. While the argument is still about whether the numbers can be trusted, the data team is a cost center defending its outputs. Once the values are right at the point they are created, the same team is answering questions instead of relitigating inputs.

Frequently asked questions

Sources

Outbound citations, named and dated:

• IBM, “What is data integrity?” (accessed 2026-09-11) — the three-commitment definition used above.
• Fortinet, “What Is Data Integrity? Why Is It Important?” (accessed 2026-09-11) — the security sense of the term.
• Qlik, “What is Data Integrity? Why You Need It & Best Practices” (accessed 2026-09-11) — the standard control set.
• Harvard Business School Online, “What Is Data Integrity and Why Does It Matter?” (accessed 2026-09-11) — integrity as the foundation for analysis.