A data retention policy is a written rule set that says which categories of business data you keep, how long each category stays, where it lives, and how it gets destroyed when the clock runs out. The condition that trips most organizations is the word "written": an unwritten practice, even a consistent one, is not a policy, because nobody can audit, defend, or hand it to a new hire. The second boundary is legal. In the United States no single federal statute dictates a universal retention schedule, so your timetable is built category by category from the rules that govern your industry and jurisdiction.

Retention Policy Versus Retention Schedule Versus Backup

These three terms get used interchangeably in meetings, and the confusion causes real errors. A retention policy is the governing document: principles, roles, approval authority, and the penalties for noncompliance. The retention schedule is the table inside it that assigns a period to each data category. Backups are copies, not the system of record, and a deletion from the live system that never reaches the backup tier is an unfinished deletion.

The distinction matters most during litigation. If a court issues a preservation order, the obligation attaches to the policy and the schedule, and the routine destruction authorized by your schedule has to stop for the affected categories until the matter closes. Companies that treat backups as exempt often discover the exemption doesn't survive a discovery request.

Microsoft's data lifecycle documentation draws the same line for its own services, describing retention as keeping content for a defined period and deletion as removal at the end of that period or when a user manually removes it. That two-part definition is a useful test for your own document: if you can't point to both the keep rule and the destroy rule, you have half a policy.

Six Data Categories Most Schedules Miss Entirely

Most first-draft schedules cover customer records, financial records, and employee files, then stop. Contractors and auditors keep finding gaps in the same six places, so scan for these before your document goes to legal review.

Category Typical placement Why it gets missed
Email and chat threads Handled by the mail or collaboration platform, not the records team Feels like infrastructure, not records
Shared-drive drafts and versions Version history settings inside the storage tool Nobody owns the version stack
Backups and snapshots IT operations, not compliance Treated as copies rather than records
Deleted-user mailboxes and drives Admin center retention settings The person who owned it has left
Vendor and processor data copies Contract exhibits Assumed to match your own schedule
Logs, telemetry, and device data Engineering or security operations Created by machines, not people

Google Workspace administrators control the deleted-user question directly through Google Vault retention rules, which can hold a departed employee's mail and Drive content for a set period independent of what the user did on the way out. A policy that doesn't name that setting is silently delegating a retention decision to whoever offboards the account.

For each category above, the schedule needs three fields: owner, system of record, and destroy authority. The owner is a named role, not a department. The system of record is the one place the authoritative copy lives, which is how backups stay copies and email threads stop being invisible.

The Tool Setting That Keeps Everything Forever

Here's the failure case that surfaces in almost every data cleanup: a company publishes a clean retention schedule, then discovers the platform was already holding data past every period on it. The mechanism is usually a preservation setting nobody remembered enabling.

In Microsoft 365, a retention policy set to retain content without a deletion action will hold items until the period expires and then leave them in place. The same platform offers a preservation lock, which prevents anyone, including a global administrator, from turning the policy off or making it less restrictive. Microsoft's own documentation states that the lock cannot be removed or overridden by anyone once applied.

That combination is correct for regulated records and catastrophic for everything else. If a well-meaning admin applies a preservation lock to a broad policy covering all mailboxes, your schedule's delete column becomes decorative. Audit the difference between "retain for N years" and "retain for N years, then delete" before you sign off, and check whether any policy in the tenant carries a lock.

Data Retention Policy and GDPR Storage Limitation

Under the EU General Data Protection Regulation, storage limitation is one of the principles in Article 5: personal data may be kept in a form that identifies people for no longer than necessary for the purposes it was collected for. The regulation doesn't publish a timetable, which is exactly why controllers have to write their own and be able to justify each period.

Article 17 adds the erasure right, and Article 30 requires most organizations to maintain records of processing activities, which is where your retention periods become visible to a supervisory authority. The European Commission's page on the regulation confirms the one-month response window for data subject requests, extendable by two further months for complex cases. That clock runs from receipt, not from whenever your team finds the request.

A retention policy that satisfies GDPR has to answer a specific question for every category: what justified this period? "Industry standard" is not an answer. A defensible entry reads more like "applicant records, 6 months after decision, justified by equal-opportunity monitoring under local labor rules," and the justification column is the part auditors read first.

Building the Schedule: Six Steps in Order

Start with the highest-risk category, not the largest one. Getting the method right on a small category first means you can defend the whole document when the big ones follow. The sequence below is the order that survives review.

1: Inventory every system that stores business records, including the ones finance and HR never mention. Use the application list from IT, not a survey. 2: Assign each system an owner by name and title, with a deputy. Owners answer for the data; IT answers for the platform. 3: Map each category to a legal or operational basis, citing the specific statute, contract clause, or business function. 4: Set the period for each entry after the trigger event, and write the trigger down. "Seven years after the fiscal year closes" is a period; "seven years" is not. 5: Record the destroy method per medium: secure deletion for live systems, media destruction for hardware, and a verified deletion path for each backup tier. 6: Draft the exceptions list, covering litigation holds, investigations, tax audits, and regulatory examinations, and name who can invoke each one.

If step one stalls, the cause is almost always shadow storage: personal cloud accounts, team drives outside the sanctioned platform, and departmental file servers. Treat discovery of those systems as a finding to remediate, not a reason to delay the schedule. A schedule covering 80 percent of systems with named gaps beats a perfect draft that never ships.

How Long to Keep Records: A Starting Table

These periods appear across common US and UK compliance frameworks, and they are starting points to confirm against the statute and regulator that govern you, not settled answers. The trigger matters as much as the number, so both columns are listed.

Record type Common period Trigger point Typical authority
Tax and accounting records 7 years End of fiscal year US IRS guidance; UK HMRC
Employee payroll and tax records 4 years End of tax year US IRS employment tax rules
Form I-9 employment eligibility 3 years or duration plus 1 year Date of hire or termination, whichever is later US Citizenship and Immigration Services
Workplace injury records 5 years Date of injury US OSHA recordkeeping rule
Customer contracts 6 years from expiry Contract end date Common limitation period, not a statute
Applicant records (unsuccessful) 6 to 12 months Hiring decision Equal-opportunity monitoring practice
Marketing consent records Duration plus 2 years Consent withdrawn GDPR accountability practice

Two rows above are deliberately softer than the rest. The contract period reflects limitation periods rather than a filing rule, and the applicant and consent rows reflect practice rather than a single statute. Confirm the version that applies in your jurisdiction on the regulator's own site before locking them into a schedule you'll defend.

A Worked Deletion: 18,000 Mailboxes and 54 Million Items

Numbers make retention decisions real, so run the arithmetic on a mid-size scenario. Assume a company with 1,800 employees, 200 of whom leave each year, each holding a 30-gigabyte mailbox. On a three-year post-departure retention rule for departed-user mail, the steady-state holding is 600 departed mailboxes, or 18 terabytes of retained mail at any moment.

Now count items, not gigabytes. An average of 3,000 messages per mailbox across those 600 exits yields 1.8 million messages held for ex-employees at any one time. Extend the rule to ten years and the steady state becomes 2,000 mailboxes, 60 terabytes, and 6 million messages, because annual departures times years retained sets the population.

That growth is why retention periods get shortened in practice. Each additional year of mail retention adds roughly 200 mailboxes to the holding at this rate, and the cost lands in search, backup capacity, and review effort during litigation rather than in storage alone. The same math applies to support tickets, CRM records, and log data: annual volume times years retained equals the archive you must govern.

The harder question is defensibility, not cost. If a departed employee's mail is held for ten years on no stated basis, that period invites a challenge the company has to answer. If it's held for three years on a documented business need and destroyed on schedule, the policy holds up under scrutiny.

Five Mistakes That Break Retention Policies

The first mistake is publishing a schedule without a destruction process, which produces a document that prohibits nothing. The second is exempting backups from the schedule, which means the data never actually leaves. The third is letting custody sit with IT alone, without a records owner who can speak to the legal basis for each period.

The fourth mistake is applying a uniform period across everything, because a single number fails both directions at once: too short for tax records, too long for routine correspondence. The fifth, and most damaging, is not suspending deletion when a legal hold lands. Courts treat routine destruction after a preservation duty attaches as a serious problem, and a policy that doesn't name the suspension step gives employees no instruction to follow.

Run a tabletop exercise once a year against these five. Pick one category, trace a record from creation to destruction, and record where custody changes hands. Any gap you find in the trace is a gap an auditor or opposing counsel will find too.

What to Check Before You Publish

Publishing is the point where a draft becomes evidence, so a short pre-publication checklist earns its place. Verify that every schedule row has an owner, a trigger event, and a destroy method, then confirm those three fields survive contact with the actual systems. A period with no matching deletion capability in the platform is an aspiration.

Check the approval chain next. Legal, privacy, IT, and the business owner of each category should sign, and the signed version should carry an effective date and a review interval. Annual review suits most organizations; a major platform migration or a new regulation justifies an off-cycle revision.

Then test deletion for real on one low-risk category. Delete it, wait for the backup cycle, and confirm the data is gone from every tier, including the archive. If the verification step finds copies, fix the process before announcing the policy, because the first failed deletion after publication undermines the rest of the document.

FAQ

Does a data retention policy have to be in writing to count?

Yes for any organization that needs to demonstrate compliance. Regulators and courts evaluate documented rules, and GDPR's Article 30 records requirement effectively forces written retention periods for processing activities. An unwritten practice can't be audited, approved, or suspended during a legal hold.

How long can an employer keep an ex-employee's email?

There's no single legal maximum in US law, and the period is set by the employer's documented business need. A three-year post-departure holding is common in mid-size companies, while regulated sectors may require longer. The period must be stated in the schedule and applied consistently to every departure in the same category.

Are backups covered by the retention schedule?

Yes, and exempting them is the most common structural flaw. Backup copies hold records too, so the schedule needs a separate deletion path for each backup tier and a verification step confirming the deletion reached the archive. A live-system delete that leaves the backup untouched fails both the policy and most erasure requests.

The deletion becomes a potential sanctions issue, because the preservation duty overrides the routine schedule from the moment it attaches. Organizations should route hold notices to the exact systems and custodians affected and pause automated deletion for those categories. Recovery from backups is not guaranteed and may itself be challenged.

Does GDPR specify how long personal data can be kept?

No, the regulation sets the storage-limitation principle in Article 5 and leaves the period to the controller's documented purpose. That's why the justification column in your schedule carries as much weight as the number. Article 17 erasure requests must be answered within one month, extendable by two for complex cases per the European Commission.

Can a retention period be shortened after a policy is published?

Yes, provided the change doesn't breach a legal minimum, a litigation hold, or a contractual obligation. Shortening the period affects future records and, where systems allow, existing ones, which is why the destructive step for existing data gets its own approval line. Document the rationale, because a sudden shortening after a dispute arises invites scrutiny.

What's the difference between retention and preservation in Microsoft 365?

Retention keeps content for a defined period and deletes it afterward; preservation holds items in place and can be locked so no administrator can reduce it. Microsoft's documentation confirms a preservation lock cannot be removed once applied. Applying a lock to a broad policy with no deletion action is how tenants end up holding mail indefinitely.

Does a retention policy apply to machine logs and telemetry?

It should, and this is the category most schedules miss. Device logs, application telemetry, and security data often contain identifiers that count as personal data, and they grow far faster than human-authored records. Assign an owner, set a short period, and confirm the platform's own rollover settings match the schedule.

This article is general educational information about records management and is not a legal opinion or compliance certification. Retention periods depend on your jurisdiction, industry, contracts, and the specific data involved, so confirm each entry with a qualified local attorney or privacy professional before adopting a schedule. Nothing here guarantees a particular regulatory outcome.