All articles Privacy & compliance

GDPR-Compliant Analytics: A Setup Checklist

Translucent glass checklist panel with glowing blue outlines floating on a black background

Making your stack analytics GDPR compliant is not a legal review you run once and file away. It is a set of defaults in your code: which events fire, what fields they carry, how long they live, and who can read them. This checklist covers the decisions that actually change your exposure, in the order most teams implement them.

It assumes you already have a privacy notice and a consent banner. What it covers is the part most teams skip: the wiring between that banner and the SDK, the retention rules nobody set, and the traffic you never meant to collect. It is engineering guidance, not legal advice.

Use it as a pre-launch gate, or as an audit for a tracker you already run.

Map each event to a purpose and a lawful basis

Under GDPR Article 6 you need a lawful basis for each purpose you process data for, and under the ePrivacy rules you generally need consent before storing or reading information on a device. That makes your tracker configuration a legal artifact. Build a table with four columns: event name, purpose, lawful basis, fields collected. If an event cannot name its purpose, delete it. If two purposes share an event, split them: product analytics (which features people use) and marketing analytics (which campaigns convert) should not travel in the same payload. Ad identifiers, campaign-stuffed user properties and cross-site IDs have no place in a product analytics stream.

Keep the product stream first-party

Your product events should be readable only inside your own tooling: your domain, your storage, your access rules. Keeping marketing pixels on a separate path also makes withdrawal simple — you can switch off one stream without losing your funnel data.

Article 7 requires consent to be as easy to withdraw as to give, and requires you to be able to prove it. A banner that writes a cookie while the SDK keeps firing on page load fails both tests. Route every call through one gate, so consent stays a single decision point:

Abstract diagram of a consent gate connecting two dark modules with a blue line

if (consent.has("analytics")) analytics.track("signup_started", props);

Store the consent state with a version, a timestamp and the wording the user saw, so you can answer “what did this person agree to, and when” six months later. When consent is withdrawn, stop collection immediately and delete or anonymize what you hold for that identifier if your policy says so. The same rule applies to every replay and heatmap tool in the stack.

Default the gate to false, and never infer consent from a page view, a scroll or a closed banner. Silent defaults are where most of the risk sits.

Honor Global Privacy Control on the client and the server

Global Privacy Control is a browser-level opt-out preference, sent as the Sec-GPC request header. Two mistakes to avoid: honoring it only in JavaScript, and honoring it only on page views. If you accept events server-side or through a mobile SDK, put the check in the same middleware that authenticates the request. Then decide what GPC means for your product — usually no non-essential analytics for that visitor, no profile building, no export to third parties — and write it down so support, sales and engineering give the same answer. Prynt Analytics honors Global Privacy Control by default.

Set retention per data type and enforce it automatically

Storage limitation (Article 5(1)(e)) is the rule teams break by accident, because retention is a dashboard setting nobody edited. Write a number for each layer: raw events, rollups, session replay, identified profiles, and the logs that tie a visitor to a company. Then verify that deletion is automatic and observable. A scheduled job with a run log you can show a customer beats a policy paragraph every time.

Layered translucent plates fading into darkness to represent data retention windows

Minimize fields before you store them

Ask of each field: does a decision change if it is missing? A coarse country code usually answers the question you had for a raw IP. An internal user ID answers the question you had for an email. Where you need a value only for matching, keep a keyed hash rather than the value itself, and check the delete-by-user path in the docs (https://pryntid.com/docs) before you need it in a hurry.

Exclude bots and agents before their sessions become personal data

Crawlers, uptime monitors and AI agents generate sessions that look like people. They inflate funnels, distort retention, and give you a reason to keep records about traffic that was never a person. Filter at ingestion rather than in a dashboard filter, and log what you dropped so your numbers stay explainable. Prynt Analytics excludes bot sessions from counts; the classification approach is described in our bot detection guide (https://pryntid.com/bot-detection).

Document the plumbing, not just the policy

A privacy notice describes intent; the DPA, the subprocessor list and the deletion path are what a customer’s security team asks for. Keep three things current: the data processing agreement you sign (https://pryntid.com/dpa), the subprocessors with their regions and transfer mechanism, and a tested way to export or delete everything tied to one userId. Test the last one with a real identifier each quarter — the failure mode is discovering mid-incident that nobody remembers which table holds the mapping.

Stream of blue light dots passing through a fine mesh that filters a few of them out

FAQ

Not automatically. If your analytics stores or reads information on a device, ePrivacy rules generally require consent unless the storage is strictly necessary. Some first-party, aggregated measurement may rely on another basis under national law, but that is a call for your counsel. What you can fix today is making consent optional at the code level, so switching is a config change instead of a refactor.

Does Global Privacy Control apply to product analytics?

Treat it as an opt-out preference for the visitor’s browser. Honoring it for non-essential analytics, on the client and on the server, is the defensible default; document exactly what you do.

How long can we keep analytics data?

GDPR sets no single number. It asks that you keep data no longer than needed for the stated purpose. Pick a window per data type, write it in the notice, delete automatically, and be able to show the job that did it.

Put the checklist to work

Prynt Analytics is built around these defaults: bot sessions excluded from counts, Global Privacy Control honored, and events you define field by field. Start a 14-day trial at https://pryntid.com and run one funnel through this checklist before you migrate the rest. If you need the paperwork first, the DPA and subprocessor list are at https://pryntid.com/dpa.

Try it free

Prynt is device intelligence with a free plan: 5,000 identifications a month, visitor IDs with confidence and bot detection — no credit card. Upgrade to Pro for VPN, proxy, tampering and behavioral signals.

Keep reading