The architecture guide covers the broader trade-offs.
Identify the measurement problem first
A gap between shop and GA4 numbers is a finding. Consent, different definitions, duplicate events, time zones and technical failures can all contribute. Identify the cause before choosing a hosting provider.
A universal cookie phase-out is not a sound justification. Google is maintaining Chrome's existing approach to third-party cookie choice and has announced the retirement of several Privacy Sandbox technologies. Source: Google, October 2025.
What server-side actually is
A server container receives events, processes them and sends configured data to destinations. Specify three things:
- Receive. Which events arrive from the browser, shop or backend?
- Process. Which fields does each destination need and receive? Which consent rules apply?
- Forward. Which tags or API integrations send the data, and how are they tested?
Transformations can allow, change or exclude event fields exposed to tags. This requires configuration and does not automatically anonymise data. Inspect parallel browser requests too. Hashing an email or truncating an IP address does not automatically remove its personal nature.
Browser limits still apply
Your own endpoint changes the transmission path. Browsers and blockers can still restrict it. Measure whether the change reduces event loss in your actual setup.
HttpOnly does not guarantee a cookie lifetime. It prevents JavaScript access. WebKit caps cookies set through detected CNAME or IP cloaking at seven days; script-writable storage has a separate limit when users do not interact with the site. A blanket comparison of “7 days in the browser, 400 days on the server” is therefore incorrect. Source: WebKit Tracking Prevention.
What server-side does not solve
Missing consent. Processing that requires consent still requires it on a server. The CMP captures the choice in the browser. With Google Consent Mode, the web container sends the state to the server and tags respond to it. The consent guide covers the chain through to each destination.
Invalid events. An extra server does not repair a wrong order ID or missing currency. The Measurement Blueprint defines valid data.
Unclear ownership. Managed hosting does not automatically maintain your tags, consent mappings or event specifications.
When it is worth it
Compare the expected benefit with implementation, hosting and ongoing QA. Consider the decisions affected, the cause of the measurement gap and the data controls needed. Neither sessions nor ad spend provide a universal break-even threshold.
A small funnel can have specific requirements for data sharing. A high-traffic site may still be adequately served by repairing its browser setup.
Stape vs Cloud Run vs your own server
The current Stape pricing comparison separates monthly and annual billing and explains request budgets and plan limits.
| Option | What to budget | What stays with your team |
|---|---|---|
| Stape | Plan, required features, incoming requests and headroom | Tracking configuration, consent, testing and data-quality monitoring |
| Cloud Run | CPU, memory, region, runtime, networking and other services | Configuration and operation of the tagging container |
| Your own server | Server, redundancy, updates, monitoring and working time | Infrastructure and tracking operations |
Cloud Run does not become cheaper at a universal request count. Compare the same period and functionality. Google's pricing page describes the cost factors. An EU region alone does not settle all data-flow or contractual requirements.
The calculator below compares Stape standard plans only. It does not calculate general ROI or a cost advantage over Cloud Run.
Incoming requests are not equivalent to sessions or browser events. Use measured requests including optional features.
1,000,000 requests per month
Business: 100 USD per month. Allowance: 5,000,000 requests/month.
Hosting only, excluding taxes, setup and ongoing QA. Also check daily limits, features and peak loads. This is not a cost comparison with self-operated Cloud Run. Checked 8 September 2026.
Stape plans and termsDisclosure
Stape.io has been our declared server-side-tagging partner since 2024 and is listed as such in our imprint. The status gives us access to roadmap calls and occasional promo codes for discovery clients. And money: stape.io pays us back 20 % of the licence fee every month as cashback, so we earn on every stape.io licence that runs through us. Know that before you read the recommendation above.
Other server-side setups (self-hosting, GCP Cloud Run / App Engine) are just as valid, we recommend them where they fit. These recommendations are based on operational experience, not on affiliate incentives.
Diagnostic checklist before deciding
- Align the comparison. Match order definitions, periods, time zones and the consent policy. Separate observed and modelled values.
- Locate failures. Inspect browser requests, server preview and destination data. The difference between two analytics tools is not an isolated adblocker rate.
- Plan acceptance. Document test orders, deduplication, permitted fields and all consent states. Name the people responsible for hosting and tracking.
Scope and implementation are covered by Measurement & Privacy Engineering.