Server-side tagging is presented as the fix for everything wrong with modern measurement. It fixes some of it, at real cost, and the honest answer for most sites is "not yet".
What server-side GTM actually fixes
Instead of the browser sending data to Google, Meta and everyone else, it sends to your endpoint on your domain, which forwards server-to-server.
First-party cookies with a full lifetime. Safari's ITP caps cookies set by client-side JavaScript at seven days — sometimes 24 hours. A returning visitor on Safari looks new, and attribution windows silently collapse. Cookies set server-side with an HTTP header are not subject to that cap. For many sites this is the single biggest gain.
Ad-blocker resilience. Blockers match known third-party endpoints. Requests to your own domain are not automatically blocked. Expect meaningful recovery, not total — blockers are not helpless here.
Control over what leaves. You can redact PII, drop parameters, and decide per-vendor what is shared. For regulated industries this is often the actual reason to adopt it.
Less client-side JavaScript. Vendor tags move off the page, which helps performance.
What it doesn't fix
It is not a consent bypass. Consent obligations attach to processing, not transport. Anyone selling sGTM as a way around consent is selling you a legal problem.
It does not fix a bad data layer. Garbage in, garbage forwarded reliably. If events are misnamed or values wrong, sGTM propagates that faster. Fix the data layer first — always.
It does not help much with cross-device identity. That needs a logged-in user id, which you can implement without sGTM.
It does not reduce complexity. It adds a server to operate, monitor and debug.
The cost model
Three real costs:
Infrastructure. A tagging server, typically on Cloud Run or App Engine, that must stay warm and scale with traffic. Modest for small sites, non-trivial at scale — and it is a production dependency: if it goes down, you lose data.
Engineering time. Initial setup, DNS and certificates, then ongoing maintenance. Debugging is harder because failures are now server-side and invisible from the browser.
Expertise. A smaller pool of people can competently operate it.
Against that, the benefit is a percentage improvement in data completeness. Whether that is worth it depends entirely on what a percentage of your data is worth.
Measurement Protocol for money events
Even without full sGTM, the Measurement Protocol is worth adopting for purchases and refunds.
The browser is an unreliable narrator for money. The tab closes, the network drops, a blocker intervenes. Sending purchase server-side, when the order is written, makes your most important number as reliable as your database.
The pattern: keep client-side events for behaviour (view_item, add_to_cart), send purchase and refund server-side with the same transaction_id so GA4 deduplicates. Refunds in particular almost never happen in a browser — an admin processes them — so server-side is the only sensible route.
Migration path
If you decide it is worth it, do not move everything at once:
Fix the data layer first. Non-negotiable. sGTM on a broken implementation makes things worse.
Stand up the tagging server on a subdomain, with proper certificates.
Run in parallel. Duplicate GA4 into a second property via sGTM and compare for a few weeks.
Move GA4 across once the numbers agree.
Move advertising tags one at a time, checking conversion counts after each.
Add Measurement Protocol for purchase and refund.
Expect this to take weeks, not an afternoon.
Decision checklist
Server-side tagging is likely worth it if most of these hold:
Significant Safari or iOS traffic where ITP is measurably hurting attribution
Meaningful ad spend where a few percent better attribution changes bidding
Regulatory pressure to control what leaves your infrastructure
An existing, documented, trustworthy data layer
Someone who can own a production service
It is probably not worth it yet if:
Your data layer is inconsistent or undocumented
Ad spend is small
Nobody can own another production dependency
You have not first fixed consent and purchase accuracy
The uncomfortable summary: most teams asking about server-side tagging have unresolved client-side problems. Duplicate purchases, no tracking plan, consent implemented incorrectly. Fixing those delivers far more than sGTM will, for a fraction of the cost.
Do the boring work first. Then, if the case still holds, go server-side.