Why the sticker price misleads

Most comparisons of server-side tagging cost stop at the hosting invoice. That is the one number that is easy to find and the one number least likely to decide the outcome. A cloud project that costs 120 dollars a month in compute can quite easily consume a day of engineering time in the same month, and a day of senior engineering time is not cheaper than a managed plan.

The honest framing is total cost of ownership across a year: infrastructure, plus the labour to build it, plus the labour to keep it running, plus the cost of the incidents you do not catch.

What you are actually paying for

Server-side tagging has four cost components, and different hosting models shift which one dominates.

• Compute: the containers that receive and forward requests. Priced per instance hour in a cloud project, or absorbed into a plan tier by a managed host.

• Egress: outbound data to ad platforms and analytics endpoints. Small per request, meaningful at scale.

• Setup: initial configuration, custom domain, SSL, client and tag configuration, consent handling.

• Maintenance: image updates, scaling adjustments, monitoring, and responding when forwarding silently fails.

The last item is the one that separates the two models. A managed host treats maintenance as their product. A client-owned project treats it as your responsibility, which is fine when you have a platform team and expensive when you do not.

A like-for-like comparison

The figures below assume a store doing around two million tagging requests per month, which is broadly typical for a brand spending between thirty and eighty thousand a month on paid media. Engineering time is priced at 100 US dollars per hour.

Note that the infrastructure lines are almost identical. The gap is entirely labour, and it compounds because maintenance recurs every month while setup happens once.

Where the crossover sits

Managed pricing tends to scale with request volume, while a well-tuned cloud project scales with instance count, which grows more slowly once you are past the minimum viable footprint. That means there is a volume above which self-hosting genuinely becomes cheaper.

In practice that crossover sits somewhere around four million requests per month for most setups, and it moves in favour of self-hosting if you already run other services on the same cloud account, already have infrastructure-as-code in place, and already have someone on call. If none of those three are true, the crossover is much further out than the arithmetic suggests.

The cold start question

Scaling to zero looks attractive on a cost sheet and is a poor idea for conversion tracking. A cold start adds latency to the request that carries your purchase event, and requests that time out are events that never arrive. Keeping at least one instance always on is the difference between a cheap setup and a working one, and it is the single largest fixed cost in a self-run project.

How to decide

Choose a managed host if tagging is the only workload you are putting in the cloud, if your volume is under a few million requests monthly, or if nobody on the team wants to own a container image. Choose a client-owned project if you have a platform team already, if data residency or contractual data ownership is a genuine requirement, or if your volume is high enough that plan pricing has become the dominant line.

What should not decide it is the belief that self-hosting is inherently cheaper. It is inherently more controllable, which is a different and often more valuable property.