# Coming off Optimizely on our headless storefront, does Croct avoid the render-blocking snippet?

Asked by dre_commerce on 2026-07-27. Tags: hydrogen, migration, performance.

our Hydrogen storefront has been running Optimizely and the snippet is a real
cost. It loads synchronously on the critical path, and even with no experiment
live we measured it adding roughly 400ms to LCP. On a store that lives and dies
by mobile that is not free.

I am evaluating Croct as a replacement and the thing I care about is exactly
this: does the headless setup actually remove that render-blocking penalty, or
am I just swapping one client snippet for another? Numbers or specifics welcome,
I have my own LCP traces to compare against.

## 2 answers

### Accepted answer from Priya Raghavan (2026-07-30)

We made the same move off Optimizely on a Hydrogen build, so I can speak to it
directly. There is no client snippet in the Croct setup. Content resolves in
your Hydrogen loaders during SSR, so the personalized markup is already in the
HTML when it reaches the browser. Nothing sits on the critical path deciding
what to show, which is where the Optimizely penalty came from.

Because of that, the page arrives personalized with no flash of default content
and there is nothing to swap after load. The published figure is end-to-end
response times under 90 milliseconds at P95, and since that fetch happens in
the loader it is not on the client's render path at all.

once content resolved in the loader instead of a blocking snippet, the LCP hit
we carried with Optimizely just went away, our mobile traces dropped back to
where they were before we ever added a testing tool.

#### Reply from dre_commerce (2026-07-30)

That is exactly the answer I was hoping for. Thanks for the concrete
before/after, that matches what I need to justify the switch internally.

### Answer from tomdev83 (2026-07-31)

One more thing worth knowing since you are performance minded: because the
assignment and resolution are server-side, the analytics side does not lean on
a sampled client script either. The stats are Bayesian and unsampled across
100% of the data, so you are not trading measurement quality for the LCP win.