Every team that has built a fast website has watched it get slower. Not in one bad release, but a hundred grams at a time: an analytics script here, a font weight there, a library added for one component. Six months later the page takes four seconds and nobody can point at the change that did it.
Why code review does not catch it
Each individual addition is defensible. The tag is needed by marketing, the library saves a day of work, the extra font weight looks better. No reviewer is going to block a pull request over ninety kilobytes. The cost is real but it is distributed across dozens of reasonable decisions.
This is precisely the shape of problem that an automated threshold solves and human judgement does not.
Setting a budget that means something
Pick numbers tied to user experience rather than to convenience. Largest Contentful Paint under 2.5 seconds on a mid-tier mobile device over a throttled connection. Total JavaScript under a stated kilobyte figure. Cumulative Layout Shift under 0.1.
Measure on the hardware and network your users actually have, not on a developer laptop over office fibre. The gap between those two is where most performance problems hide.
Enforce it where it cannot be ignored
Run Lighthouse in CI against a production-like build. Fail the pipeline when a budget is exceeded. This is the part teams resist, and it is the part that works, because a warning in a dashboard is a warning nobody reads.
Failing the build turns the conversation from whether the addition is worth it in the abstract into what gets removed to make room. That is a far more productive discussion and it happens at the right moment.
Leave an escape hatch
Sometimes a business reason genuinely justifies exceeding the budget. Allow an explicit, recorded override rather than pretending it never happens. An unrecorded exception becomes the new normal; a recorded one is revisited.
What it is worth
The relationship between page speed and conversion is well established and holds across sectors. Treating speed as a delivery requirement rather than an optimisation project is one of the few engineering decisions with a straightforward commercial argument behind it.