Photo: free stock photography (Unsplash licence) — see imprint
Why the score misleads
Lighthouse simulates one page load on one throttled connection. It cannot see that your logged-in B2B customer skips the homepage entirely and lives on a 400-row order list — which the score never tests and your revenue depends on.
Optimise the pages that carry money
Pull your top ten pages by revenue-influenced sessions, not by traffic. Usually that is the category listing, the product detail page and the cart — not the homepage everyone is measuring.
The four fixes that always work
Serve modern image formats at the right size. Defer every script that is not needed for the first interaction. Cache the HTML. Remove the two tracking pixels marketing forgot they installed in 2021. That is most of the win, on most shops.
Set a budget and defend it
Agree a performance budget in the project — e.g. LCP under 2.5s on the product page — and fail the build when a new plugin breaks it. Speed is not a sprint you do once; it is a limit you enforce.
- Optimise revenue pages, not the homepage score.
- Images, deferred JS, HTML cache, dead pixels.
- Enforce a performance budget in CI.
Frequently asked questions
Lighthouse loads one page, once, on a throttled connection. Your logged-in B2B buyer skips the homepage entirely and works on a long order list the score never tests. A weak score on a page nobody buys through is not an emergency. A slow product page with a good score is. The score and your customers are measuring different things.
No. We have taken shops to 98 and watched them convert worse, because the effort went into the homepage while the category listing and the cart stayed slow. The score is a proxy for experience, not a replacement for it. Chase it while it still points at something a buyer would notice, and stop when it stops.
Sort your pages by revenue-influenced sessions, not by traffic. On most B2B shops that puts the category listing, the product detail page and the cart at the top — and the homepage, which everyone measures, further down than expected. Fix those three and you have fixed what buyers actually feel. The homepage can wait its turn.
Agree a performance budget in the project — for example LCP under 2.5s on the product page — and let the build fail when a change breaks it. Without that, every optimisation you pay for gets quietly eaten by the next installation. Speed is not a sprint you run once; it is a limit somebody has to enforce, in CI, automatically.
We do this for a living — Shopware, Node.js, React, ERP integration and automation for B2B.
Talk to an engineer