Marketing teams celebrate campaign launches. Performance engineers watch bounce rates on mid-range Androids in second-tier cities. Those two worlds meet in Core Web Vitals, and the meeting is often uncomfortable. A beautiful creative that never paints is just a billing line for media. Largest Contentful Paint, Interaction to Next Paint, and Cumulative Layout Shift remain useful because they approximate what a person feels. In markets where 4G is shared with an entire household and devices are three years old, a heavy homepage is not a branding statement. It is a closed door. Lab scores on a developer laptop are a starting rumor, not a verdict.
What we keep seeing
Hero images shipped at desktop resolution. Third-party tags injected without budgets. Client-side rendering for content that could have been HTML. Fonts blocking text. Each item is familiar. Together they turn a 1.2 second lab LCP into a 5 second field LCP on real devices. The paid click arrives. The page thinks. The user leaves. Attribution models blame the channel.
A Vietnam e-commerce brand cut LCP by roughly 40 percent by serving appropriately sized WebP images, deferring non-critical scripts, and moving product detail above the fold to server-rendered HTML. Paid traffic conversion improved enough to fund the engineering time twice over in a quarter. The marketing team noticed first because CPA stopped drifting for no obvious creative reason.
INP is the quiet killer
After FID retired, INP forced teams to confront main-thread congestion. Heavy analytics bundles and chat widgets fighting for the thread make taps feel sticky. Users do not file bugs. They leave. A sticky add-to-cart button is a silent refund of your media spend.
Caution: optimizing only the homepage for lab scores while checkout remains sluggish is a classic self-deception. Measure the money path. Field data from Chrome UX Report and your own RUM should guide priorities, not a single Lighthouse screenshot in a slide. Festive campaigns amplify the gap: traffic rises exactly when the page is least ready.
A practical performance diet
- Set performance budgets in CI for JS weight and LCP candidates.
- Audit third parties quarterly; remove what sales no longer uses.
- Prefer responsive images and modern formats with sensible fallbacks.
- Test on mid-tier devices and throttled networks, not only flagship phones on office Wi-Fi.
- Make a performance owner visible in campaign launch checklists.
Campaign launches and the performance contract
Treat every major campaign as a performance event. Freeze non-essential third parties before the media spend starts. Pre-negotiate tag budgets with marketing the way you pre-negotiate creative. If a new pixel cannot prove incremental value, it does not get to sit on the critical path. Engineers should not be the only people who feel latency; media buyers should see field LCP next to CPA in the same weekly report.
Image pipelines deserve the same seriousness as checkout. Automate responsive variants. Reject uploads that skip compression. Teach content teams why a 4MB hero is a conversion bug. The cultural fix matters: performance stops being a surprise firefight when it becomes a shared contract with dates, owners, and the power to delay a launch that would burn budget on a slow page. A useful internal test is whether a skeptical finance partner can understand the unit economics without a translator from engineering slang. If the story only works in a specialist room, it is not ready for production funding. Translate early. Funding follows comprehension more often than it follows novelty. None of this removes the need for craft. It simply refuses to confuse craft with theater. Craft shows up in the details customers and operators feel. Theater shows up in diagrams that never change a Monday morning workflow. Keep the craft. Cut the theater. Repeat until the program is dull in the best sense.
Takeaway
Treat Core Web Vitals as product quality metrics tied to revenue. Fast enough is a market requirement in Asia, not a niche engineering hobby. If your acquisition budget is rising while field LCP is drifting, you are not buying growth. You are buying bounce.