Every performance conversation starts with load time and ends with a CDN. Then the owner opens the site on their own phone, taps the WhatsApp button, and it does nothing for half a second. That gap is what Interaction to Next Paint measures, and it is the metric most Indian sites fail while patting themselves on the back for a green PageSpeed score.
What the number actually means
INP watches every interaction across a real visit – taps, keystrokes, pointer events – and reports the longest wait you subjected a visitor to, with a small allowance for the single worst outlier. Google grades it on the 75th percentile of field traffic, not on your laptop: under 200 milliseconds is good, over 500 is poor. The important word in that sentence is field. A page can pass in a lab and fail in Sonipat, because the lab machine has a processor your customers do not have.
Why it fails on the phone in their hand
Input delay is a queue problem. When you tap, the browser cannot run your click handler until the main thread finishes whatever it is doing. On a mid-range Android, a single 400 ms script is enough to swallow the tap entirely. Three things cause almost all of what we see:
- Third-party scripts nobody owns. A chat bubble, a heatmap recorder and a tag container that loads four more containers. They ship to every page, they run at the worst moment, and no one can say what breaks if they go.
- JavaScript bundled for the whole site. A slider library on a page with one slider. A form framework loaded where the only form is three fields.
- Work done in the wrong order. An accordion that animates height and forces layout on every frame, or a click handler that fires a fetch and then re-renders a list of two hundred items.
How we diagnose it
Field data first: which URLs are actually failing, and at what percentile. Then a performance trace on a throttled device – 4x CPU slowdown, mid-range profile – with the interaction recorded rather than the load. We look for the longest task that starts within 50 ms of the click. That task has a name on the flame chart, and it is usually not the theme.
The fix order, in the order that pays
- Delete. The fastest code is the code that is not shipped. Half of our INP wins are a vendor script removed because nobody could tell us what it was for.
- Defer. A WhatsApp launcher that shows a button and loads its SDK on the first tap. A tag container that starts after the page has settled.
- Split. Long loops broken into chunks so the thread breathes between them.
- Replace. CSS scroll-snap instead of a carousel library. A native
<details>instead of an accordion plugin.
What will not fix INP: a faster server, a CDN, smaller images, or a theme update. Those move the network, and this metric is not about the network. Measure the interaction, not the page weight, and the list of things to do gets short and very specific.
Leave a Reply