A headless D2C replatform that took LCP from 4.9s to 1.6s

Kesari & Co. sells ayurvedic skincare direct to consumers — 2,300 SKUs once you multiply shade by size, ₹40 crore of annual GMV, and 74% of their traffic arriving on a mid-range Android phone over a patchy 4G connection. They had outgrown their Shopify theme in every direction at once.

The storefront was the bottleneck

Largest Contentful Paint on a Redmi-class device over throttled 4G was 4.9 seconds. Core Web Vitals were failing on every template that mattered. The cause was not mysterious: fourteen Shopify apps were each injecting their own script bundle, and the storefront was shipping ~900 KB of JavaScript before a single product image appeared. Every app had been added to solve a real problem, and collectively they were the problem.

Three constraints an off-the-shelf theme could not satisfy

• Their promotion logic had outgrown the platform. Build-your-own-routine bundles with tiered discounts, subscribe-and-save, and a loyalty tier multiplier — expressed as rules the marketing team could author themselves. The theme-level discount tooling they were using could not express it, so they were running promotions by manually creating and destroying variants. In one Diwali campaign this produced a bundle that could be bought for less than the cost of one of its components.
• Inventory truth lived somewhere else. Stock synced from their 3PL's warehouse system on a six-hour cron. During a flash drop, six hours is an eternity: they oversold a hero SKU by 400 units on one launch and spent three weeks apologising.
• ONDC was coming. They wanted to list on the Open Network for Digital Commerce without bolting a second, divergent catalogue onto the side of the business — and without letting network traffic anywhere near the funnel that pays the bills.

The brief they arrived with was "rebuild our store". The brief we agreed on was narrower and more useful: get the storefront off the platform's rendering path, get promotions into a rules engine, get inventory to real time, and make ONDC an adapter rather than a fork.

We cut over in two stages — collection and product pages first, running on 20% of traffic behind a split for eleven days, then the rest. There was no big-bang weekend.

Speed turned directly into revenue

• LCP on a mid-tier Android over throttled 4G: 4.9s → 1.6s. All Core Web Vitals green, on mobile, on the templates that actually carry traffic — not just on the homepage that everyone benchmarks.
• Mobile conversion rate rose 34% over the eight weeks following full cutover, measured against the same traffic mix. Nothing else changed in that window: no new creative, no new discount structure, no new acquisition channel. This is the cleanest read we have ever had on the value of a fast storefront.
• 80% less JavaScript shipped before first paint. The number that mattered more internally: their marketing team can no longer accidentally add 60 KB of blocking script by installing an app, because there are no apps.
• Zero oversells across four subsequent flash drops, including one that moved 3,100 units in nine minutes. The reservation ledger held.

What changed for the business

Promotions that used to take a developer and two days now take a marketer and twenty minutes, and the preview mode has caught three pricing mistakes before they reached a customer — including one that would have repeated the Diwali below-cost bundle almost exactly.

They went live on ONDC four months later. That was a two-week piece of work, not a replatform, because the adapter was already sitting there speaking to the same catalogue. That is the entire argument for building it as an adapter, and it is worth more than any performance number on this page.