JD Sports, Contract
Overview
JD Sports had just put a new storefront live, with five months to clear a large backlog of product work before peak trading. I joined as one of a small contract team delivering against it.
My work sat on the three surfaces closest to the sale: the checkout, the click & collect store finder, and the sizing content behind every product page.
Duration: May 2026 to December 2026
Stack: TypeScript 路 React 路 Next.js 路 Amplience 路 Playwright 路 Contentsquare
Role: Principal Software Engineer, contract
Checkout: five pages to one
Checkout ran across five pages. Every page was another load, another wait, and another opportunity for a customer who had already decided to buy to change their mind.
I delivered the collapse to a single page. The customer sees their preferred delivery and payment options on one screen, and the clicks between deciding and ordering drop with the page count. After the main release, Contentsquare session data let the UX team see where the friction points were, which allowed us to ship three further iterations.
What moved: Largest Contentful Paint improved by 35%. Once checkout was one page, a lot of JavaScript loaded on it, so we deferred that loading until after the content showed. Contentsquare showed fewer friction signals and fewer stuck sessions. The measure the work was held to was the share of sessions that go on to purchase once they reach checkout.
Click & collect: address lookup from 5s to 1.7s
The address lookup in the click & collect store finder took around five seconds to return. Five seconds is long enough that a customer stops trusting the field: they retype, they click again, or they give up on collection entirely.
What moved: Lookup response time went from roughly 5 seconds to 1.7 seconds. The store finder used to search on every keypress, so we added a submit button, and we moved the address lookup to Loqate typeahead.
Size guides: from six weeks to two days
This began as a small ticket. Product pages listed every size guide in a dropdown, and the ask was to show only the one that applies to the product in front of the customer.
Delivering it exposed a larger problem. Hiding the button where no guide existed made visible how many products had no size guide at all, and the fix was content work, owned by a content team already fully committed to Black Friday.
It also made the underlying constraint obvious: the size guides lived in the codebase. Correcting a single sizing table meant gathering the information, routing it through three teams, then a developer, a pull request and a deploy. Roughly six weeks, for a content change.
I argued that sizing belonged in the CMS rather than in code, and took the whole piece rather than hand the team a dependency:
- Amplience content type, schema and a custom authoring extension. The extension presents a sizing table as the spreadsheet grid the content team already worked in, so adoption cost them no new tooling.
- 1,250 size guides migrated out of the codebase into Amplience, across every country.
- A programmatic backfill for the gaps, so the content team did not carry them into peak. An agent loop reads a brand's published sizing, normalises it to the Amplience schema, writes it into the sandbox through the Amplience CLI, and logs every row to a shared spreadsheet that acts as both the pipeline's memory and the stakeholders' live view of progress. Promotion out of sandbox stays with a human.
What moved: The first new brand through the pipeline was 51 size charts in two days, against a six week manual route per change. Sizing content now publishes without a developer and without a deploy, and the gaps were cleared ahead of peak rather than during it.
Quality at delivery pace
A backlog that size, a five month window and heavy AI assisted delivery move the bottleneck off writing code and onto review.
Playwright journeys cover checkout and store search. Alongside them, every pull request went through an agent driven pass with Claude in Chrome reading the DOM and the accessibility tree rather than the rendered page: missing accessible names, skipped heading levels, elements hidden visually but still in the tab order, stale ARIA labels left behind by a rename. Reading the tree rather than the screen surfaces the defects a visual check cannot see. It became a standing part of our pull request process.
Outcomes
- Checkout collapsed from five pages to one, with a 35% improvement in Largest Contentful Paint and fewer friction and stuck signals in Contentsquare
- Click & collect address lookup reduced from roughly 5 seconds to 1.7 seconds
- 1,250 size guides migrated out of the codebase into Amplience across every country, behind a custom authoring extension built around how the content team already worked
- Backfill pipeline delivered 51 size charts for a new brand in two days, against a six week manual route per change
- Sizing content now publishes with no developer, no pull request and no deploy
