Launch day problems are almost never novel. Missing images, a category that resolves to an empty page, a price that is right in one place and wrong in another, and a marketplace feed that goes out with four hundred rejects. The same list, every season.
A list that repeats is a list you can check in advance.
Short answer: run the checks in order of how long the fix takes, not in order of how the site is built. Identifier and rights problems take days. Content gaps take hours. Layout problems take minutes. Most launches check them in exactly the reverse order and discover the expensive ones last.
Two weeks out: the things that take days
Identifiers. Every sellable unit has an internal SKU that is unique and has never been used before. Every unit sold through a third party has a valid GTIN whose check digit computes and whose prefix belongs to you. Obtaining or correcting identifiers is an external process with its own lead time, which is why it is first.
Rights and releases. Every asset going live has a licence that covers the use and does not expire during the campaign, and any asset with a recognisable person or property has a release on file. This is the check that can stop a launch outright, and it is the one most often done by assumption.
Channel eligibility. Each product is actually allowed on each channel it is scheduled for. Restricted categories, market availability, regulatory fields. Discovering a restriction two days out means either a fast exemption process or a smaller launch.
Data completeness against every rule set. Not a spot check. Every product scored against every channel it is destined for, with the missing fields named and assigned. If your platform exposes this over the API, it belongs in an automated report; the reasoning is in syndicating product data to marketplaces.
One week out: the things that take hours
Image coverage. Every product has every mandatory frame from its category shot list, and every variant that needs its own image has one. Run it as a query against the association data rather than by looking, because looking finds about eighty per cent.
Alt text. Every image that will appear has a description. Not a filename, not the product title repeated. The WCAG guidance is the reference for what makes one adequate, and the practical rule is that the description should let someone who cannot see the image understand why it is there.
Copy completeness and localisation. Every localisable field has a value in every live market. Empty fields fall back to another locale in most systems, silently, which means a market can launch looking complete and reading in the wrong language.
Category and navigation. Every category that will be linked has products in it. Empty category pages are the most common launch-day embarrassment and they are trivially detectable in advance.
Price and stock reconciliation. The price shown matches the price in the authoritative system, in every market and channel. Where the values are cached, the cache has actually refreshed. This is where the field ownership list from single source of truth applied to product data earns its keep.
Two days out: the things that take minutes
Rendition spot check. Pull the actual URLs a product page will use, in each size, and look at them. Content-aware cropping is right most of the time and wrong occasionally, and wrong looks like a chair with no back.
Format and weight. Confirm modern formats are being served where supported and that no hero image is arriving at two megabytes. Images are usually the largest contentful paint element on a product page, so this is the single check with the most measurable effect on the launch.
Structured data. Product markup present and valid on product pages, with price, availability and image. Wrong or missing markup does not break the page and does quietly cost you the rich result.
Redirects. Every URL that existed before and does not now resolves somewhere sensible. This applies to discontinued products as much as to restructured categories.
Feeds. Generate each channel feed and validate it without publishing. Count the rows. A feed with fewer rows than expected is a filter you did not know about.
Launch day: watch four things
Resist the urge to watch everything. Four signals tell you whether it worked.
- Error rate on product pages, which catches missing assets and broken references immediately.
- Feed acceptance counts per channel, compared against what you sent.
- Page weight on a real device, because a synthetic test on a fast connection will not show you what a customer sees.
- The support inbox, which is the highest-fidelity monitoring you have and the one nobody puts on the dashboard.
The day after
Two jobs, and the second one is the one that gets skipped.
Reconcile. Full comparison between what you intended to publish and what is live on each surface. Divergence found here is a bug in the change path, not a one-off, and it will recur next season if it is only patched.
Write down what went wrong. Every launch produces three or four surprises. Each one becomes a line on this checklist, permanently. A checklist that grows by three lines a season and gets used is worth more than any tool you could buy, and the operating model it belongs to is in running product content as an operation.
What can be automated?
More than most teams automate, and the criterion is simple: any check that is a query should be a query.
Completeness, image coverage, alt text presence, empty categories, identifier validity, feed row counts and price reconciliation are all queries. Write them once, run them nightly from six weeks out, and let the list of failures be the work queue. The value is not the automation, it is that the failures surface while there is still time to fix them cheaply.
What cannot be automated is judgement. Whether a crop looks right, whether the copy reads well in a market you do not speak, whether the product photography matches last season. Reserve human attention for those, and stop spending it on things a query answers better.
The upstream work that makes launch week calm rather than heroic is in the product image workflow and product photography without a studio backlog, both of which are about removing the queue rather than working through it faster.