Product feed management

Feeds go wrong when three people maintain three versions: the platform export, a spreadsheet somebody uploads on Fridays, and a set of manual overrides inside a channel. We decide where each field genuinely lives, usually the store itself with a small number of supplements for data the platform cannot hold, and everything downstream is generated from that. The spreadsheet either becomes a documented supplemental feed with an owner, or it goes away. This single decision removes most recurring feed problems before we touch a title.

Shopping has no keywords, so the title carries the matching. Most store titles are written for someone already on the page, where the brand name and a product code are enough. A feed title needs the words a stranger types: product type, key attribute, size or capacity, colour, then brand. We build the pattern per category rather than per product, using the fields you already have, so 4,000 titles rewrite themselves and stay correct when a variant changes. Length limits differ by channel, so the important words go first.

Every destination has fields it uses to decide relevance and fields it ignores. Google leans on product type, the Google category, identifiers, condition, and for clothing the age group, gender, colour and size. Other channels want a category from their own taxonomy. We map yours to theirs once, fill what is missing from data you already hold, and stop populating fields nobody reads. Where an attribute genuinely does not exist for your products, we say so rather than inventing a value, because a wrong value is worse than a blank one.

Custom labels are free text fields that let a campaign see something the channel cannot work out for itself. We normally use them for margin band, stock depth, seasonality, price band and whether an item is discounted. Once they exist, campaigns can be split by profitability instead of by category, low-stock items can be capped before they sell out mid-flight, and seasonal lines can be pushed and pulled without touching campaign structure. The labels are generated from your data, so they stay accurate on their own.

Rather than exporting several files by hand, we build one clean base and apply rules per channel: different title patterns where character limits differ, category mapping per taxonomy, image selection where a channel forbids overlays or text, and exclusions for products a channel will not accept. Rules are documented, so a person who is not us can see why a field looks the way it does. When a channel changes a requirement, the change happens in one place rather than in every export.

Every feed gets a weekly check against your live catalogue: how many products should be there, how many arrived, how many were rejected, and which fields lost data since the last run. Sudden drops almost always trace to one cause, such as an import that overwrote colours or a template change that broke a URL pattern. Catching it in a week rather than a quarter is most of the value. You get the numbers in plain language along with what we changed.