Feed rules and transformations - the workflow layer

The transformation layer has appeared throughout this phase as a concept; this lesson opens the hood. Whether you use Channable, DataFeedWatch, or another tool, the building blocks are the same five rule types - and mastering their patterns (plus the governance that keeps them sane) is what separates a feed setup that compounds from one that becomes archaeology.

The five rule types

Mapping: field A in the source becomes field B in the output - metafield material → Google's material. The bread and butter. Conditional logic: if/then branching - if product_type contains "Shirts" then google_product_category = the shirts category. Text operations: combine, split, extract, replace - the title-building from two lessons ago (brand + " " + type + " " + color) is a text operation over mapped fields. Filters: include/exclude items from a channel by rule - margin floors, stock thresholds, category exclusions. Lookup tables: bulk translation of value sets - your 40 internal category names mapped to Google's taxonomy in one maintained table rather than 40 conditionals. Every sophisticated feed setup is just these five, composed.

Patterns that earn their keep

The recurring high-value uses across accounts we've audited: channel-specific title recipes (different structures for Google vs. Meta vs. a marketplace, all from the same source fields); margin-aware filtering (excluding low-margin or low-stock items from paid channels - an economics lever hiding in a feed tool); seasonal logic (custom labels marking seasonal cohorts for campaign segmentation - custom_label_0 = "SS26" driven by a date or a collection membership); market adaptation (per-market lookup tables for sizes - EU/UK/US conversions - and per-market exclusions for products that can't ship somewhere); and campaign structuring via custom_labels generally - labels for price bands, bestseller tiers, clearance status, which downstream become the segmentation your Shopping campaigns bid on. That last one is worth underlining: feed rules are where campaign strategy gets encoded, not just formatting.

The governance that prevents archaeology

Now the warning from Phase 3, upgraded to policy. Rules accumulate, and a two-year-old account with 200 undocumented rules - half compensating for data problems that were since fixed, half depended on by campaigns nobody remembers - is a real and common artifact. Three disciplines prevent it. The repair test, applied at rule creation: does this rule correct wrong source data (belongs in the catalog - fix it there, where every channel benefits) or reformat correct data for a channel (belongs here)? Hold the line; every violation is debt. Documentation as you go: every rule gets one line - what it does, why it exists, when added. Feed tools have description fields; use them. Quarterly rule review: delete what's obsolete, flag repairs that should migrate to the catalog, confirm the lookup tables still match reality. Thirty minutes, same calendar slot as your other quarterly audits - by now you have a standing date.

The layer's job, said once more precisely: adapt truth, never manufacture it. Source data true in the catalog, dialects handled in rules, and the whole thing legible to whoever inherits it. That's a feed setup that scales.