Daily workflows

Check Metrics

Start here every morning. Open Metrics → Search Metrics and look at the Summary tab. You're checking for anything that moved unexpectedly since yesterday.

The two numbers to check first:

Zero-result rate — should be close to zero. Anything above 1–2% means shoppers are searching for things the catalog can't answer. If it spiked overnight, something changed — a new product was announced, a category went live, or a term people are suddenly searching for isn't in your vocabulary.

View rate and conversion rate — if these dropped without a corresponding drop in search volume, results are getting worse for some queries. The Queries tab will tell you which ones.

If the Summary looks healthy, a quick scan of Trending Queries is worth 30 seconds. Queries gaining momentum are opportunities — if a trending term isn't well-served by the current catalog, you want to know before it peaks, not after.

If something looks off: switch to the Queries tab and sort by zero-result rate descending. The queries at the top are actively failing shoppers. Note them — you'll deal with them in Step 3.

Work the Review queue

Open Attribute Enrichment → Review & Publish and go to the Needs review tab. This is the queue of AI-generated attribute values that didn't meet the auto-publish confidence threshold — they're waiting for your sign-off before going live.

On most mornings this queue is manageable. The AI handles high-confidence values automatically overnight. What's left for you is the borderline cases.

A practical way to work through it:

• Filter by Product Type to focus on one category at a time — faster to review 40 color values for jackets than to jump between unrelated product types.
• Sort by Confidence: High to Low. The nearly-certain values at the top are usually quick approvals. Work those first before spending time on the genuinely uncertain ones.
• For each row, check the AI-generated value against the original. If it looks right, Approve. If it's wrong, Revert — or click the edit icon to correct the value before approving.

If you spot a pattern — the AI keeps getting a specific attribute wrong in a particular product type — that's feedback for the enrichment configuration, not something to fix one row at a time.

Before closing the tab, flip to Auto-applied and skim what went live automatically. If you catch a bad auto-applied value, hit Revert and it's undone immediately.

Act on what you find

This is where you close the loop on whatever Metrics flagged in Step 1. Two tools cover most situations.

Zero-result queries → Linguistic Override

If a query is returning zero results, the most common cause is a vocabulary gap: shoppers are using a term that doesn't appear in your product data. Open Linguistic Overrides and add a synonym rule mapping the failing term to a term your catalog uses.

For example: if "softshell jacket" is returning zero results but your products are tagged "soft shell", a two-way synonym rule fixes it instantly — no re-indexing needed.

If it's a brand-new product name — something announced in the news or on social media — shoppers may be searching for it before it's in the catalog. A synonym rule bridges the gap in the short term.

Poor results for known queries → Discovery Rules

If Metrics shows a query with decent volume but low view or conversion rates, the ranking is probably the problem. Create a Boost rule to surface the right products for that query, or a Bury rule to push down products that are consistently misleading shoppers.

This is also where you manage anything business-driven: a new collection that needs visibility before it has engagement data, a promotion that should dominate specific queries for a campaign window, or out-of-stock products that keep appearing at the top of results.

Rules take effect immediately on activation and can be deactivated just as quickly.

Verify enrichment runs

Open Attribute Enrichment → Enrichment Runs and check that the overnight run completed. You're looking for a recent entry with a Success status and a sensible Updated count.

If everything looks normal, you're done. If something's off:

• Failed status — click into the run and read the error message. Common causes are model unavailability (transient — re-trigger the run) or a data format issue (needs investigation before re-triggering). A failed run doesn't affect previously published values, so the catalog is still in a clean state.
• Low Updated count — if far fewer products were updated than usual, the catalog snapshot may have been smaller than expected, or enrichment is hitting the same products repeatedly without finding new values to add.
• Run didn't appear — if there's no run from last night, the scheduled trigger may have been missed. Trigger a manual run and flag it with your engineering team.

If a specific product is missing an attribute you expected to see, check here first. Find the relevant run, confirm the product's type was included, then check Review & Publish — the value may be sitting in the Needs review queue rather than missing entirely.