Skip to main content
Use Insights to get a daily, plain-language read on what is moving in your storefront. Each card describes a change Layers detected in recent search and purchase activity, why it matters, and what you can do about it. Insights are useful when you want a single place to scan for movement without building reports.

​
Prerequisites

  • Your store must be connected to Layers.
  • Your catalog should be synced.
  • A few days of shopper search and purchase activity must exist so Layers has a baseline to compare against.

​
Open Insights

  1. In the Layers dashboard, open Lab.
  2. Select Insights.
The top insight also appears in the Insights carousel at the top of the Home page. Use the prev and next controls to page through additional insights, or select View all insights to jump to the full feed. The Home page shows a single compact rollup of AI curation activity — for example, “AI curated N searches in the last 30 days” — that links to the Analytics surface, rather than individual per-decision cards. The detailed curation opportunity cards remain on this Insights page.

​
How the feed is organized

Insights are grouped by category. Categories only appear when Layers has at least one insight to show for them.
  • Trends — Search demand moving up or down. Use these to spot terms gaining or losing interest before they show up in headline metrics.
  • Attribution — Revenue influenced by search and discovery. Use these to see which queries, collections, or products are driving the biggest share of search-led revenue. Attribution cards highlight wins, so a window where attributed revenue dropped is suppressed in favor of the up window for the same subject. Genuine revenue declines surface as anomalies instead.
  • Anomalies — Sharp breaks from your recent baseline, such as a spike in zero-result searches or a sudden drop in click-through. Use these to investigate issues before they affect more shoppers.
  • Opportunities — Demand you are not fully capturing yet, such as high-traffic queries with low engagement. This category also holds sort health insights that flag buried risers and mis-windowed sort orders.
  • Sequence suggestions — Signals about your product sequences based on how often their members are actually purchased together. Layers surfaces two kinds of sequence cards:
    • Substitute — Members of an existing sequence rarely appear in the same order and one product clearly dominates the others in sales. Shoppers are treating them as alternatives, not a set. Consider splitting the sequence or reordering it so the leader is first. Layers only raises this card when the dominance gap is large and the trailing product has enough orders to be a reliable sample. A normal 2:1 popularity split between styled-together items will not trigger it.
    • Suggestion — Two products are frequently purchased together but are not in any sequence yet. Consider creating one from the sequences page.
    Non-merchandise items (for example, tip, gift-wrap, and shipping-protection line items) are excluded so pair suggestions stay focused on real products. Pair titles are sorted alphabetically, so “A & B” and “B & A” collapse to a single card.
  • What Layers did — A collapsed section that lists AI curation decisions applied to broad queries. When AI reranking reorders a high-volume query’s front page, a card lists how many products were promoted or demoted, the rationale, and the specific products that were demoted. Expand the section to review the demoted products and retag any that were misclassified. Cards only appear here when a demotion occurred or Layers recorded a rationale for the decision.
If the same subject — for example a query term like “anklet” — would otherwise produce several near-duplicate cards, Layers collapses them to the single highest-signal card so the feed stays scannable. Conversion-rate “converts at” cards are also capped per page so a single class of insight cannot crowd out the others. Each category has its own pagination controls beneath its cards. Paging through Trends does not affect Opportunities, Anomalies, or any other section. You can dig into older cards for one category while keeping the others on their first page.

​
What each card shows

Every insight card includes:
  • A category badge — Trend, Attribution, Anomaly, or Opportunity.
  • A percent change versus the baseline. Movement is colored green when up is good, red when down is bad, and amber for anomalies.
  • A timeframe row with the time window the insight covers (for example, 7d), the period label, and the date Layers detected the change.
  • A headline describing what changed.
  • The subject the insight applies to, such as a query cluster, collection, or product group.
  • The current value for the underlying metric.
  • Comparison bars that show the current value next to the baseline value so you can see the size of the gap at a glance.
  • A short explanation of what changed and why it matters.
  • A recommended action when Layers has one, such as reviewing a query, adjusting a sort order, or adding a merchandising rule.

​
Anomaly review

When Layers detects a store-wide drop in click-through or engagement, it automatically re-checks the search results for the query patterns most likely to be dragging the metric down. Each anomaly card then reports one of two states:
  • Reviewed healthy — Layers re-ran the affected searches and confirmed the results look right. The drop is probably driven by external factors (seasonality, traffic mix, a specific campaign) rather than a search-quality problem.
  • Needs merchant action — Layers found the results for one or more affected query patterns look off. The card surfaces the specific clusters and products to review so you can tune merchandising, sort orders, or product data.
This review runs in the background once per day and never changes live ranking on its own. It only flags where to look.

​
Sort health insights

Layers checks your published, metric-first sort orders every day and raises cards when the ranking is likely to be missing recent movement. These cards appear in Opportunities and Anomalies and always target a specific sort order or product, so the recommended action is a concrete change instead of a general note. Three subtypes are worth calling out:
  • Buried risers (Opportunity) — A product is in the top 20 by units over the last 3 days but sits below position 40 or is unranked in a published sort order. This usually means the sort order’s window is too long to catch the pickup, or the product is losing to a hard tiebreaker.
  • Product spike (Anomaly) — Yesterday’s units for a product jumped well above its own 28-day baseline. Use it to decide whether to promote the product in your active sort orders while the spike lasts, and to check for a stock or content issue if the spike is unexpected.
  • Window mismatch (Opportunity) — The sales window configured on a sort order’s primary metric trails the window Layers found actually predicts next week’s sales best for your store. For example, a sort order using a 30-day window when a 7-day or trend-weighted window wins the backtest by a clear margin.
Layers runs a per-store sort window backtest once a week that compares trailing 7/14/30/60-day windows against 7-day and 14-day half-life decayed windows. When a short-horizon window wins, the Trending sort recipe and the Trending measures in the metric picker unlock, and the window-mismatch insight can recommend switching to them. Stores without enough recent purchase history for the backtest to be reliable don’t get these insights.

​
Apply an insight to a sort order

Sort health cards carry one or two action buttons under the recommended action. Selecting an action opens the target sort order in the editor with the change already staged. Nothing on the storefront changes until you publish. Two actions are available:
  • Boost recent risers — Adds a soft boost to the sort order that lifts products in the top 10% by units over the last 3 days. Best when a specific product is buried and you want to catch other risers at the same time.
  • Switch metric — Swaps the sort order’s primary metric to a Layers recipe with a better-fitting window (for example, the Trending recipe or a shorter trailing-units window). The Switch to trending variant only appears when the sort window backtest supports a short-horizon window for your store.
When you open the editor from an insight action, a Prefilled from an insight banner appears above the sort form. The banner reminds you the change is staged and not live yet. Review the staged expressions, tweak anything you want to adjust, then publish through the normal review flow. Dismiss the banner to hide it for the current session; the staged changes stay in the form either way.

​
Troubleshooting

​
The feed is empty

Layers needs several days of search and purchase activity before insights start appearing. If your store is new or recently re-launched, check back after more traffic has accumulated.

​
A card I expected is missing

Insights are filtered to keep the feed high-signal. Movements below the configured thresholds — for example, very low search volume or a small percent change — are intentionally hidden. Cards also drop off the feed once the underlying signal returns to baseline. If you expected several cards for the same query term, note that Layers collapses near-duplicates to the strongest card per subject.

​
An anomaly stays in Needs merchant action

The state updates after the next daily review. If the affected query patterns still look weak the next day, the card stays flagged until either the results improve or the anomaly itself clears.

​
Next steps