Recommendations Activity Stalls After Catalog Feed Change
A retailer runs a Recommendations activity using the 'People Who Viewed This, Viewed That' algorithm on product detail pages. The activity has been live for 6 weeks with a 12% lift in revenue per visitor. Last Tuesday, the merchandising team switched the product catalog feed from a nightly CSV upload to an API-based feed that updates every 15 minutes. Since then, the Recommendations activity has shown zero recommendations for 40% of product pages, and overall click-through rate on recommendations dropped from 3.2% to 1.1%. The feed contains 50,000 SKUs, but only 35,000 have complete 'entity' attributes (name, image, price, category). The remaining 15,000 are missing the 'category' attribute. The API feed includes a 'last_updated' timestamp, but the Target implementation team did not configure the feed to include the required 'entity.id' and 'entity.name' fields in the correct format. The activity uses a default include rule that excludes products with no category. The QA team tested only 10 product pages after the feed change, all of which had complete attributes. The business wants to restore performance within 48 hours.
1. What is the most likely root cause of the zero recommendations on 40% of product pages, and what immediate diagnostic steps should the architect take?
Read the worked answer
The most likely root cause is that the new API feed does not correctly populate required entity attributes for all products, specifically the 'category' attribute missing on 15,000 SKUs. Because the activity's include rule excludes products with no category, those products are ineligible to appear as recommendations. Additionally, the feed may not be formatting 'entity.id' and 'entity.name' correctly, causing Target to fail to ingest those entities entirely. Diagnostic steps: 1) Download the feed and validate that all 50,000 SKUs have 'entity.id', 'entity.name', and 'category' in the exact case and format required by Target (e.g., 'entity.id' must be a string, 'category' must match the taxonomy used in the include rule). 2) Check the Recommendations feed status in Target (Recommendations > Feeds) for ingestion errors or warnings. 3) Use the Target Recommendations API or the feed diagnostics to count how many entities were successfully indexed. 4) Compare the list of product pages showing zero recommendations against the feed to confirm they correspond to SKUs missing 'category'. 5) Temporarily relax the include rule to not require 'category' to see if recommendations appear, confirming the hypothesis.
2. What remediation steps should the architect take to restore performance within 48 hours, and what longer-term process changes are needed to prevent recurrence?
Read the worked answer
Immediate remediation (within 48 hours): 1) Fix the feed to include 'category' for all SKUs-either by enriching the API response or by adding a default category value for missing ones. 2) Correct the formatting of 'entity.id' and 'entity.name' to match Target's requirements. 3) Re-ingest the feed and verify that all 50,000 SKUs are indexed. 4) Temporarily modify the activity's include rule to not exclude products without category, or set a fallback category, to restore recommendations immediately while the feed is fixed. 5) Monitor the feed ingestion and recommendation coverage. Longer-term: 1) Implement automated feed validation that checks for required attributes and formatting before ingestion, with alerts. 2) Expand QA to sample product pages across all attribute completeness levels, not just complete ones. 3) Establish a change management process for feed changes that includes Target architect review. 4) Document the required feed schema and enforce it in the API contract. 5) Set up monitoring for recommendation coverage and CTR anomalies.