Suggested Responses Ignored in Email Channel
A retailer handles 2,100 customer emails per day in Pega Customer Service 23.1. The email case type uses the standard email composer with a suggested responses panel. There are 1,240 published knowledge articles. Only 38 articles are marked Suggested; the rest are Public. The article taxonomy has three levels: Department, Product, Issue. Agents handle 45 emails per day each; average handle time is 14 minutes against an 8-minute target. Click-through on suggested responses is 7%, and 62% of agents say the suggestions are irrelevant. The top five email intents are delivery delay, return label, damaged item, password reset, and warranty claim. Those five intents account for 58% of volume. There is no feedback capture on suggestions, and the knowledge team publishes ad hoc.
1. Diagnose why suggestion relevance is low, using the configuration facts.
Read the worked answer
The suggestion engine is not seeing most content. Only 38 of 1,240 articles are marked Suggested, so the other 1,202 are effectively invisible to the panel even if they are published and Public. The five intents that drive 58% of volume may have no Suggested article at all. Relevance also depends on context: the article taxonomy is Department, Product, Issue, but the email case must pass product and issue properties for ranking; if it does not, the panel falls back to text matching, which explains the 7% click-through.
Check the audience setting as well; an article marked for customers but not for CSRs will not appear to agents. Finally, the absence of feedback means no one can see which suggestions were shown, accepted, or rejected, so the knowledge team is publishing without a signal. The 14-minute AHT is consistent with agents searching manually after rejecting suggestions.
2. Design a 30-day improvement plan with measurable targets.
Read the worked answer
Start with the five intents, not the whole taxonomy. For each, identify the top three products and publish a Suggested article that includes the agent action, the customer-facing wording, and any verification step. Target 15 Suggested articles, not 1,240. Map case properties Product and IssueCategory to the article's relevant context so ranking uses the case, not just keywords. Add thumbs-up and thumbs-down feedback on the panel and capture not-relevant reasons: wrong product, outdated policy, missing step.
Set targets: suggestion click-through 35%, AHT 10 minutes, and 80% of the five intents covered by at least one Suggested article within 30 days. Review search terms weekly and retire or rewrite articles with high impressions and low acceptance. This is a content and context fix, not a UI redesign.
3. What governance would prevent the same decay from recurring?
Read the worked answer
Assign an owner per department and a monthly review for the top 20 intents. Define Suggested criteria: current, agent-facing, product-specific, and tested against a live case. Require a taxonomy value for Department, Product, and Issue before publish. Use knowledge reports for impressions, acceptance, and search-with-no-result, and feed them into the monthly review.
Add a pre-publish check that the article's audience includes Customer Service and that the product and issue properties match. Keep a small set of canned responses for high-risk steps such as password reset and warranty claim, where free text is not appropriate. Without this loop, the same 38 Suggested articles will decay and the 7% will return.