Home / Patterns / Deflection of Unresolved Issues

Deflection of Unresolved Issues

Goal Self Service Deflection Frequency Very Common Category Support Services Published View source on GitHub ↗

Issue: Self-Service Agent Marks a Conversation as “Deflected” (Resolved Without Human Agent) Based on the Customer Not Replying, Rather Than the Issue Actually Being Resolved

Frequency: Very Common

Symptoms

  • Conversation is logged as successfully self-service-resolved whenever the customer does not send a follow-up message within a timeout window, regardless of whether the bot’s last response actually answered the question
  • Deflection rate (a key cost-saving metric for self-service tooling) looks strong while actual issue resolution, measured independently, is lower
  • Customers who give up due to frustration (bot loop, unhelpful answer) are counted identically to customers whose issue was genuinely resolved, since both produce the same “no further reply” signal
  • Customers who abandon self-service and contact support through a different channel shortly after are not linked back to the “deflected” conversation, so the false deflection is never corrected in the metric

Root Cause Deflection is operationally defined and measured as “conversation ended without escalation to a human agent,” which is a behavioral proxy for resolution, not a direct measurement of it. Silence after a bot’s response is consistent with both successful resolution and customer abandonment from frustration, and without an explicit confirmation signal or downstream tracking to detect re-contact through another channel, the metric cannot distinguish between the two — it structurally rewards driving customers away as much as it rewards actually helping them.

Example

Scenario: Customer asks self-service bot about a billing discrepancy
Bot: Provides a generic FAQ answer that does not address the specific discrepancy
Customer: Does not reply further (gives up, frustrated), closes the chat window
System: Logs conversation as "deflected" (successful self-service resolution)
Customer: Contacts support via phone the next day about the same unresolved billing issue
Linkage: Phone contact not connected back to the original "deflected" chat in metrics
Impact: Deflection rate is overstated; actual resolution rate and customer effort are misrepresented

Key Statistics

  • Industry research on self-service and chatbot deflection metrics consistently flags the gap between “deflected” and “actually resolved” as a major measurement validity concern
  • Cross-channel re-contact within a short window after a “deflected” self-service conversation is a recommended proxy for detecting false deflections, per customer experience measurement practice
  • Explicit resolution confirmation (asking the customer to confirm their issue was resolved) has been shown to produce meaningfully lower apparent deflection rates than silence-based deflection counting, revealing the gap between the two measures

Mitigation Strategies

  1. Explicit Resolution Confirmation: Require an explicit “was this resolved?” confirmation step before counting a conversation as deflected, rather than inferring resolution from the absence of a follow-up message
  2. Cross-Channel Re-Contact Linking: Track whether a customer re-contacts through any channel within a defined window after a “deflected” conversation on the same topic, and retroactively reclassify linked conversations as false deflections
  3. Resolution Rate as the Primary Metric, Deflection as Secondary: Report actual confirmed-resolution rate alongside deflection rate, and treat deflection-without-confirmation as a lower-confidence, secondary signal
  4. Bot-Loop and Abandonment Detection: Detect behavioral signals of frustration (repeated rephrasing of the same question, abrupt mid-flow exit) and exclude those conversations from the deflected/resolved count by default

Metrics

  • Confirmed resolution rate vs. silence-based deflection rate (the gap between them)
  • Cross-channel re-contact rate within a defined window after a “deflected” conversation
  • Bot-loop/abandonment-pattern detection rate among conversations counted as deflected

Alerts

  • Cross-channel re-contact rate for “deflected” conversations on the same topic exceeds a defined threshold → P2
  • Gap between confirmed-resolution rate and silence-based deflection rate widens beyond a defined margin → P2

References

Related Patterns