Definition
Percentage of chatbot-deflected interactions where the customer's issue was genuinely resolved, separating true resolution from deflection that resurfaces through repeat contacts, channel switching, or defection.
The Measurement Gap
AI chatbots handle an increasing share of customer service interactions. The headline metric is deflection rate: interactions handled without a human agent. But deflection is not resolution. A customer who asks for help, receives a generic FAQ link, and gives up has been deflected, not resolved. A customer told the issue is "resolved" but who calls the phone line the next day was deflected from one channel to another. A customer who gets nowhere and switches retailers has been deflected into competitive defection. Most retailers track first-contact resolution (FCR), which counts any interaction the AI closes without escalation as resolved. TRR-R subtracts the evidence of non-resolution: tickets reopened within 14 days, duplicate tickets on the same topic, channel switches (chat to phone to store visit), and auto-closed tickets without customer confirmation. The gap between FCR and TRR-R is the "deflection illusion," often the number the service director suspects but cannot prove.
Decision Supported:
Is the AI customer service system genuinely resolving issues, or is it deflecting effort while customers find workarounds, give up, or switch retailers?
Formula
Formula:
TRR-R = (AI_resolved - Reopened - Duplicate - Channel_switch - Auto_closed) / AI_resolved
Variables:
- AI_resolved = interactions marked resolved by AI without human involvement
- Reopened = interactions reopened within 14 days
- Duplicate = new interactions from same customer on same topic within 14 days
- Channel_switch = same customer contacts a different channel (phone, email, store) on same topic within 14 days
- Auto_closed = interactions closed by timeout without explicit customer confirmation
Units: Percentage. Compare to deflection rate and FCR. The gap between FCR and TRR-R is the deflection illusion.
Edge Cases and Exclusions:
Same customer with a different topic should not count as a duplicate. This requires topic matching at the category level, not keyword level, to avoid overcounting. If the customer resolves the issue themselves after the AI interaction (for example, the AI pointed them to an FAQ and they successfully self-served), that IS resolution and should not be subtracted. Only subtract evidence of non-resolution. In-store follow-up requires cross-channel customer identity matching, which is difficult. Use loyalty or account ID where available. Require a minimum of 200 AI-resolved interactions per measurement period.
Proposed formula, pending validation pilots.