Support Deflection with Escalation
Answer common issues instantly; route edge cases to the right place.
Outcome at a Glance
Who it’s for: Support, success, and operations teams. What you’ll achieve: High deflection on repeat questions (with citations ON), safe handoffs for exceptions, and a continuously improving support loop. Time to first win: 45–60 minutes.
Why it Matters
Faster answers, fewer tickets: Grounded responses resolve questions instantly and reduce ticket volume.
Safer boundaries: Clear “no answer” messaging and escalation flows preserve trust and compliance.
Continuous improvement: Weekly content revisions turn gaps into new documentation.
Delphi becomes your front line for support — answering what it can, flagging what it can’t, and learning from every case.
What You’ll Set Up
Mind — Upload policies, how-tos, and troubleshooting trees. Add Context and Publish Dates.
Mind Settings
Citations = Show: Links every answer to its verified source.
Message on No Answer: Defines safe fallback and escalation instructions.
Custom Instructions — Explain escalation rules: what Delphi should hand off, to whom, and via what channel.
Actions
Alert Triggered → Auto-response to policy-sensitive topics (refunds, outages, cancellations).
Routing Follow-ups → Tag or notify appropriate team members.
Alerts
Detect intent, compliance issues, or escalation signals (e.g., “urgent,” “not working,” “refund,” “breach”).
Integrations — Website Embed on Support/FAQ pages with Suggested Questions for self-service.
Build → Launch → Operate
Build (30–40 min)
Upload Core Support Docs:
Top 20 FAQs
Key policies (refunds, cancellations, privacy, billing)
Step-by-step troubleshooting guides Add Context and Publish Dates for each item.
Enable Citations & Safe Boundaries:
Turn Citations = Show in Mind Settings.
Add a Message on No Answer: “I don’t have enough information to answer that safely. I can help with [A/B/C]. If this is urgent or account-specific, please [escalation path].”
Define Custom Instructions:
Specify when to escalate (refunds, downtime, security).
Name who to notify and via which channel (Slack, email, Zendesk).
Set Up Alerts for Hot Topics:
“If user mentions ‘refund’ or ‘cancellation,’ tag as Refund Inquiry and trigger refund response.”
“If user says ‘site down,’ tag as Outage Alert and send status link.”
“If user mentions ‘urgent,’ tag as Escalation and notify ops.”
Configure Actions:
A. Alert Triggered → Send Message (User): Policy-based response with link.
B. Alert Triggered → Notify Team: Sends Slack/email alert for follow-up.
C. Conversation Ended → Review Tag: Track unresolved or handoff cases.
Launch (10–15 min)
Embed Delphi: Add on Support or FAQ pages with a short CTA: “Ask about policies, troubleshooting, or how-tos.”
Activate Alerts & Actions:
Test refund, outage, and urgent triggers.
Confirm policy responses include citations and links.
Verify “No Answer” Handling:
Ask a non-documented question to confirm Delphi shows fallback safely.
Tagging & Tracking:
Use tags like Refund Inquiry, Outage Alert, Escalation, Unhandled, and Resolved.
These power future reports and Broadcasts.
Within an hour, your Delphi becomes a reliable support triage assistant — handling FAQs instantly and routing edge cases to humans safely.
Operate (Weekly, 30 min)
Review Conversations: Filter by unresolved or long threads. Revise answers or create new documents to fill gaps.
Audit Alerts: Check which triggers fired most (e.g., Refund vs Login Issues). Adjust keywords or add context.
Update Docs: Refresh Publish Dates and upload new versions as policies evolve.
Track Trends: Use tag data to identify what users struggle with most and which answers drive deflection.
Recipes (Copy → Tailor)
Message on No Answer
“I don’t have enough information to answer that safely. I can help with [A/B/C]. If this is urgent or account-specific, please [escalation path].”
Alert-Triggered Policy Reply
Trigger: Alert Triggered (keyword: “refund”) Action: Send Message (User) Message: “Here’s our refund policy, including steps and timelines: [link]. If your case is unique, reply ‘refund exception’ and I’ll escalate.”
Outage Notice
Trigger: Alert Triggered (keyword: “down,” “outage”) Action: Send Message (User) Message: “We’re investigating a temporary service issue. You can check live updates here: [status page link]. If you’re fully blocked, reply ‘urgent’ and I’ll flag it for review.”
Escalation Action
Trigger: Alert Triggered (keyword: “urgent,” “breach,” “security”) Action: Notify (Team) Message: “Urgent escalation: user {name} flagged {keyword}. Check thread [link].”
Tagging Actions for Insight
Trigger: Alert Triggered (any policy keyword) Action: Add Tag (User) Tags to use: Refund Inquiry, Billing Question, Outage Alert, Escalation, Unhandled Case.
This tagging system lets you track recurring issues, segment users for follow-ups, and target improvements in documentation.
Metrics to Watch
Deflection Rate: % of support questions answered without ticket creation.
Unanswered → Revised Trend: Decline indicates documentation improving.
Alert Volume by Topic: Top drivers of user support demand.
Time-to-First Response: Should remain near-instant for FAQs.
Escalation Accuracy: % of alerts correctly routed to humans.
Troubleshooting
Users still open tickets: Strengthen first-line responses; make escalation path clear; ensure visible citations.
Policy confusion: Simplify copy; add examples; update Publish Date for clarity.
Too many alerts: Narrow keywords; add cooldown window (e.g., one trigger per session).
Missed escalations: Check Custom Instructions; ensure Alert → Notify mapping is correct.
Pre-Flight Checklist
Policies, how-tos, and troubleshooting guides uploaded with Context + Publish Date.
Citations = ON and Message on No Answer configured.
Custom Instructions for escalation set (who, how, when).
Alert Triggered Actions active for hot topics (refund, outage, urgent).
Website Embed live on Support/FAQ pages.
Weekly Review & Revise slot scheduled.
In Practice
Your Delphi becomes a high-trust support layer — grounded in real policy, ready to triage exceptions, and constantly improving through feedback loops. Every alert becomes both an action and a learning opportunity.
“If someone mentions a refund, tag them as Refund Inquiry and send the policy instantly. If they say ‘exception,’ escalate automatically.” That’s Delphi running your support frontline — fast, consistent, and always learning.
Last updated