Cortex — support copilot

Agentic assist for tier-1 support—grounded answers, suggested replies, and escalation hints.

OpenAIVector DBZendesk

Cortex — support copilot product preview

Highlights

Retrieval over curated docs with freshness scoring
Human-in-the-loop feedback loops to improve drafts
Escalation policies based on sentiment and topic
PII redaction pipelines before model calls
Latency budgets for real-time chat assist

Overview

Cortex sits beside Zendesk/Intercom: it retrieves from your knowledge base, drafts replies in brand voice, and flags when a human should take over. Every suggestion cites sources; toxic or off-policy outputs are blocked by guardrails tuned per account. Managers get weekly quality dashboards—not just CSAT vanity metrics.

At a glance

Snapshot signals we optimize for on this engagement—useful for stakeholders comparing fit and scope.

Stack: RAG · LLM · Helpdesk
Safety: Citations · PII
Mode: Suggest + draft

How we delivered

Phased cadence—discovery through hardening—so expectations stay aligned with risk and learning cycles.

Content audit
Normalized KB articles and deprecation tags.
Pilot
Shadow mode with agent scoring.
Assist
Live suggest with rollback switches.

Grounding

We bias toward abstention when evidence is weak—better a safe handoff than a confident wrong answer.

Operations

Playbooks define when macros vs freeform drafts are allowed per brand.

Measurement

Outcome metrics tie to handle time and re-open rate—not just thumbs up/down.

Screenshots

Additional UI captured in desktop framing—same product, more surface area.

Want something similar for your product? Start a conversation.

Zentisys

Loading experience…