FAQ
What is the fastest way to get value from Jardine AI?
Pick one high-volume support question, add the source content for it in Knowledge Library, validate it with realistic phrasing, and run it through one connected channel. That first reliable path gives your team a concrete baseline and reduces setup anxiety.
Do we need to configure everything before launch?
No. You can launch with a focused setup and expand in stages. In practice, teams move faster when they make one flow dependable first, then add more channels, routing controls, and connector-backed use cases over time.
Is multiple workspace switching required for setup?
Not in the current product behavior. Organizations are operating in single-workspace mode right now, so most teams should treat setup as one active support environment.
What should we upload first in Knowledge Library?
Upload the material your support team already trusts in live conversations: billing policies, account policies, product usage FAQs, and troubleshooting guides. Avoid broad internal notes that were never written for customer-facing clarity.
How do we know a document is actually usable by the system?
Check ingestion status. A document that is still processing or failed will not provide dependable grounding. Active indexed content is what you can trust for answer behavior.
When should we use Knowledge Validate?
Use it before launch, before major updates, and whenever behavior feels uncertain. Advanced validation is especially useful because it lets you inspect evidence, rate responses, capture corrections, and simulate expected outcomes.
Why do answers look generic even when documents exist?
Usually because source content is vague, stale, or conflicting. The fix is typically document quality, not more routing rules. Improve source clarity first, then retest the exact scenario.
Do we need connectors from day one?
Only if your first workflow depends on account-specific data. If your early flows are policy or product questions, documentation grounding alone is often enough. Add connectors when you need runtime facts like customer status or subscription context.
Which connector providers are available in the current app?
Postgres, MongoDB, Supabase mode, and Stripe are supported in the Data Connectors interface.
What is the difference between automatic and manual routing?
Automatic routing gives sensible default behavior and is usually best early on. Manual routing exposes explicit policy controls through tags, destinations, and rules. Use manual controls when you need stricter, auditable handling behavior.
How should we design tags?
Keep the taxonomy small and behavior-driven. If two tags do not produce meaningfully different handling, they should probably be one tag. Overlapping tags make routing less predictable.
Why did a message escalate instead of getting an AI response?
Common reasons are weak grounding, policy conditions that require human handling, or missing runtime context for account-sensitive requests. Escalation is often the correct outcome when confidence or context is insufficient.
Is escalation a failure?
No. In support operations, escalation is a quality control mechanism. It protects trust in sensitive scenarios and keeps AI from forcing uncertain answers.
How do we test human handoff behavior?
Use internal test threads and inspect conversation detail. Confirm ownership transitions between AI and human are clear and timely, and confirm escalated threads reach the right destination.
What should we monitor after launch?
Watch conversation outcomes across layers: answer quality patterns, escalation correctness, channel reliability, connector health for account-aware paths, and unresolved-thread trends. Monitoring one metric alone can hide real issues.
Can we connect multiple channels at once?
Yes, but staged expansion is safer. Teams usually get better outcomes by stabilizing one channel first, then adding others once baseline behavior is predictable.
We connected Zendesk or Intercom, but conversations are not appearing. What now?
Treat this as an end-to-end channel setup check. OAuth status is only one part. Validate webhook setup, secret configuration where required, and trigger/event flow so inbound messages can actually enter conversations.
How can we reduce risk during rollout?
Roll out in stages: internal traffic, small canary, then broader release. Use validation before each major expansion. If behavior becomes uncertain, tighten scope instead of broadening it.
Who should own Jardine day to day?
The strongest model is shared ownership: support operations owns behavior expectations, and engineering supports integration reliability and advanced configuration when needed. Neither group should run blind without the other.
What is the right response when we are unsure about automation behavior?
Route sensitive or uncertain cases to humans, investigate by layer, and apply one focused fix at a time. Safety and clarity beat aggressive automation during uncertainty.
Where should we go next after this FAQ?
If you want practical launch flow, read Setup Guide. If you want operational scenario patterns, read Key Workflows. If you want deeper capability mapping, read Feature Deep Dive.