Core Concepts

If Jardine AI ever feels confusing, it is usually not because the product is complicated. It is because support teams naturally think in tools, while Jardine is built around decisions. Once you shift from “which feature do I click?” to “how does this conversation get decided?”, everything starts to click.

A useful mental model is this: Jardine sits between incoming support messages and your team, then decides the safest next action for each conversation. Sometimes that action is an AI response. Sometimes it is escalation. Sometimes it is continued ownership by a human. What matters is that the system makes those decisions using context you control.

How Jardine Decides

Every conversation passes through a decision loop, whether the message came from Email, Intercom, or Zendesk.

First, a message enters through a connected channel. Channel type matters for ingestion and delivery details, but not for the core decision logic. That shared decision model is why teams can keep support behavior consistent even when they run multiple channels.

Second, Jardine classifies the request and gathers context. Context can come from two places. The first is knowledge: the documents you upload and keep current in the library. The second is runtime data: connector-backed facts used for account-specific questions.

Third, Jardine evaluates whether the response path is trustworthy enough. If grounding is strong and policy conditions are satisfied, AI can respond. If risk is higher, context is missing, or policy requires a person, the conversation is escalated or handed over.

Finally, the result is visible in operations. You can inspect conversations, see status and ownership, review confidence and intent signals, and move ownership between AI and human when needed.

That loop is the center of the product.

Knowledge, Routing, and Ownership

The three concepts that drive most outcomes are knowledge quality, routing policy, and ownership state.

Knowledge quality answers the question: “Can Jardine say something accurate here?” If your documents are clear, current, and customer-facing, answer quality rises quickly. If content is stale or contradictory, you will see uncertainty, weak replies, or unnecessary escalations. This is why high-performing teams treat the library as an operational asset, not a storage folder.

Routing policy answers the question: “What should happen after intent is understood?” In the current interface, routing is automatic by default. If you need stronger control, you can switch to manual controls and define tags, destinations, and rules. Tags classify intent, destinations define where escalated work goes, and rules connect the two with optional conditions like confidence thresholds or message patterns.

Ownership state answers the question: “Who is actively handling this thread right now?” In conversation detail, you can explicitly assign a thread to AI or to a human. This is useful for sensitive edge cases, coaching, and cleanup when a conversation changes direction.

When these three are aligned, the system feels predictable. When they are misaligned, outcomes feel random even though each component is technically working.

Why Escalation Is Part of Good Automation

Many teams assume escalation means failure. In support operations, the opposite is often true. Escalation is what protects trust when a request is sensitive, ambiguous, or under-contextualized.

A billing dispute, account lockout, or emotionally escalated customer message may not be the right place for fully automated handling. Jardine should route those moments cleanly to humans when policy or confidence says it should. That behavior is not a fallback; it is part of correct operation.

The real goal is not maximum automation. The real goal is reliable resolution. Reliable resolution often includes a healthy mix of AI responses and intentional handoff.

Validation and Analysis as Feedback Loops

Another concept that matters early is the difference between setup and learning.

Setup is where you connect channels, add knowledge, and configure routing. Learning is where you validate behavior, rate responses, inspect evidence, run suites, and watch analysis signals over time. In Jardine, both are first-class.

Validation gives you controlled testing before broader rollout. You can run playground chats, mark responses as correct or wrong, add correction notes, and simulate expected outcomes for intents, route types, and template keys. This lets teams tune behavior before customers absorb the mistakes.

Analysis gives you health signals on the knowledge side: indexed ratio, failed documents, collection-level coverage, and recommendations. It helps you detect quality drift that starts in ingestion, long before it shows up as obvious conversation failures.

Together, these tools form a continuous improvement loop. That loop is how teams move from “it works sometimes” to “it works reliably under pressure.”

Scope and Team Model

In the current product state, organizations are operating in single-workspace mode. You will still see workspace concepts in role and access flows, but day-to-day setup should be approached as one active support environment. Team management happens in People, where roles control who can invite members and change sensitive settings.

That model is helpful because it reduces early operational complexity. You can focus on getting one system reliable before introducing broader structural concerns.

When you approach Jardine with these concepts in mind, individual features stop feeling disconnected. Channels are ingress paths. Knowledge provides grounded context. Connectors add runtime facts. Routing applies policy. Conversations expose outcomes. Validation and analysis keep quality improving.

If this mental model feels clear, continue with Setup Guide to translate these concepts into a practical production setup sequence.