Crachá de Contexto and Bandeja de Contexto: The Difference That Prevents the Reset

The biggest mistake in conversational AI initiatives isn't "getting the answer wrong." It's passing the baton poorly. In most companies, the handoff is the most expensive point of the cycle—because that is where context is lost. The customer who spent 15 minutes chatting with a bot and is "transferred to an agent" who asks "how can I help you?" has just experienced a reset: the invested time was wasted and trust was restarted.

MCI solves this with two artifacts that complement each other—and should not be confused.

The Crachá de Contexto: The Traveling Record

The Crachá de Contexto is the minimum data contract that travels with the cycle across departments and systems. It is the persistent record—it always exists throughout the entire life of the cycle. It has seven canonical fields:

• Origin — where it came from, which entry channel.
• Channel — current and historical.
• Intent — with confidence level.
• Pain/Context — the real constraint that moves or stalls the decision.
• Cycle State — where it is in the graph.
• Summary — a decision-making synthesis, not a message dump.
• Next Step — action + SLA.

Without a standardized Crachá, "passing the cycle" becomes "sending a Slack message." With a Crachá, the crossing has a contract—and a contract can be audited.

The Bandeja de Contexto: Presentation at the Moment of Decision

The Bandeja de Contexto is the actionable presentation of the Crachá for a specific moment of decision—when a human takes over. It is generated from the Crachá during the handoff. It has six components:

• Current State and Evidence — where the decision stands and which signals support this reading. Not "what the customer said," but "what the system understands they want," with a level of certainty.
• Dominant Blocker — the system's hypothesis on what is preventing progress. This is the most valuable contribution of AI because it connects signals that a human would rarely link in time.
• Summarized History — the 3 to 5 most relevant pieces of information, not "every message."
• Recommended Next Step, with Options — not a rigid script. The Bandeja suggests; the human decides.
• Current SLA — at risk, expired, and what has already been triggered.
• Recent State Change — where the decision came from and what changed.

The AI doesn't need to announce "I'm going to transfer you to an agent." The ideal handoff is silent: the human takes over with context in seconds, and the customer feels continuity, not a "department change." In practice, the Bandeja changes the human's first sentence. Instead of "how can I help you?" (reset), it becomes "I see the question regarding the ERP integration remained open—I've brought in our specialist to resolve it now" (continuity). From a question to a contextual statement: that is the operational difference between reset and trust.

The Quality of the Bandeja: 3 Criteria

MCI Ops monitors three criteria. Completeness — are the six components present? If the blocker hypothesis is empty, the human loses the primary benefit. Accuracy — does the reading match reality? If the AI says "budget blocker" and the human discovers in 30 seconds it was "decision-maker on vacation," the Bandeja failed and needs recalibration. Actionability — can the human act without additional investigation? The goal is to read, confirm, and act in less than 2 minutes.

Anti-patterns

• History Dumping. Sending "all messages since the first contact" is not context—it's noise. The Bandeja is distillation, not an archive.
• Generic Recommendation. "Suggestion: follow up" describes the obvious. The recommendation needs to be contextual and specific.
• Static Bandeja. If it is generated once and never updated, it ages quickly. It needs to be dynamic—updated with every relevant interaction.