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

The biggest mistake in AI initiatives is not "getting the answer wrong." It’s a poor handoff. In most companies, the overflow — the moment the conversation moves from bot to human, or from one department to another — is the most expensive point in the cycle. Not because it costs a lot of money per se, but because that is where the context is lost.

Everyone has experienced this: you talk for 15 minutes with a chatbot, explain your situation, provide your ID number, describe the problem, and then receive that phrase that throws everything away — "I'm going to transfer you to an agent." Seconds later, the human agent opens the chat with "Hello, how can I help you?". Reset. The time invested turned into trash. Trust, restarted from zero.

MCI (Marketing Conversacional Integrado) solves this with two artifacts that complement each other — and must not be confused: the Crachá de Contexto and the Bandeja de Contexto. This text is the definitive reference on them: what they are, what they are for, which fields compose them, how they relate, and what differentiates them in practice.

Why most companies fail at the handoff

Before diving into the artifacts, it is necessary to understand the problem they solve.

When a conversation changes channels (WhatsApp → phone), systems (bot → CRM), or people (SDR → closer, customer service → retention), three things almost always happen:

1. History doesn't travel in a useful way. Either it simply isn't passed along, or it is dumped entirely into the next person's lap — 300 messages that no one will read.
2. Emotional context is lost. The next system doesn't know if the customer is calm, frustrated, decisive, or hesitant. It treats everyone the same.
3. The next step needs to be reinvented. Since the new agent doesn't know where the conversation left off, they restart: "tell me what happened."

The result is the reset. And a reset is expensive: it kills NPS, increases average handle time, forces the customer to repeat information they already gave (which they hate), and, most importantly, breaks the perception that there is a company behind the channels — there simply becomes disconnected sectors.

The Crachá de Contexto and the Bandeja de Contexto exist to eliminate the reset. One ensures that context always exists during the cycle; the other ensures that it arrives in the right format at the moment of decision.

The Crachá de Contexto: the record that travels with the conversation

The Crachá de Contexto is the minimum data contract that accompanies a conversation cycle from start to finish. It exists the entire time, is updated at every relevant turn, and travels between channels, systems, and people. It is the persistent record.

A useful analogy: think of the ID badge a patient receives when entering a hospital. It stays with the patient throughout their stay, is read by every professional who treats them (reception, triage, nursing, doctor, pharmacy, discharge), and concentrates everything the next person needs to know to act without having to ask the same questions again. This is the role of the Crachá in a commercial or service conversation.

The Crachá has seven canonical fields — and all seven are mandatory. If one of them comes back empty, the badge is broken, because the minimum information for the next person (or IA) to continue is missing.

The 7 fields of the Crachá de Contexto

1. Origin (Origem) — Where this conversation was born: which acquisition channel, which campaign, which link, which referral. It is not "where it is now," it is "where it entered through." This is the field that connects the cycle to the marketing strategy.

Examples: "Meta Ad — Black Friday campaign," "Active customer referral," "Spontaneous return via WhatsApp Business."

2. Channel (Canal) — The medium through which the conversation is happening now, and the history of previous channels (if there was a migration). It is the "where," not the "why." It serves to adapt the tone and density of the response to the medium.

Examples: "WhatsApp (started in Instagram DM)," "Website chat," "Voice — call center."

3. Intent (Intenção) — What the customer wants to resolve in this conversation, expressed in one sentence, with a confidence level (high, medium, low). It is the explicit request, the declared goal. Without this field, the response becomes generic.

Examples: "Simulate financing for model X (high confidence)," "Compare 2 plans and choose (medium)," "Just researching, no urgency (low)."

4. Pain / Context (Dor / Contexto) — The implicit reason behind the request. The real constraint, the life situation, the emotional trigger that moves or stalls the decision. It is the "why" behind the "what." This is where a well-trained IAm adds value over a form.

Examples: "Family grew, needs to trade car for SUV," "Competitor increased price, seeking alternative," "Needs delivery by Friday or misses internal deadline."

5. Cycle State (Estado do ciclo) — Where this conversation stands in the MCI decision graph (discovery, evaluation, comparison, decision, post-sales, retention). It is the field that tells you what to do next — there is no point in sending a proposal to someone who is still in discovery. Without a state, the IAm speaks out of turn.

Examples: "Evaluation — comparing 2 brands," "Decision — waiting for credit approval," "Retention — customer showing signs of churn."

6. Summary (Resumo) — A decisional synthesis of the conversation, not a message dump. One or two sentences that anyone in the company could read in 3 seconds and understand the essentials: who they are, what they want, where they are, what has already been agreed upon. It is the most read field in daily operations — it is worth the effort to get it right.

Example: "Customer Aurora Prime, 3rd stay, wants to upgrade spa package for 1st week of September. Already accepted preliminary quote; pending confirmation of payment method."

7. Next Step (Próximo passo) — The most logical action to follow, with an SLA (deadline to happen) and an owner. It is not a vague suggestion like "follow-up": it is a specific action with a deadline. It is the field that ensures the cycle moves forward instead of stalling.

Example: "Send final quote with 10% discount by 5 PM today — owner: South regional sales."

Why 7 fields and not 4, nor 20

Fewer than 7 and there are missing variables for the IAm (or human) to decide without asking again. More than 7 and the Crachá becomes a form — no one reads it, no one fills it, no one trusts it. The practice of thousands of cycles has shown that these 7 are the minimum necessary and sufficient: they cover origin, medium, what, why, where, summary, and what comes next. Any other data (name, ID number, product, plan) already comes from the CRM/ERP and does not need to be in the Crachá — it references, it does not duplicate.

The Operational Context: the 5 fields that let AI negotiate alone

The Crachá de Contexto describes the conversation. But for an AI to truly converse (not just answer FAQs), it also needs company data regarding that conversation: the product at stake, the current commercial policy, what the client has already contracted, the authority limits, and a short memory of what was said in the last few turns.

This block is called Operational Context and consists of 5 mandatory fields. It complements the Crachá de Contexto and is the foundation of AI autonomy. If you only had these 5 fields and no conversation history, you should still be able to answer the client's next message with precision. That is the benchmark.

1. Product in focus — The concrete item on the table: name, model, version, price, stock, delivery time. Without concrete data, the AI responds generically ("we have several options...") — which is the opposite of selling.

Example: "Zeekr 001 YOU version — R$ 339,900 • 2 units in stock • delivery in 7 business days".

2. Commercial policy — What the AI is authorized to offer TODAY: current discounts, payment terms, campaigns of the month, gifts, trade-in bonuses. It defines the limits of the negotiation and prevents promises outside the agreement.

Example: "Trade-in bonus up to R$ 15k • 1.29% p.m. financing in 60x • June campaign: 1 year free maintenance".

3. Client assets — What they already have with the company that can serve as a hook: current product, active plan, loyalty points, current contracts, purchase history. It is the bridge for upsell, cross-sell, renewal, or trade-in. Without this field, the AI treats a long-time client as if it were the first time they spoke to the company.

Examples: "2021 T-Cross — internal appraisal R$ 98k", "12,450 Aurora Prime pts — 3rd stay", "300Mb Plan for 2 years — eligible for 1Gb upgrade".

4. AI Scope — What the AI can close alone and the clear trigger for human handoff. It is the dividing line that prevents two opposite disasters: AI promising what it cannot, and AI calling a human for every trivial thing.

Example: "Can schedule test-drives and simulate financing. Human handoff for discounts above 5% or specific unit reservation."

5. Last interactions (memory) — 2 to 4 short bullets with what has already been said in this conversation. It is the condensed memory. Without this field, the AI repeats questions the client has already answered — the cardinal sin of automated service.

Example: "Asked for Zeekr price", "Confirmed interest in SUV version", "Mentioned they have a current T-Cross".

How Crachá de Contexto and Operational Context combine

The Crachá de Contexto answers the question "who is this client and what do they want now?". The Operational Context answers the question "what does the company have to offer them, and how far can I go?".

Together, the 12 fields (7 + 5) form the minimum package that must accompany every conversation cycle so that an AI (or a human) can continue the conversation on any channel, with any agent, without forcing the client to repeat themselves and without making promises outside the agreement.

If one of the 12 fields comes back empty, the package is broken. It is the golden rule of MCI.

The Bandeja de Contexto: the presentation at the moment of decision

If the Crachá de Contexto is the record that exists all the time, the Bandeja de Contexto is the actionable presentation of that record at a specific moment: the moment of handoff. It is generated from the Crachá de Contexto when the conversation moves from AI to a human, or from one human to another.

The difference is subtle but fundamental:

• The Crachá de Contexto is structured data — it exists to be read by systems and as a reference.
• The Bandeja de Contexto is the actionable reading of that data by a person in 30 seconds, the moment they take over the conversation.

A second analogy: if the Crachá de Contexto is the patient's complete medical chart, the Bandeja de Contexto is the 30-second briefing the nurse gives the on-call doctor before they enter the room — distilled, prioritized, with the diagnostic hypothesis and the next decision on the table.

The 6 components of the Bandeja de Contexto

1. Current state and evidence — Where the decision stands and which signals support that reading. Not "what the customer said," but "what the system understands they want," with a level of certainty. The Bandeja de Contexto needs to show its "proof of work" — if the human trusts it, they act; if they distrust it, they ignore it and start over.

2. Dominant block — The system's hypothesis on what is preventing progress. This is the highest-value component of the Bandeja de Contexto because it connects signals that a human would hardly link in time. Example: customer asked twice about the deadline, mentioned a trip, and didn't reply to the last message in 40 minutes → hypothesis of the block is "deadline urgency due to travel," not "price."

3. Summarized history — The 3 to 5 most relevant pieces of information, not "all messages." The Bandeja de Contexto is distillation, not an archive. If the human needs to read the entire conversation, the Bandeja de Contexto has failed.

4. Recommended next step, with options — Not a rigid script. The Bandeja de Contexto suggests 2 or 3 possible paths with a short justification for each; the human decides. The IAm proposes, the human disposes.

5. Current SLA — At risk, expired, or within the deadline — and what has already been triggered. Without an SLA, the handoff has no sense of urgency. This is the field that makes the human prioritize this conversation among the 12 others they have open.

6. Recent state change — Where the decision came from in the last turns and what changed. It contextualizes the moment — helping the human understand why this conversation is being passed to them now and not before.

The ideal overflow is silent

One of the most important principles: the IAm does not need to announce "I will transfer you to an agent." This breaks the experience. A well-executed overflow is invisible to the customer — on their end, the conversation simply continues, richer.

On the company's side, the human takes over with context in seconds because the Bandeja de Contexto is already ready. The difference appears in the first sentence the human writes:

• Without Bandeja de Contexto (reset): "Hello, how can I help you?" — erases everything that came before.
• With Bandeja de Contexto (continuity): "I saw that the question regarding ERP integration remained open — I've brought in our specialist to resolve it now" — recognizes, validates, and moves forward.

From a generic question to a contextual statement. This is the operational difference between reset and trust. And it is measurable: in cycles with a Bandeja de Contexto, the post-handoff closing rate consistently rises between 15% and 40%, depending on the sector.

How Crachá and Bandeja Relate (The difference that avoids the reset)

This is the point that generates the most confusion. It is worth a direct summary:

In one sentence: the Crachá is the data, the Bandeja is the reading of that data at the right time, for the right person, in the right format.

Bandeja Quality: 3 Audit Criteria

MCI Ops monitors the quality of Bandejas based on three objective criteria:

Completeness. Are the six components present? If the friction hypothesis is empty, the human loses the main benefit — and falls back into the reset. An incomplete Bandeja is a broken Bandeja.

Accuracy. Does the reading correspond to the reality the human finds when talking to the client? If the AI says "stuck on budget" and the human discovers in 30 seconds that it was "decision-maker on vacation," the Bandeja failed and the model needs to be recalibrated. Accuracy is measured post-handoff, comparing the hypothesis with what the human confirmed.

Actionability. Can the human act without additional investigation? The goal is to read, mentally confirm, and act in less than 2 minutes. If the human needs to open 3 CRM tabs to complete the picture, the Bandeja failed due to a lack of density.

The anti-patterns (what not to do)

History dumping. Sending "every message since the first contact" isn't context — it's noise. The human receiving 50 messages spends more time reading than they would spend asking the questions again. The Bandeja de Contexto is distillation, not an archive.

Generic recommendations. "Suggestion: follow up" describes the obvious. The recommendation must be contextual and specific: "Suggestion: call today by 4 PM focusing on the trade-in condition — the client mentioned urgency and hasn't yet seen the value of the used vehicle appraisal".

Static Bandeja. If it's generated once and never updated, it ages fast. Did the client reply later? Did the state change? New evidence? The Bandeja de Contexto needs to be recalculated at every relevant interaction — otherwise, the human takes over with an expired snapshot.

"Optional" Crachá. Teams that treat Crachá de Contexto fields as optional destroy the system's value. If "pain" can be left empty, it will be left empty — and the next human will ask again. All 12 fields (7 from the Crachá + 5 Operational) are mandatory. Period.

Bandeja as a copy of the Crachá. If the human receives the 7 Crachá fields in raw format, that isn't a Bandeja de Contexto — it's raw data. The Bandeja needs to reorganize, prioritize, and suggest. If it doesn't do that, it's just one more screen in the CRM.

Where Crachá and Bandeja live in the stack

From a technical standpoint, a typical implementation places:

• Crachá de Contexto as a structured, versioned JSON object, updated by a server-side function at every relevant turn of the conversation. Persisted in the database, referenced by cycle ID, with version history for auditing.
• Operational Context hydrated from the CRM/ERP at the opening of the cycle and updated whenever commercial policies change or the client acquires/consumes something new.
• Bandeja de Contexto as a rendering (view) of the Crachá + Operational + last N turns, generated on demand when the handoff event triggers. It is not stored as the source of truth — it is regenerated from the Crachá to ensure freshness.

This separation (data vs. presentation) is what allows the same database to feed different versions of the Bandeja de Contexto for different agent profiles (SDR, closer, technician, retention) without duplicating information.

Conclusion: the antidote against the reset

The conversational reset costs more than most companies realize. It is not just a lower NPS — it is the perception that the brand does not organize itself internally to serve well. A customer who needs to repeat themselves learns, over time, to expect little.

the Crachá de Contexto ensures that information exists throughout the entire cycle, in a minimal and standardized format. The operational context ensures that the IAm has real company ammunition to negotiate. The Bandeja de Contexto ensures that this information arrives in the right format at the exact moment a human needs to make a decision.

The three, together, transform the handoff — historically the most fragile point of the journey — into the moment when the company best demonstrates that it cares. It is the difference between "I'm going to transfer you" and "I saw exactly what happened — let's move forward from here."

This is the bible of the Crachá. The rest is disciplined execution.