When Your AI Agent Gets Stuck, a Real Person Takes Over.

• Customers who need a human get one, without starting over
• Your agent keeps the easy work and passes on the hard cases
• Every handover comes back with a written outcome your system can log

1. An AI chat agent for an online store meets a customer disputing a double charge. It hands the case to a person, who reviews the order history provided, explains the next step to the customer and returns a resolution note.
2. A voice of the customer agent detects a frustrated, repeat caller. A human picks up by email or chat in the customer's language and closes the loop.
3. A subscription app's agent is not allowed to approve exceptions. It escalates refund exception requests to a person who applies your written policy and records the decision.

Reads the AI transcript, resolves or routes the case within the customer's written policy, records the outcome in the customer's helpdesk.

Every request is reviewed by a person. We decide case by case whether we deliver it with our own team or through a vetted partner, starting with teams in the Philippines. If a job isn't something we can do well or lawfully, we'll tell you plainly.

Human agents take over conversations your AI support cannot or should not finish.

Venture ID: nbvn.venture.escalation. Status: manual_matching.

What to send

Common to every Venture: requester_contact (required), authorized_by (required), agent_note, languages (required), customer_geography (required), start, budget, data_sensitivity (required).

Specific to the Escalation Venture:

• channels (required): Channels the handover arrives on: chat, email, ticket
• helpdesk_system (required): Helpdesk the human works in, e.g. Zendesk
• escalation_triggers (required): When your agent should hand over
• monthly_volume (required): Rough escalations per month
• hours (required): Hours of coverage, with time zone
• knowledge_base_url: Your help content or written policy
• handoff_payload: What each handover contains, e.g. transcript plus customer id

What comes back

• case_id
• resolution
• notes
• handled_by_role
• closed_at
• needs_customer_action

Status moves through received, reviewing, fulfillment_decided, scoping, in_delivery, delivered and closed, or declined with a reason_code.

What we won't do

• legally regulated work requiring a license or registration (legal advice, medical, financial advice, insurance, licensed investigation)
• debt collection or collections calls
• cold outbound sales or telemarketing without documented prior consent
• anything legally stretchy or unclear in law, at our discretion
• payment card handling beyond what an agreed compliant setup allows
• impersonation, deception, or contacting people who have not agreed to be contacted
• giving legal, medical or financial advice

How to submit

Use the form at /request/?venture=escalation or POST https://nbvn.com/api/request with JSON. Include venture_id (service_id is accepted as an alias). You get a reference number straight away and a reply by email from a person. API and MCP access are planned.

{
  "venture_id": "nbvn.venture.escalation",
  "description": "[TEST] An AI chat agent for an online store meets a customer disputing a double charge. It hands the case to a person, who reviews the order history provided, explains the next step to the customer and returns a resolution note.",
  "requester_contact": {
    "name": "Dana Test",
    "email": "dana@nbvn-qa.example",
    "org": "Example AI Inc"
  },
  "authorized_by": "Customer ops lead, Example AI Inc",
  "agent_note": "Submitted by our agent on behalf of Example AI Inc",
  "languages": [
    "English"
  ],
  "customer_geography": [
    "US"
  ],
  "start": "2026-12-01",
  "budget": "to be discussed",
  "data_sensitivity": "pii",
  "channels": [
    "chat",
    "email"
  ],
  "helpdesk_system": "Zendesk",
  "escalation_triggers": "refund over policy, angry customer, bot loop",
  "monthly_volume": 3000,
  "hours": "Mon to Sun 08:00 to 22:00 PT",
  "knowledge_base_url": "https://example.com/help",
  "handoff_payload": "conversation transcript plus customer id",
  "consent": "yes"
}

How does my AI agent hand a customer to a human through NBVN?

Today you submit a request describing your escalation flow through /request/ or a JSON POST to /api/request with venture set to escalation. A person reviews it and agrees a handover arrangement with you. A live API and MCP handover are planned, not available yet.

What context should the handover include?

Describe it in handoff_payload, for example the transcript plus a customer id, and link your written policy in knowledge_base_url. Without a written policy we will ask for one before delivery starts.

Can the human hand the customer back to my agent?

Yes, if your setup allows it. Each case comes back with a resolution, notes, the role that handled it and whether the customer still needs to act.

Which languages can escalations be handled in?

Tell us which languages you need. Teams in the Philippines are our first option, and we confirm language coverage case by case before agreeing anything.

• Call Overflow Venture

  Start this Venture →

• Booking Venture

  Start this Venture →

• Field Verification Venture

  Start this Venture →

• Back Office Venture

  Start this Venture →