# Bot-to-human handoff transcripts as AI data

Canonical: https://refer.sourcex.si/data-types/ai-to-human-handoff-data
Last updated: 2026-10-20

AI-to-human handoff data is the transcript before a transfer, the escalation reason, the handoff summary and the human resolution. Contact centers and BPOs with years of linked records hold the boundary agent builders are tuning, and qualifying companies can license them directly to SourceX once rights and redaction are settled.

## What is bot-to-human handoff data?

Bot-to-human handoff data is the record of the moment an automated assistant passes a conversation to a person: the transcript up to the transfer, the reason for it, the summary handed over and how the human resolved the case. It marks the boundary agent builders are tuning, which is why contact center and BPO owners with years of these transcripts hold something AI labs and data users look for.

The phrase "human data" in AI circles covers material produced by people doing real work, as opposed to synthetic text generated by a model. A handoff transcript is a clear example: both the customer's words and the agent's resolution are human-produced, and the boundary between bot and person is labeled by the system itself.

Typical triggers are that the customer asks for a person, frustration shows, the bot's confidence is low, or the topic is one policy reserves for people. If your platform logs a reason, those triggers are already written into your records.

## What does a handoff record contain?

| Element | Example content | What it teaches |
|---|---|---|
| Pre-transfer transcript | The bot conversation, intents detected | What the bot attempted and where it stalled |
| Escalation reason | Customer request, sentiment, low confidence, policy topic, repeated failure | When a system should stop |
| Handoff summary | The context passed to the agent | What a useful summary contains |
| Warm transfer note | Agent's first message, what they did not need to ask again | Whether context survived |
| Human resolution | Steps taken, tools used, final answer | The correct behavior at the boundary |
| Disposition and outcome | Closed, callback, refund, follow-up ticket, repeat contact | The label |

The reason code and the resolution together are the pair. A record of a transfer with no resolution is half an example.

## Why are these records hard to get elsewhere?

Most vendors see only their own bot's side. The resolution stays in the contact center platform, the CRM or the ticketing tool, so the bot never learns what happened next. A company that kept both sides, linked by a conversation ID, holds the complete loop. That linkage is the asset.

For the same reason, the exception handling work in [exception handling records](/guides/exception-handling-ai-agents-data) applies directly: the cases where the bot failed are the cases agent builders most want to study.

## Where do the records live?

- Chat and messaging platforms with bot logs
- Contact center software with transfer events and agent notes
- Ticketing or CRM systems with case histories
- Quality assurance reviews and coaching notes
- Knowledge base articles the agent used
- Analytics exports of containment and escalation rates

BPOs and contact centers often run several of these across client programs. Ask owners to map them with the [data inventory builder](/tools/data-inventory-builder), which lists systems and records without capturing any content. A contact center owner might also read [shift handover logs](/data-types/shift-handover-log), where supervisors pass context to the next team.

## What are the rights and privacy considerations?

This is the part that decides whether a contact center can proceed.

1. **Whose conversations are they?** A BPO often works on behalf of clients. Transcripts from client programs generally belong to the client or are governed by the client contract. Without consent, they are out.
2. **What did customers consent to?** Recording or transcript notices, privacy policies and any consumer-facing terms matter. Consumer personal data without a licensing basis is a red flag, and recording and privacy rules vary by state. This is general information, not legal, tax or financial advice. Confirm with your own counsel.
3. **What sensitive data appears?** Payment details, health information and identity numbers need removal or exclusion under rules agreed before any work starts.
4. **Is the bot vendor's data in the mix?** Your chatbot platform's terms may address ownership of logs. Read them.
5. **Redaction.** Names, contact details and account identifiers are removed or replaced under terms agreed with the company, and data is delivered only after an executed agreement and the company's authorization.

The company's counsel decides these questions, not the partner. A buyer's perspective on what a rights-cleared record looks like is covered in [what an AI data buyer is](/proof/what-is-a-data-buyer).

## How can I tell whether my company has strong handoff data?

- [ ] A chatbot or virtual agent has been live for more than a year, ideally several
- [ ] Transfers log a reason code or a reason in free text
- [ ] The human agent's notes and disposition are tied to the same conversation ID
- [ ] Transcripts are retained rather than deleted after 30 or 90 days
- [ ] Records come from the company's own operations, not only client programs
- [ ] The company has 50+ full-time employees at peak (contractors excluded)
- [ ] An owner or executive can authorize a license

Retention is the most common gap. If transcripts are deleted on a rolling schedule, the history may be shorter than it appears. Related records, such as [insurance agency service requests](/data-types/insurance-agency-service-requests-ai), show how another servicing workflow holds request-to-resolution loops.

## Illustrative example

Illustrative and fictional: a regional customer support company has run a chat assistant for three years across its own help desk. Transfers carry one of five reason codes, agents add a resolution note, and the QA team scored a sample of each month's handoffs. The company can list the chat platform, the ticketing tool and the QA spreadsheets with years of history for each. That is a stronger starting point than a company with a recent bot and no retained transcripts.

## How do partner rewards work?

Partners earn 25% of the compensation SourceX actually pays the company you refer, capped at $100,000 per referred company. A reward becomes payable after SourceX has paid the referred company under its signed agreement and the payout conditions in the program terms are met, including finance approval and your tax and payment details. Anyone can join from any country; the introduced company must be a company in the U.S., Canada, the EU, the UK, Australia, New Zealand, India, Singapore, South Africa, Latin America, the UAE or Saudi Arabia (outside the U.S., Canada and the EU, at least 90% of the data must be in English). Read the [program terms](/terms).

## When should you skip this?

- The only transcripts belong to client programs without consent.
- Records are mostly consumer personal data with no licensing basis.
- Chat logs are deleted every few weeks.
- The company has not kept the human side of the resolution.
- The data was already licensed for AI training.

## Next step

Business owners can confirm fit on [who qualifies](/who-qualifies) and apply at [sourcex.si/apply](https://sourcex.si/apply). Partners who know a contact center or BPO owner can [register as a partner](/register) and introduce them.