cx-training-data-eligibilitylisted

# Using support conversations as training data Support transcripts are the most valuable training and evaluation data a customer-service AI could have, and the most legally and ethically loaded. The request usually arrives as an engineering task and it is not one — the gating questions come first, and if they fail, no amount of filtering fixes it. **This skill does not determine lawfulness.** It structures the assessment, surfaces the decisions someone else has to make, and — if they clear it — builds the dataset properly. ## The gating questions, before any data moves Work through these in order and stop at the first unresolved one: **1. What is the purpose, precisely?** Training a production model, fine-tuning, building an evaluation set, or prompt examples. These carry very different risk: an eval set of 200 hand-reviewed conversations is a different proposition from fine-tuning on two million. **2. Is it compatible with why the data was collected?** Customers contacted support to get help. Whether using that to train a model is compatible with that purpose — and on what lawful basis — is the central question and it is legal's to answer. **3. Where will it go?** In-house training on your own infrastructure, a provider's fine-tuning API, or a third party's platform. Each is a different transfer and processor question, and a provider's *default* terms on retention and training use frequently differ from the negotiated ones — only the configured arrangement matters. *