AI Model Class Selection Criteria
Tags: Frameworks
TL;DR
- Enterprise models offer stronger security, admin controls, and IP protections; public models are cheaper but pose data leakage and control risks.
- Your stance on using prompts/inputs for model training is a pivotal decision that drives contractual terms.
Why it matters for HK marketers: Choosing the right model class impacts confidentiality, compliance, and who owns or reuses your creative and audience data.
Key trade-offs
- Enterprise models: Security, scalability, admin controls, greater IP protection; higher cost and governance burden.
- Public models: Lower/no cost, broader training inputs; limited control/auditability, one-sided ToS.
Gating questions to ask
- Model type and purpose: Generative, predictive, or decision-support—and business use.
- Training data provenance: Proprietary, licensed, synthetic, or third-party—and lawful sourcing.
- Governance and audits: Policies, testing cadence, and update cycles.
- Compliance evidence: Documentation for legal and contractual obligations.
Operational safeguards
- Input-use restrictions: Bar reuse of your prompts/inputs for training outside your enterprise instance.
- Retention/deletion: Define timelines and feasibility, especially given black-box constraints.
- Employee policy: Acceptable use training to prevent sensitive data exposure in public tools.
Enterprise models typically agree not to train on your data without consent; public models often don’t.
So what for marketers
Standardize a due diligence checklist and require enterprise controls when handling sensitive or proprietary data.
← Back to Knowledge Base