Your Models or Theirs? The Architectural Bet Behind CDP AI Decisioning

Angle

CDP vendors have taken divergent positions on where AI models live: Tealium's IYOM (Invoke Your Own Model) lets organizations trigger their own models in their own cloud and activate in real-time; Oracle Unity bundles 27+ pre-built industry AI models; GrowthLoop ships three purpose-built AI agents (Audience, Journey, Insights). These aren't just product differences — they're bets about whether AI decisioning should be controlled and portable vs. optimized and pre-packaged. The article names what each bet costs and when each is worth it, without declaring a winner.

Key decision this helps with

Should your CDP AI decisioning use vendor-native models, or does your organization need to invoke its own models in your own infrastructure?

Tradeoffs the article will map

• Vendor-native AI models (faster time-to-value, lower infra overhead) vs. BYOM/IYOM (model portability, IP control, ADMT audit-readiness)
• Pre-built industry accelerators (Oracle 8+ vertical models, faster deployment) vs. custom models (higher differentiation, higher ongoing maintenance)
• AI agent orchestration bundled in CDP (GrowthLoop) vs. MCP-based interop with external AI systems (Tealium IYOM, Adobe MCP + Agent2Agent)

Open questions / uncertainties

• Whether vendor-native AI models can satisfy CCPA ADMT logic-disclosure requirements is unresolved — no vendor has published an ADMT compliance roadmap for bundled decisioning
• The operational cost of maintaining custom models in BYOM architectures over 3+ years vs. vendor model quality drift is difficult to benchmark before commitment

Knowledge-graph nodes this draws from

• concept.real-time-decisioning
• vendor.tealium
• vendor.oracle
• vendor.adobe-experience-platform