A marketer can predict which customers may leave, yet still need to decide what to say to them. Generative modeling addresses the second problem: AI models learn patterns from large amounts of text, images, and video, then produce new responses to prompts. Unlike a predictive model that returns a defined outcome such as a churn score, a generative model can produce open-ended text or images in response to instructions. Generative marketing applies this capability to marketing work. section 9.1 section 1.6.2
Useful applications span the path from insight to campaign:
- Explore customer data through natural-language questions and report summaries, then use the findings to inform audience segmentation. section 9.2.1
- Suggest audiences, targeting, personalization, creative ideas, and ways to interpret results. Marketers can refine suggestions through conversational feedback rather than accept a first answer. section 3.3.4 section 9.4
- Draft campaign content, including personalized emails and visuals. These are aids to specific tasks, not a fully autonomous process for designing, running, and measuring campaigns. section 1.6.2 section 9.2.1
From generic to tailored email
A fashion retailer tailored email suggestions to customers’ style preferences instead of sending generic recommendations. Click-through rates rose compared with the earlier approach. The example illustrates a measurable use for generated content, not a guarantee that every campaign will improve. section 1.6.2
Adoption depends on more than the model. Establish self-serve access to data and standardized measurement before adding AI-assisted ideation. Suggestions become more relevant when the system can use unified first-party data, available channels, and organizational objectives. Keep people in the feedback loop: models can hallucinate, and organizational policies may restrict use. Privacy-preserving clean rooms may support work with aggregated, anonymized shared data, but compliance and technical overhead remain considerations. section 3.3.4 section 4.4.5 section 8.5
Table 4.6 Artificial intelligence considerations
| Solution |
Considerations |
| Composable CDPs |
(+) Composable CDPs can connect directly to your team’s predictions, and the cloud-native architecture makes it easier to incorporate new AI capabilities provided by your cloud platform. |
| Marketing clouds/engagement platforms |
(-) MCs often provide limited prebuilt AI capabilities that focus on ease of use but lack integrations with major cloud platforms that offer more-advanced AI capabilities. |
| Monolithic CDPs |
(-) Monolithic CDPs often provide limited prebuilt AI capabilities that focus on ease of use but lack integrations with major cloud platforms that offer more-advanced AI capabilities. |
| In-house builds |
(+/-) The applications of AI and the availability of best-in-class AI capabilities are subject to the design of the in-house build, and they vary widely from one build to the next. |
Compare how first-party data solution types accommodate AI: direct access to team predictions and cloud capabilities differs from limited prebuilt features, while in-house capability depends on design. section 4.4.5
