CVS Health Analytics and Behavior Change AI Content Builder

Employ AI to create and optimize multi-channel marketing campaigns—generate copy, craft subject lines, source creatives, and test performance across seven customizable templates tailored to business needs and campaign inputs.

Overview

Marketing Managers and Data Scientists struggled with a slow, manual, and a resource-intensive campaign creation process. The workflow lacked personalization, metadata, and robust testing, leading to inefficiencies, increased costs, and lower team morale. The existing Next Best Action (NBA) experimentation process was time-consuming and manual, resulting in limited behavior change opportunities.

Solution: AI-Powered Campaign Creation & Experimentation – A 0-to-1 Innovation at CVS Health

The Foundry Team within the Analytics and Behavior Change organization at CVS Health created a semi-automated system utilizing large language models to address these challenges by:

• Accelerating campaign generation with automated content creation.

• Enhancing personalization using metadata-driven AI insights.

• Enabling extensive testing for subject lines and campaign elements.

• Reducing costs, manual effort, manager and legal reviews, improving team efficiency and morale.

• Create editable templates for 7 lines of business each with their own brand identity

Create Content Workflow

Usability Testing

To develop a user-centered Gen AI content builder tailored to the distinct needs of marketers and data scientists, we conducted a rigorous and iterative UX research process. This involved establishing a core user group of five participants, strategically selected to represent diverse roles and experience levels within these two key personas. We ensured a mix of professionals from various lines of business, encompassing both marketing and data science disciplines, to capture a broad spectrum of perspectives.

Our research methodology centered around a series of weekly design thinking workshops, fostering a collaborative and dynamic environment for exploration and refinement. These workshops were structured to move through distinct phases of the design thinking process:

• Empathize:

  • We began by deeply understanding the user's current content creation workflows, pain points, and aspirations. This involved detailed interviews, contextual inquiries, and the analysis of existing documentation and tools.

  • We explored the specific tasks marketers and data scientists undertake in content generation, identifying the nuances of their individual needs. For marketers, this focused on brand voice, campaign alignment, and creative asset creation. For data scientists, the focus was on data visualization, report generation, and technical documentation.

  • We investigated their existing use of AI, their perceptions of generative AI, and their expectations for a content builder.

• Define:

  • Based on the insights gathered during the empathize phase, we synthesized key user needs and challenges.

  • We defined specific problem statements and user stories, outlining the desired outcomes and functionalities of the Gen AI content builder.

  • We created detailed user journey maps, visualizing the end-to-end experience and identifying critical touchpoints.

• Ideate:

  • We conducted collaborative brainstorming sessions, encouraging participants to generate a wide range of innovative ideas for the content builder's features and functionalities.

  • We utilized techniques like "Crazy 8s" and "How Might We" questions to stimulate creative thinking and explore diverse solutions.

  • We specifically wanted to explore the best ways to allow the users to control the AI, and how to allow for customization, and accuracy.

• Prototype:

  • We rapidly developed low-fidelity and high-fidelity prototypes, ranging from paper sketches to interactive digital mockups.

  • These prototypes were designed to visualize key features and user flows, allowing participants to experience the proposed solutions firsthand.

  • We created prototypes that allowed for varied levels of user input, and control of the gen AI.

• Test:

  • We conducted usability testing sessions, observing participants as they interacted with the prototypes and providing feedback.

  • We employed various testing methods, including think-aloud protocols, task-based scenarios, and surveys, to gather both qualitative and quantitative data.

  • We paid close attention to user interactions, identifying areas of confusion, frustration, and delight.

• Iterate:

  • Based on the feedback gathered during testing, we iteratively refined the prototypes, making necessary adjustments to the design, functionality, and user interface.

  • This iterative process continued throughout the workshop series, ensuring that the final solution was aligned with user needs and expectations.

  • We made sure to document all feedback, and changes, to ensure that the process was transparent, and repeatable.

By engaging in this comprehensive and iterative UX research process, we were able to create a Gen AI content builder that not only addresses the specific needs of marketers and data scientists but also provides a seamless and intuitive user experience."

Personas