AI-Powered Onboarding System for Personalization & A/B Testing
A phased evolution from manual training materials and live sessions into a fully AI-driven onboarding and campaign creation system — using a Gemini Gem, a Claude Skill, and automated workflows to help teams launch personalization campaigns faster and more consistently.
Outcome
Reduced campaign build time from 3 hours to 30 minutes through AI-assisted onboarding and automated workflows.
The Challenge
Onboarding new users to a personalization and A/B testing program is time-intensive. Each new team member or stakeholder needed to understand internal processes, tooling, naming conventions, and quality standards before they could launch independently. Relying on manual documentation, live training, and ongoing Q&A created a bottleneck and inconsistency as the program scaled.
The opportunity was to systematize this knowledge — first through structured documentation, then through AI tooling that could guide users interactively and automate the most repetitive parts of the campaign build process.
The Approach: A Phased Evolution
Phase 1 — Manual Foundation
Built out a structured onboarding program from scratch: documentation covering internal A/B testing and personalization processes, live training sessions, Q&A, and published internal Wiki articles. This created a single source of truth and reduced the time needed for individual coaching — but still required manual delivery and ongoing maintenance.
Phase 2 — AI-Assisted Onboarding
Evolved the program into AI-native tooling to make onboarding interactive and available on-demand. This included:
- ▸Gemini Gem — a custom AI assistant trained on internal processes and documentation, allowing users to ask questions and receive guided, contextually accurate answers without requiring a live session.
- ▸Claude Skill — a structured skill built to guide users step-by-step through the campaign creation process, enforcing consistency in naming, targeting logic, and QA checklists.
Phase 3 — Automated Campaign Workflows
Extended AI tooling into the campaign build process itself — automating steps that previously required manual configuration, reducing the time to build a campaign from 3 hours to 30 minutes. Required thoughtful design of how end-users would interact with the tools, iterative testing of adoption patterns, and a structured feedback loop post-rollout to refine the system.
A key part of Phase 3 was an AI-powered offer generation tool that allowed users to produce campaign-ready content variants without writing a single line of copy. Users select a Cloudflare product, choose a design template (e.g. Toaster Banner, Modal), set their campaign goal (Awareness, Trial, Conversion, Retention), and pick a tone (Professional, Conversational, Urgent). The tool then generates multiple offer variants — each with a headline, CTA, HTML preview, and structured JSON — tailored to the visitor's company context.
Each generated variant is saved to an offer repository, where it can be reviewed, previewed in context, and pushed directly to the personalization system — Adobe Target, in this case — with a single click. The repository stores both the structured JSON (for system ingestion) and the rendered HTML (for visual QA), keeping everything needed for launch in one place.
My Role
Senior Manager, MarTech — owned the onboarding program from initial design through AI transformation. Responsible for building the original training materials, designing the AI tool interactions, testing adoption, and gathering and acting on user feedback after rollout. This project combined domain expertise, instructional design, and hands-on AI tool development.