Overview

Tealium is an enterprise Customer Data Platform that helps companies collect, unify, transform and activate customer data in real time to power personalized marketing and customer experiences across channels.

I led the evolution of Tealium’s AI capabilities, first designing AI Note Generator and AI Chat Assistant, then defining and launching the company’s first AI Decisioning Agent framework.

This shifted the platform from assistive AI to governed, predictive automation embedded directly into audience creation and activation workflows.

Impact

Platform

• Introduced Tealium’s first AI decisioning capability

• Established scalable agent framework for predictive automation

• Established reusable agent framework

• Built governance model for automated AI workflows

• Defined governance lifecycle reused across AI initiatives

Business

• Directly helped secure the company’s largest deal of the year, a $10M+ enterprise win with Europe’s third largest bank, Crédit Agricole Group, beating Adobe.

• Strengthened AI positioning in competitive enterprise sales cycles

• Contributed to Tealium’s recognition as a Leader in the 2025 Gartner Magic Quadrant for CDPs

• Advanced the platform from AI assisted workflows to an intelligent decisioning infrastructure.

Problem

Assistive AI improves productivity.
Decisioning AI influences customer outcomes.

Before this initiative:

• AI did not create audiences

• AI did not trigger connectors

• AI did not automate lifecycle decisions

• No governance model existed for automated activation

Introducing AI into activation workflows raised enterprise-level risks:

• Who owns the decision?

• What is editable vs system-managed?

• How do we prevent over-automation?

• How do we make invisible processing trustworthy?

The challenge was not introducing AI.
It was designing accountable AI.

My Role & Contribution

Foundation for Assistive AI

Designed:

• AI Note Generator

• AI Chat Assistant

Established:

• AI interaction patterns

• Transparency standards

• Trust guidelines

Decisioning AI Leadership

I led:

UX Strategy

• Defined decisioning AI as a new product category

• Positioned automation as structured infrastructure

• Proposed governance-first lifecycle (Draft → Review → Published)

• Framed Early Retention Signals as the initial release use case

• Structured roadmap between MVP and later phases

System & Interaction Design

• Designed the end-to-end agent builder flow

• Defined editable vs system-managed boundaries

• Structured inspectable inputs for trust

• Designed system-generated output architecture

• Introduced performance insight layer to measure impact

• Designed AI-detected missing attribute guidance

• Partnered with engineering on async architecture alignment

Cross-Functional Alignment

• Partnered with Product to define automation scope

• Worked with Engineering on architecture implications

• Partnered with Customer Success to validate churn pain

• Hosted Early Product Workshop with AI/ML experts and domain leaders

• Aligned Activation and Audience teams for safe integration

This introduced a new AI category into the platform.

Validation Strategy

Partnered closely with PM and Customer Success and spoke directly with enterprise customers, including:

• Country Road Group

• Abercrombie & Fitch

• AstraZeneca

• ASICS

• Sapient

• EMF

• Liberty Mutual

Across conversations and renewal cycles, we identified:

• Repeated manual churn segmentation workflows

• Duplicated lifecycle rules across teams

• Hesitation toward black-box automation

Insight:
Enterprise customers wanted automation — but with visibility and control.

Trade-Off Decisions

Automation vs Control

• No auto-publish

• Required Review state before activation

Transparency vs Cognitive Load

• Exposed model inputs clearly

• Hid background event attribute creation

• Surfaced system-generated artifacts in summary

Flexibility vs Model Integrity

• Inputs editable

• System-generated audiences locked

Speed vs Governance

• Avoided one-click automation

• Required structured configuration before publishing

Each decision protected enterprise trust.

End-to-End Agent Flow

Step 1 — Select Decision Objective

Initial release: Early Retention Signals (Churn Detection)

Step 2 — Review & Configure Inputs

• Preloaded attributes surfaced

• Users can replace or add inputs

• Model inputs visible and inspectable

Step 3 — System-Generated Outputs

The system automatically generates:

• At-risk audience

• Not-at-risk audience

• Supporting output attributes

Step 4 — Activation Configuration
Audience → Connector → Action

AI influences activation but does not hide it.
Human oversight remains intact.

Step 5 — Lifecycle Governance

Draft → Configured → Review Needed → Published

AI is treated as a managed entity, not a toggle.

Step 6 — AI Summary

• A summary slide-out appears after agent creation as a quick decision checkpoint

• Explains in plain language what’s happening, summarizing insights so users understand, not just raw data

• Shows expected impact and updates performance as real data comes in, with controls to adjust strategy

• This matters because users can quickly grasp and monitor results in one place

Phased Delivery Strategy

To balance speed, governance, and technical feasibility, I collaborated with the Scrum team (Product and Engineering) to define a staged rollout.

Rather than delay launch for full analytics infrastructure, we sequenced core decisioning first, followed by advanced performance insights.

Phase 1 — MVP

• Agent creation flow

• Inspectable inputs

• System-generated audiences

• Governance lifecycle (Draft → Review → Publish)

• Inline missing attribute detection

• Initial performance visibility

Goal: Introduce accountable decisioning safely and quickly.

Phase 2 — Advanced Insights

• Before vs after churn comparison

• Ongoing agent performance tracking

• Optimization recommendations

• Expanded explainability

Required additional backend investment, so it followed the MVP release.

Solution Validation

Early Product Workshop
Cross-functional pressure test with Product, Engineering, AI/ML, and Customer Success to validate governance boundaries and automation risk.

Usability Testing (UserTesting.com)
Ran moderated and unmoderated studies to assess:

• Clarity of what the agent will do before publish

• Understanding of editable vs system-managed logic

• Confidence configuring inputs and activation

• Whether lifecycle states (Draft/Review/Published) reduced automation anxiety

Insights informed guardrails, labeling, and summary clarity before rollout.

Direct User Feedback
Validated input transparency, trust in generated audiences, and activation oversight expectations.

Internal Release
Refined lifecycle states, summary explanations, and background artifact communication.

Early External Release
Monitored configuration completion, activation adoption, and Customer Success hesitation signals.

What I Would Improve

• Confidence scoring surfaced in UI

• Add explainability summaries

• Improve Agent performance dashboards 

• Optimization recommendations over time

Automation is phase one. Adaptive AI is phase two.