a leading advertising-technology platform
Designed how an AI-powered learning and help module should live inside an established, high-traffic ad-tech product without disrupting existing workflows.
Role
Product & UX design
Client and product details are described at an industry level to respect confidentiality; the problem, decisions, and reasoning shown here are unchanged.
The problem
The product needed an AI-assisted learning/help layer, but the product already had an established, high-traffic workflow used by advertisers daily. The real problem wasn't designing an AI assistant — it was deciding how much of the product an AI layer should be allowed to touch before it started working against the people already using it well.
Approach
- 01
Mapped where users were actually getting stuck in the existing product before proposing where AI assistance should appear — avoided defaulting to a bolted-on chat widget.
- 02
Designed the module to answer questions and surface guidance in context, without interrupting or gating the existing task flow.
- 03
Treated AI output as suggestion-and-verify rather than autonomous action inside a product where mistakes have real ad-spend consequences.
Decisions
Contextual help over a chat-first interface
Rejected a standalone chat panel as the default entry point. The AI layer surfaces relevant guidance inline, next to the task the user is already doing, because the product's existing users are task-focused specialists, not people who want a conversation with a bot.
AI proposes, the user confirms
In a product tied to real advertising budgets, the AI module never takes an action on the user's behalf without an explicit confirmation step — the interface always shows what the AI is suggesting and why, not just an outcome.
Designing with AI
A direct human-in-the-loop case: every AI suggestion is visible and requires confirmation before it affects a live campaign — the opposite of an autonomous-agent pattern, chosen deliberately given the stakes.
Outcome
Shipped design direction for integrating an AI learning/help layer into an existing, high-traffic product without disrupting its core workflow.
Outcome as reported at the time of the engagement.