Generative AI assistant

Overview

A unified generative AI assistant across the Amazon Ads console

I owned design and research for a unified generative AI assistant across the Amazon Ads console, a 12-month, cross-functional effort with 1 PM, 4 SDEs, and 2 data scientists. The long-term vision was an assistant that could walk an advertiser from a question to a launched campaign without leaving the page, surfacing spend recommendations and creative tools along the way.

Recognizing that teams across the console were independently building their own AI interfaces, I saw that fragmentation would make that vision impossible. I expanded scope to design a unified platform built on a multi-agent orchestration layer capable of invoking specialized agents for image generation, reporting, keyword generation, and campaign creation. I collaborated closely with engineering on data architecture decisions to ensure the UI could handle the complexity of multi-agent responses, established four design principles, navigated a VP-level org conflict, and wrote design system guidelines for pattern adoption across teams.

The assistant launched with 10 skills and three interaction types — reactive, ambient, and inline — resolving 10% of support cases, eliminating 50k+ tickets, cutting average handle time from 25 minutes to 1.5 minutes, and saving nearly $600k in cost per contact. These early metrics reflected the assistant's initial launch scope. The bigger bet was on what came next — an advertiser that never needed to leave the page to go from insight to action.

The opportunity

From support chatbot to a unified AI platform

The long-term vision for the assistant was to help advertisers create better campaigns, faster, and surface spend recommendations that would keep them active and growing in the console. Support was the entry point because it was the highest volume problem, but the real opportunity was an assistant that could eventually walk an advertiser from a question to a launched campaign without ever leaving the page.

After starting work on the support chatbot, I noticed other teams across the console were independently building their own AI-powered experiences — one for image generation, one for reporting, one for campaign creation — each siloed, each starting from scratch. That fragmentation was going to make the bigger vision impossible. An advertiser couldn't move fluidly from support to campaign creation to creative generation if each of those lived in a different interface built by a different team.

I made the case to expand scope from a support tool into a unified platform with a multi-agent orchestration layer that all of those agents could plug into.

The goal

Provide advertisers with a cohesive assistant experience across the Ad Console

We defined five success targets going in:

• +5% resolved support cases — resolve 5% or more of support cases with the assistant within the first quarter.
• Launch 5+ agents — integrate multiple specialized agents under a single unified experience.
• −80% avg handle time — reduce support handling time from associate to assistant by at least 80%.
• +5% reports generated — increase NLP-generated reports by 5%.
• +10% campaigns with AI creative — get at least 10% of SMB advertisers to use AI-generated creative from the assistant.

Research

Four findings that shaped the direction

• Advertiser skepticism toward AI — Past bad chatbot experiences created real hesitation. Trust had to be earned through quality, not promised upfront.
• Low engagement with existing GenAI tools — Standalone experiences on console were underused, signaling a need to reframe how advertisers interact with AI entirely.
• Workflow-integrated access — Advertisers managing complex campaigns needed faster access to recommendations, creative tools, and answers — without leaving their workflow.
• Advertisers prefer chat over search — A strong signal that a well-designed assistant could become the primary interaction model on console.

Proposed solution

A cross-page unified assistant with inline and nudge entry points

This experience would provide not only support but also access to recommendations, image generation, keyword generation, and more — living in the existing help panel with three distinct interaction types for advertisers.

Data structure

Multi-agent orchestration layer

Working with product, engineering, and data science teams to define the assistant architecture. The system is built around a multi-agent orchestration layer with a multi-agent workflow handling intent extraction, skill/agent matching, response moderation, and suggested prompts — backed by a specialized agents registry, conversation history management, advertiser profile, and a responsible AI guardrails layer.

Design principles

Four principles to guide every design decision

• Design for imperfection — Generative model outputs may not always meet expectations. Highlight potential flaws and help users understand and work with imperfect outputs.
• Advertiser and consumer needs define AI capabilities — Work backwards from customers to identify needs, then invent on their behalf. We don't work backwards from technology.
• Experimentation and refinement before expansion — Maintain a high bar by testing, learning, and improving AI models before scaling.
• Design optimistically; anticipate rapid AI advancement — Solve for current limitations in a way that enables learning for both AI and customers as the technology evolves.

Design exploration

From intent detection to three interaction types

Early 2024

We started with a redesigned contact us experience, moving toward a conversational intent detection path. This groundwork trained our intent detection accuracy to almost 95%.

This phase was fundamentally intent-based — a finite set of advertiser intents routed to predefined responses. Reliable, but not the future.

Mid 2024

The next exploration revolved around a conflicting team working on the same solution. This escalated to a VP-level decision where we aligned on our team moving forward with our planned design and the conflicting team owning agent onboarding.

End of 2024

At this point the architecture shifted from intent classification to a true LLM backbone — capable of understanding natural language freely, reasoning across context, and invoking agents dynamically. The design challenges changed fundamentally: instead of designing for known states, we were now designing for infinite possible outputs.

Nearing the final design, we expanded the functionality by introducing two additional interaction types: proactive nudges and inline experiences contextual to each page. During this time there was also a design system refresh.

Final design

Three interaction types, 10 skills and counting

We launched three interaction types with 10 different skills and counting, focusing on quality of responses before launching each skill completely. At launch we supported: campaign moderation issues, image generation, report generation, billing issues, campaign creation, campaign optimization, advertiser suspensions, invalid clicks, promotional credits, and all education questions.

Guidelines & handoff

Design system guidelines for conversational AI

I wrote design system guidelines for the conversational AI assistant pattern in our Storm design system, providing a framework for pattern adherence and adoption across teams.

In tandem with the written guidelines I created a framework for both designers and engineers to work from — covering feedback patterns, the conversational layer, input fields, and response sources.

Outcomes

Exceeded our goals

The assistant resolved 10% of support cases, eliminating 50k+ tickets and nearly $600k in cost per contact. Average handle time dropped from 25 minutes to 1.5 minutes. We onboarded 4 partner teams under our framework and introduced more than 10 skills.

What's next

The next phase was moving beyond reactive responses toward an assistant that could proactively surface recommendations, initiate actions, and guide advertisers from insight to execution — all without leaving the page. The support metrics were proof of concept. The real product was still ahead.

Reflections

Designing for AI outputs taught me that trust is the hardest problem. Advertisers needed to act on recommendations confidently — which meant every response had to feel grounded, transparent, and within their control. The biggest design lesson was learning to design for uncertainty: how do you create interfaces that feel reliable when the underlying system is inherently probabilistic?