Evolving roles
Product designers learn prompt and model behavior patterns, content designers shape assistant behavior, and researchers evaluate real model output.
01 Background
I joined Meta as a Senior Design Manager on the Ads Experience Platform group, leading design for Meta AI Business Assistant, an effort at the intersection of AdTech and AI.
The mission is to bring AI-driven insights and assistant-style workflows directly into the tools advertisers use every day, so that creating, managing, and optimizing campaigns feels less like operating software and more like collaborating with a knowledgeable partner.
Ads is Meta's largest business, and the surface our AI work has to earn its place on. At this scale, small experience decisions ripple across a very large advertiser base.
02 Team setup & operating model
Leading a 15-person design organization across MAIBA, Ads Manager Mobile, and usability.
15
team members
2
design leads
13
ICs
5
workstreams
The Ads Experience Platform design group was organized around two core product pods — Meta AI Business Assistant and Ads Manager Mobile — supported by specialized craft anchors in compliance, business reporting, and usability. I designed the org for clarity, scale, and decision velocity.
Org footprint
Ads Experience Platform Design
Team 1
MAIBA
1 lead + 5 ICs
Team 2
Ads Manager Mobile
1 lead + 5 ICs
Compliance
Quality & policy
1 IC
BR3
Business reporting
1 IC
Usability
Research quality
1 IC
Measured by REACH
R
Results
Business impact
E
Efficiency
Protect maker time
A
Accountability
Clear owners
C
Clarity
One source of truth
H
Health
Team growth
03 The product
Meta Ads at a glance
The scale behind every campaign.
3.98B
Users globally
205M+
Daily active users
in NAM
in NAM
10M+
Active advertisers
$201B
Total revenue
97%+
Share from ads
Where MAIBA fits
Meta is building AI products for many surfaces and audiences. MAIBA is the assistant built for advertisers, living inside Ads Manager and Business Support Home rather than in a separate chatbot destination.
Product
Built for
Lives in
Meta AI
For users on Meta platforms
Meta AI App, WhatsApp, Instagram, Facebook, Messenger, web, glasses, and Quest
Biz AI
For users interacting with businesses
Messenger, WhatsApp, and websites
MAIBA
My focus
For advertisers
Inside Ads Manager and Business Support Home
Sales AI Companion
For sales teams
Inside CRM workflows
What MAIBA does
MAIBA is an always-on generative AI chat assistant that gives advertisers personalized guidance, campaign insights, recommendations, and account issue support inside their existing workflow.
Embedded in Ads Manager
No tool switching or copy-pasting campaign context into a separate chatbot.
Trend analysis and insights
Performance shifts explained in plain language, not buried in metrics tables.
AI ad recommendations
Personalized recommendations based on each advertiser’s campaigns and goals.
✓
Account issue support
Instant diagnostics for routine issues, with escalation paths when human help is needed.
Why this matters
Advertisers were increasingly dissatisfied with traditional support: advice felt generic, continuity was poor, outreach was intrusive, and resolution could be slow. Some advertisers had started turning to third-party AI tools, moving sensitive campaign context outside Meta’s ecosystem.
○
Pain point
Lack of personalization
Support advice often failed to understand the advertiser’s actual business context.
↻
Pain point
No continuity
Advertisers had to restate their situation as ownership changed across sessions.
☎
Pain point
Intrusive communication
Calls and outreach could interrupt the workflow instead of helping at the right moment.
◷
Pain point
Inconsistent resolution
Guidance was sometimes too broad or too late to support high-stakes campaign decisions.
MAIBA opportunity
Deliver personalized, always-on guidance natively inside Meta’s advertising tools, making the product clearly better than external workarounds while keeping data where it belongs.
04 Design process
This was our team's first AI-native business tool. We had no internal blueprint — the model layer was changing weekly, and we needed a shared North Star before moving too far into execution.
A five-day design sprint in Menlo Park brought together a cross-functional team to align on the vision, interaction model, and evaluation criteria.
01
Design sprint
Challenge
Designing without a blueprint
This was our first AI-native business tool. The team needed patterns for an assistant that could reason, respond, and take context from live product surfaces.
Constraint
Trust, latency, and evolving model behavior
Design had to communicate confidence, boundaries, and waiting states where model behavior was still evolving.
Strategy
Align the system before the screen
The sprint forced the right trade-offs early: user value, model capability, business impact, and operational feasibility.
Who needed to be in the room
✎Design
⌕Research
♙PM
</>Engineering
▥Data Science
◁PMM
☆Leadership
♧XFN teams
02
From sprint to vision
The sprint produced a North Star vision. The months that followed turned it into a shipping product through capability frameworks, prototypes, research, model-informed iteration, and trade-off decisions.
1
North Star
A shared vision that defined what success meant for users, the business, and the system.
→
2
Product direction
Guided by capability frameworks, research, and model-informed iteration.
→
3
Shipping system
Shipped capabilities with measurable impact and a foundation for scale.
Detailed case study available upon request
Specific design decisions, prototypes, research findings, model behavior trade-offs, and outcome details for Meta AI Business Assistant are available for those interested — feel free to contact me to learn more.
✉ Contact me on LinkedIn or at keeprain@gmail.com.
05 Launched experience
Beta launch
Oct 2025
In Oct 2025, we launched the beta of Meta AI Business Assistant (MAIBA) inside Ads Manager and Business Support Home.
MAIBA delivers personalized, proactive guidance that helps advertisers analyze performance, solve account issues, and take action—faster.
Using current and prior campaign data, the assistant can surface insights, compare results to benchmarks, and recommend improvements directly from the chat.
Key features
Impact (Beta results)
Early results from advertisers using Meta AI business assistant in beta.
12%
median decrease in cost per result
for advertisers who applied opportunity score recommendations from Meta AI business assistant.
20%
increased resolution rates
of common account management issues with Meta AI business assistant, which is currently in Beta.
30%+
task time saved
estimated time saved on routine performance analysis and issue resolution.
4.8/5
user satisfaction (beta)
average rating from beta testers on the quality and usefulness of recommendations.
“
1. Based on internal Meta analysis, Oct–Dec 2025.
2. Based on internal data, Oct–Dec 2025.
3. Based on user survey with beta testers, Oct–Dec 2025.
4. Based on NPS-style survey with beta testers, Oct–Dec 2025.
06 Reflections and learnings
Across this work, I developed a stronger point of view on what it takes to design AI products. Designing for AI is not simply adding a smart layer on top of a normal UI. The interface itself becomes a partner in a system of intelligence.
“
The design challenge shifts from "here are the features" to "here is a system that learns, suggests, adapts, and interacts."
Shifts in how we design
From
To
Key design practice
From
Fixed states
→
To
Dynamic systems
Key design practice
Communicate model boundaries and make ambiguity explicit.
From
Static screens
→
To
Adaptive scaffolding
Key design practice
Preserve user control through prompts, hints, feedback loops, and recoverable decisions.
From
Completion rates
→
To
Response quality
Key design practice
Evaluate clarity, creativity, and intent alignment, not only task completion.
From
Specs
→
To
Field testing
Key design practice
Close the loop with model runs, eval data, and human feedback.
Building an AI-native design team
Augmented workflows
Critique shifts from "does the screen work" to "does the system behave," with live dogfooding and eval data as part of review.
AI-native pods
Small cross-functional pods carry end-to-end ownership so the team can move faster without losing accountability.
© Copyright 2026 -
Designed & coded by Frank Wu