Synthesizing decision-fatigue research into universal AI/UX patterns.
- AI/UX
- Pattern Systems
- Building for AI
101010+
Universal patterns delivered
Synthesizing decision-fatigue research into universal AI/UX patterns.
- AI/UX
- Pattern Systems
- Building for AI
101010+
Universal patterns delivered
Synthesizing decision-fatigue research into universal AI/UX patterns.
- AI/UX
- Pattern Systems
- Building for AI
101010+
Universal patterns delivered
TL;DR
The 30 Sec Version!
TL;DR
The 30 Sec Version!
TL;DR
The 30 Sec Version!
Testbed
Adobe Experience Platform
Adobe Experienece Platform (AEP) helps marketers build audiences from large customer datasets. We used Audience Builder as the testbed for studying AI-assisted decision-making.
My role
UX Designer & Researcher
- Research synthesis on decision fatigue, AI transparency and accessibility
- Audit of how AEP's AI presents its outputs
- Designed 3 of the 10+ AI/UX patterns
Problem
Decision fatigue
AI made audience building faster, but added cognitive overload. Marketers had to interpret what the AI inferred, why it suggested something, and whether it could be trusted.
Solution
Universal AI/UX patterns
We created 10+ reusable AI/UX patterns to reduce decision fatigue across AI workflows. Each was grounded in one of three recurring challenges and demonstrated inside AEP.
Impact
- 10+Design patterns deliveredEach one tied to a single challenge
- 3Challenges mappedDrawn from 25+ research sources
- 1→AnyTestbed to universalBuilt on AEP, designed to travel
Testbed
Adobe Experience Platform
Adobe Experienece Platform (AEP) helps marketers build audiences from large customer datasets. We used Audience Builder as the testbed for studying AI-assisted decision-making.
My role
UX Designer & Researcher
- Research synthesis on decision fatigue, AI transparency and accessibility
- Audit of how AEP's AI presents its outputs
- Designed 3 of the 10+ AI/UX patterns
Problem
Decision fatigue
AI made audience building faster, but added cognitive overload. Marketers had to interpret what the AI inferred, why it suggested something, and whether it could be trusted.
Solution
Universal AI/UX patterns
We created 10+ reusable AI/UX patterns to reduce decision fatigue across AI workflows. Each was grounded in one of three recurring challenges and demonstrated inside AEP.
Impact
- 10+Design patterns deliveredEach one tied to a single challenge
- 3Challenges mappedDrawn from 25+ research sources
- 1→AnyTestbed to universalBuilt on AEP, designed to travel
Testbed
Adobe Experience Platform
Adobe Experienece Platform (AEP) helps marketers build audiences from large customer datasets. We used Audience Builder as the testbed for studying AI-assisted decision-making.
My role
UX Designer & Researcher
- Research synthesis on decision fatigue, AI transparency and accessibility
- Audit of how AEP's AI presents its outputs
- Designed 3 of the 10+ AI/UX patterns
Problem
Decision fatigue
AI made audience building faster, but added cognitive overload. Marketers had to interpret what the AI inferred, why it suggested something, and whether it could be trusted.
Solution
Universal AI/UX patterns
We created 10+ reusable AI/UX patterns to reduce decision fatigue across AI workflows. Each was grounded in one of three recurring challenges and demonstrated inside AEP.
Impact
- 10+Design patternsEach one tied to a single challenge
- 3Challenges mappedDrawn from 25+ research sources
- 1→AnyTestbed to universalBuilt on AEP, designed to travel
You just wanted a recommendation
Now you're comparing fifty of them.
You just wanted a recommendation
Now you're comparing fifty of them.
You just wanted a recommendation
Now you're comparing fifty of them.
Simple ask
50 options
Filters
30 new options
You ask an AI where to eat. It gives you fifty options.
You narrow it down by cuisine, price, location.
Somehow, there's more to decide than when you started.
Simple ask
50 options
Filters
30 new options
You ask an AI where to eat. It gives you fifty options.
You narrow it down by cuisine, price, location.
Somehow, there's more to decide than when you started.
Simple ask
50 options
Filters
30 new options
You ask an AI where to eat. It gives you fifty options.
You narrow it down by cuisine, price, location.
Somehow, there's more to decide than when you started.
Now picture that same spiral except the stakes are a campaign budget, and the options are audiences of millions.
Now picture that same spiral except the stakes are a campaign budget, and the options are audiences of millions.
Now picture that same spiral except the stakes are a campaign budget, and the options are audiences of millions.
This is where marketers build the audience
Adobe Experience Platform
This is where marketers build the audience
Adobe Experience Platform
This is where marketers build the audience
Adobe Experience Platform
Adobe Experience Platform helps marketers build audiences from massive, messy customer data. Its Audience Builder turns broad campaign goals into precise
targeting rules.
Adobe Experience Platform helps marketers build audiences from massive, messy customer data. Its Audience Builder turns broad campaign goals into precise
targeting rules.
Adobe Experience Platform helps marketers build audiences from massive, messy customer data. Its Audience Builder turns broad campaign goals into precise
targeting rules.
AI was added to help by summarizing data, suggesting audiences, interpreting intent.
But it also added a new layer of work. Now users had to understand what the AI inferred, why it suggested something, and whether any of it could be trusted.
AI was added to help by summarizing data, suggesting audiences, interpreting intent.
But it also added a new layer of work. Now users had to understand what the AI inferred, why it suggested something, and whether any of it could be trusted.
AI was added to help by summarizing data, suggesting audiences, interpreting intent.
But it also added a new layer of work. Now users had to understand what the AI inferred, why it suggested something, and whether any of it could be trusted.
Before we could fix anything
We had to understand it
Before we could fix anything
We had to understand it
Before we could fix anything
We had to understand it
We studied two things in parallel: what makes a decision feel hard, and how AEP's audience-building workflow actually works.
We studied two things in parallel: what makes a decision feel hard, and how AEP's audience-building workflow actually works.
We studied two things in parallel: what makes a decision feel hard, and how AEP's audience-building workflow actually works.
Literature review
Studied decision fatigue, cognitive load, choice overload, and AI transparency research.
DESIGN INSIGHT
Users often fail because they never know how AI interpreted what they said.
Literature review
Studied decision fatigue, cognitive load, choice overload, and AI transparency research.
DESIGN INSIGHT
Users often fail because they never know how AI interpreted what they said.
Literature review
Studied decision fatigue, cognitive load, choice overload, and AI transparency research.
DESIGN INSIGHT
Users often fail because they never know how AI interpreted what they said.
JTBD research
Mapped the decisions marketers make while building audiences.
DESIGN INSIGHT
Most decision fatigue starts before the AI even responds - at the moment marketers have to decide what to type.
JTBD research
Mapped the decisions marketers make while building audiences.
DESIGN INSIGHT
Most decision fatigue starts before the AI even responds - at the moment marketers have to decide what to type.
JTBD research
Mapped the decisions marketers make while building audiences.
DESIGN INSIGHT
Most decision fatigue starts before the AI even responds - at the moment marketers have to decide what to type.
Current-state audit
Studied how AEP's Audience Builder and AI agent present outputs today.
DESIGN INSIGHT
AEP's AI gave confident answers with no way to see how it got there.
Current-state audit
Studied how AEP's Audience Builder and AI agent present outputs today.
DESIGN INSIGHT
AEP's AI gave confident answers with no way to see how it got there.
Current-state audit
Studied how AEP's Audience Builder and AI agent present outputs today.
DESIGN INSIGHT
AEP's AI gave confident answers with no way to see how it got there.
Literature review
Reviewed how HubSpot, Klaviyo, Salesforce Einstein, and Perplexity structure AI recommendations.
DESIGN INSIGHT
Every tool we studied did the same thing. Summarize, then let users edit.
Literature review
Reviewed how HubSpot, Klaviyo, Salesforce Einstein, and Perplexity structure AI recommendations.
DESIGN INSIGHT
Every tool we studied did the same thing. Summarize, then let users edit.
Literature review
Reviewed how HubSpot, Klaviyo, Salesforce Einstein, and Perplexity structure AI recommendations.
DESIGN INSIGHT
Every tool we studied did the same thing. Summarize, then let users edit.
This showed us where AI was reducing effort, and where it was quietly creating more decisions.
This showed us where AI was reducing effort, and where it was quietly creating more decisions.
This showed us where AI was reducing effort, and where it was quietly creating more decisions.
Then came the question
Could this work beyond AEP?
Then came the question
Could this work beyond AEP?
Then came the question
Could this work beyond AEP?
We brought our sketches back to Adobe, and they saw something we hadn't: we were redesigning AEP. Our ideas came from the research, but we'd turned them into AEP features when the goal was something that could work across all of Adobe's AI workflows.
We brought our sketches back to Adobe, and they saw something we hadn't: we were redesigning AEP. Our ideas came from the research, but we'd turned them into AEP features when the goal was something that could work across all of Adobe's AI workflows.
We brought our sketches back to Adobe, and they saw something we hadn't: we were redesigning AEP. Our ideas came from the research, but we'd turned them into AEP features when the goal was something that could work across all of Adobe's AI workflows.
Early Concept and Problems They Solved
Early Concept and Problems They Solved
Task Decision Timeline
IN FINAL WORK
Not knowing what you have already decided, or whether you can undo it.
Task Decision Timeline
IN FINAL WORK
Not knowing what you have already decided, or whether you can undo it.
Task Decision Timeline
IN FINAL WORK
Not knowing what you have already decided, or whether you can undo it.
Sticky context
Forgetting what the task was in a long conversation.
Sticky context
Forgetting what the task was in a long conversation.
Sticky context
Forgetting what the task was in a long conversation.
Task navigator
Losing progress when you jump between steps.
Task navigator
Losing progress when you jump between steps.
Task navigator
Losing progress when you jump between steps.
Must-have and optional input
Not knowing what the AI needs from you and what can wait.
Must-have and optional input
Not knowing what the AI needs from you and what can wait.
Must-have and optional input
Not knowing what the AI needs from you and what can wait.
Consequence preview
IN FINAL WORK
Not knowing what will happen before you choose.
Consequence preview
IN FINAL WORK
Not knowing what will happen before you choose.
Consequence preview
IN FINAL WORK
Not knowing what will happen before you choose.
Customized output
Getting more information than you can actually use.
Customized output
Getting more information than you can actually use.
Customized output
Getting more information than you can actually use.
We didn't have to start over. We just had to step back and ask what problem each one was solving.
We didn't have to start over. We just had to step back and ask what problem each one was solving.
We didn't have to start over. We just had to step back and ask what problem each one was solving.
Which left us with a better question to work on.
How Might We…
Which left us with a better question to work on.
How Might We…
Which left us with a better question to work
How Might We…
How might we help people stay confident in their decisions with AI, beyond a single workflow?
How might we help people stay confident in their decisions with AI, beyond a single workflow?
How might we help people stay confident in their decisions with AI, beyond a single workflow?
We went back to the research with that question.
The same three breakdowns kept showing up.
We went back to the research with that question.
The same three breakdowns kept showing up.
We went back to the research with that question.
The same three breakdowns kept showing up.
Instead of chasing one-off friction points in AEP, we looked for the breakdowns that kept repeating across products and workflows. Those breakdowns became our three core challenges.
Instead of chasing one-off friction points in AEP, we looked for the breakdowns that kept repeating across products and workflows. Those breakdowns became our three core challenges.
Instead of chasing one-off friction points in AEP, we looked for the breakdowns that kept repeating across products and workflows. Those breakdowns became our three core challenges.
Decision Fatigue in AI Workflow
Decision Fatigue in AI Workflow
CHALLENGE 1
CHALLENGE 1
Translating messy intent into system language
Translating messy intent into system language
People struggle to turn messy goals into clear prompts, often unsure what information the AI needs.
People struggle to turn messy goals into clear prompts, often unsure what information the AI needs.
What do I tell the AI?
What do I tell the AI?
What do I tell the AI?
2 patterns
2 patterns
2 patterns
CHALLENGE 2
CHALLENGE 2
Keeping track in a multi-step process
Keeping track in a multi-step process
Users lose track of what they’ve decided, what the AI is doing, and where they are in the flow.
Users lose track of what they’ve decided, what the AI is doing, and where they are in the flow.
Where am I? What have I done?
Where am I? What have I done?
Where am I? What have I done?
3 patterns
3 patterns
3 patterns
CHALLENGE 3
CHALLENGE 3
Balancing too many trade-offs at once
Balancing too many trade-offs at once
Users have to compare several attributes and outcomes without enough help seeing what changes with each choice.
Users have to compare several attributes and outcomes without enough help seeing what changes with each choice.
What changes if I pick this one?
What changes if I pick this one?
What changes if I pick this one?
2 patterns
2 patterns
2 patterns
From here, our research focused on the micro moments that come before decision fatigue.
From here, our research focused on the micro moments that come before decision fatigue.
This is where the research landed.
Patterns built to work beyond AEP.
This is where the research landed.
Patterns built to work beyond AEP.
This is where the research landed.
Patterns built to work beyond AEP.
We turned the three recurring challenges into universal AI/UX patterns designed to work across products and workflows. AEP gave us a real enterprise context to show what each pattern could look like in practice.
We turned the three recurring challenges into universal AI/UX patterns designed to work across products and workflows. AEP gave us a real enterprise context to show what each pattern could look like in practice.
Goal-Oriented Alignment
Sometimes people don't have a clear way to phrase what they want.
The first input comes back as a structured view of what the system captured, what it inferred and what is still missing, with a way to edit it.
We were studying how AI adds decisions
and we used AI along the way.
We were studying how AI adds decisions
and we used AI along the way.
We were studying how AI adds decisions
and we used AI along the way.
It felt odd to design for AI workflows without working in one. We used AI where it sped up our thinking, and built the final designs ourselves.
It felt odd to design for AI workflows without working in one. We used AI where it sped up our thinking, and built the final designs ourselves.
Thinking and writing
Synthesizing research, and structuring how we explained it.
Claude
Chatgpt
Thinking and writing
Synthesizing research, and structuring how we explained it.
Claude
Chatgpt
Thinking and writing
Synthesizing research, and structuring how we explained it.
Claude
Chatgpt
Exploring layouts
Quick layout options to react to, with real AEP screens as the reference.
Lovable
Figma Make
Exploring layouts
Quick layout options to react to, with real AEP screens as the reference.
Lovable
Figma Make
Exploring layouts
Quick layout options to react to, with real AEP screens as the reference.
Lovable
Figma Make
Prototyping to test
Quick prototypes to try ideas ourselves and see how a pattern behaves, before building the real screens.
Lovable
Vercel
Prototyping to test
Quick prototypes to try ideas ourselves and see how a pattern behaves, before building the real screens.
Lovable
Vercel
Prototyping to test
Quick prototypes to try ideas ourselves and see how a pattern behaves, before building the real screens.
Lovable
Vercel
Building the design system
AI helped set up the components and styles our prototypes and screens were built from.
Claude Design
Chatgpt
Building the design system
AI helped set up the components and styles our prototypes and screens were built from.
Claude Design
Chatgpt
Building the design system
AI helped set up the components and styles our prototypes and screens were built from.
Claude Design
Chatgpt
You made it through this one
But I have got a few more worth seeing!
You made it through this one
But I have got a few more worth seeing!
You made it through this one
But I have got a few more worth seeing!
