Log in or create a free Rosenverse account to watch this video.
Log in Create free account100s of community videos are available to free members. Conference talks are generally available to Gold members.
Why AI projects fail (and what we can do about it)
Summary
Most AI projects fail. Somewhere between 50-90% of them, which is double the rate of more traditional tech projects. This Rosenverse Session will draw on years of Carnegie Mellon HCII research to dive into the five traps that AI projects can fall into, and then talk about what designers and project managers can do to avoid those traps. Including one startling finding: user-centered design alone isn’t enough.
Key Insights
-
•
Most AI projects fail due to choosing the wrong problems rather than poor execution.
-
•
The AI innovation gap occurs because data scientists focus on technically hard but low-value problems.
-
•
User-centered design alone often identifies problems unsuitable for AI solutions.
-
•
AI works best for low-risk, moderate accuracy (~90%), narrow tasks like spam filtering or smart text suggestions.
-
•
Executives push AI fast to avoid falling behind, risking deployments without clear user or business value.
-
•
Consequences scanning is a vital method to detect and mitigate ethical risks before AI product launch.
-
•
Matchmaking AI capabilities with real user needs and organizational goals increases project success.
-
•
Agents add new AI capabilities to sense, think, and act, requiring adapted design processes.
-
•
AI lacks context, aesthetic taste, and common sense—designers must provide these to create effective AI products.
-
•
There is currently little corporate or legal accountability for AI ethics; responsibility largely falls to product teams.
Notable Quotes
"Welcome to the AI party, we've got 10 years of research to share."
"Executives said it's better to go fast and fail than to go slow and succeed."
"AI can be magical, but it's also just not that smart."
"Don Norman was named the top accomplished woman in UX by AI, even though Don is a man."
"Companies hunt for AI innovation in technically challenging areas instead of simpler, more valuable places."
"Most AI fails because projects require near perfect accuracy, which AI can't reliably deliver."
"The traditional user-centered design process doesn't work well for AI problems."
"Matchmaking connects AI capabilities to user needs so you find the low hanging fruit projects."
"AI struggles with context, taste, and common sense, and designers bring that to the table."
"Consequences scanning helps detect unintended harms and risks before launching AI features."
Or choose a question:
More Videos
"If you launch in the US and politeness is an issue, first try to fix it with prompts; only if that fails should you build an eval."
Peter Van DijckBuilding impactful AI products for design and product leaders, Part 2: Evals are your moat
July 23, 2025
"The tapestry exhibition will be a three-dimensional map co-created with refugees and local school kids to reflect their stories."
Xenia Adjoubei Sean BruceEmpowering Communities Through the Researcher in Residence Program
March 29, 2023
"Astronauts hated digital controls; they wanted durable, glove-friendly hard buttons for the lunar rover."
James RamptonThe Basics of Automotive UX & Why Phones Are a Part of That Future
July 25, 2024
"I message participants regularly during studies to reduce no-shows and troubleshoot any issues immediately."
Lily Aduana Savannah Hobbs Brittany Rutherford5 Reasons to Bring Your Recruiting in-House (and How To Do It)
March 12, 2021
"Agents are business owners who are incredibly spread thin and scrappy in how they manage everything."
Kayla Farrell Chelsey Glasson Sean Fitzell Jared LeClercWhat It's Like To Be a User Researcher at Compass
March 12, 2021
"I recommend asking about user needs rather than disability, like whether someone needs captions or larger fonts."
Kate KalcevichIntegrating Accessibility in DesignOps
September 23, 2024
"Explainability requires transparency at every stage with governance ensuring fairness and equity in model building."
Ovetta SampsonResearch in the Automated Future
March 11, 2022
"Conversational analysis is not about replacing human understanding, it’s about amplifying it."
Deanna MitchellDesigning with culture: Unlocking impactful insights for Product and UX
March 12, 2025
"Graph databases use nodes and relationships instead of tables to store information, providing a network of knowledge."
Ren PopeBuilding Experiences for Knowledge Systems
June 6, 2023
