Roux Institute researchers reveal the hidden reason AI tools fail (hint: it's not the tech)

Your organization just spent months and significant resources building an AI product. The technology works. The algorithms are sound. So why isn’t anyone using it?

The problem is that you may have built the wrong thing in the first place.

The $100 Smart Speaker That Does Two Things

Dr. Melanie Tory, Professor of the Practice at Roux Institute, experienced this firsthand with her family’s smart speaker purchase.

“When we first got it, everyone was really excited. We had all kinds of ideas—look up recipes, do web searches, order things online,” Tory explained during a recent Roux Institute webinar. “And then as we tried all these things, almost all failed. The smart speaker just disappointed.”

This moment—when users realize the gap between promise and reality—is where AI products go to die. But Tory’s family found two things the speaker did well: serving as a kitchen timer and checking the weather. They adjusted their expectations and now use it consistently.

Understanding how to guide users through this journey is the difference between products that thrive and products that gather dust.

The Real Problem: AI-Centered Design

Traditional AI development focuses on technology first: build the most sophisticated algorithm, solve the hardest problems, create something impressive. Dr. Mahsan Nourani, Research Professor at Roux Institute, calls this “AI-centered design”—building tools without considering the people who use them.

The result? Recommendation engines that ignore expertise. Predictive models that don’t explain their reasoning. Dashboards that answer questions no one asked.

Human-Centered AI: A Different Approach

Human-centered AI (HCAI) builds technology around user needs from the beginning, creating “AI systems that amplify and augment people’s capabilities, rather than trying to displace those abilities,” as Nourani explained.

This approach de-risks development through three phases: Customer Discovery (understanding users before writing code), Co-Design (evolving solutions with stakeholders throughout development), and Continuous Evaluation (testing whether the product is relevant to users, not just whether it works).

Four Strategies for AI Products That Succeed

1. Design for Specific Use Cases

Stop trying to boil the ocean. AI tools that succeed solve one or two problems exceptionally well. What’s the one thing your tool does better than any alternative? Start there, do it exceptionally well, then expand based on user feedback.

2. Involve Users Throughout Design

Bring users into customer discovery, prototype design, and feature prioritization—not just beta testing. This continuous engagement serves as an early warning system. Feedback is infinitely cheaper to address during design than after launch.

3. Set Realistic Expectations Early

Be transparent about what the AI can and cannot do, when it’s likely to make mistakes, and how confident it is in predictions. Honesty about limitations builds calibrated trust—helping users know when to rely on AI and when to rely on their own expertise.

4. Empower Users Through Strategic Training

Design onboarding and contextual help that guide users toward successful patterns of use. Training should help users discover how the tool fits into their workflow.

The Bottom Line

Building AI products that people actually use isn’t primarily a technology challenge—it’s a design, user research, and trust-building challenge.

Organizations that embrace human-centered AI principles are building products that succeed where others fail. The question is whether you’re willing to invest in understanding your users as much as you invest in perfecting your algorithms.

As Nourani emphasized: “By knowing more about what trust means in the context of your users and your product, you can help them navigate trust and support trust calibration.”

Your AI product doesn’t have to end up as an expensive kitchen timer. But getting from the drawing board to meaningful adoption requires human-centered design, continuous user engagement, realistic expectations, and strategic rollout.

Ready to Build AI Products That Last?

The Human Data Interaction Research Team at Roux Institute brings interdisciplinary expertise in human-centered AI, data science, and domain-specific applications. Whether you’re just beginning to explore AI adoption, building new products, or looking to improve existing tools, our team can help.

Connect with our team to discuss your AI strategy and product development needs below!

Watch the full webinar: Human-Centered AI: Turning Complexity Into Better Products

Northeastern University’s Roux Institute is activating an innovation economy across northern New England through graduate education, translational research, and entrepreneurship programs. Our research faculty combine academic excellence with real-world industry application.