Generative AI accelerates creation. Human-centred design (HCD) helps direct that speed toward problems worth solving and outcomes that people value.

Human understanding is a competitive advantage

Generative AI models are inherently prone to "regression toward the mean,” where outputs can exhibit less variation than real-world experiences.1 While prompting techniques can introduce greater diversity, organizations still need human insight to identify opportunities that are distinctive rather than merely probable.

Humans are remarkably good at finding ways around broken experiences, creating our own workarounds when products, services, or experiences are suboptimal. These are often undocumented and don’t show up in large language models.

HCD is an approach to problem-solving that puts people at the centre of the design process, using direct observation, conversations, and prototyping to uncover unmet needs, real-world behaviours, and to shape solutions. It’s a process that intentionally wrestles with divergence and extremes as ways to pressure test desirability and demand. As inclusive design has long shown, when we take the time to understand complexity, we come up with solutions that benefit a broader range of needs and contexts.

When we understand people better, we make better things for them.

Generative AI has lowered the barrier to creation: a few well-crafted prompts and some experimentation are often enough to turn an idea into something tangible. But speed doesn’t always mean value—we need to build things people want instead of just building things faster.

Case study: supporting inclusion

In collaboration with a national non-profit organization, we developed an AI-powered workplace accommodations tool that helps organizations offer consistent, effective, and more equitable disability accommodations.

Problem

Research found that it was a perceived challenge for employers to build the proper infrastructure and an inclusive culture such that they could confidently hire and retain persons with disabilities. We also discovered that persons with disabilities are often hesitant to disclose their disabilities due to fear of bias or negative career impacts. The team sought to explore if a GenAI-based solution could help. 

Solution

We conducted extensive user research, stakeholder interviews, and co-creation sessions with persons with disabilities to identify opportunities to improve workplace inclusion. By partnering with individuals with lived experience, we explored, tested, and refined promising concepts. We quickly learned that workplace inclusion is ultimately experienced most through interactions with direct managers. Even the most thoughtful policies and well-intentioned leadership can fall short if managers are not actively fostering inclusion every day. 

Through a series of collaborative design workshops, we ultimately co-created Adaptive Accommodations, a GenAI-enabled solution designed to support employees with disabilities and their direct managers with developing and managing personalized and effective workplace accommodations.  

Adaptive Accommodations is designed to provide personalized workplace accommodation recommendations, manager guidance, accommodation planning and tracking, proactive support, and streamlined workflows to make accommodations more accessible, timely, and consistent while maintaining privacy and human oversight.

We could not have developed a solution that resonated so strongly without the active participation of those we were designing for. Their participation helped us move rapidly to a working proof of concept while ensuring we were solving the right problem in the right way.

How Deloitte can help

It’s not a matter of generative AI versus human insights but how and when to leverage what each does best to differentiate and not just replicate.

Here are some questions that we’d love to explore with your organization.

Establishing problem-solution fit

• How and where can we leverage human insights to test assumptions that generative AI comes up with before something gets built?

Choosing and testing MVPs

• Given how easy and cheap it is to build a prototype, what’s the value of human insights at this stage?
  
  
• How and where can AI improve the prototyping process? Where do you need human input?

Launching new businesses

• How do we balance the trade-offs between speed to implementation when there’s no one answer and where services need to meet the needs of diverse populations and contexts?
  
  
• How do we preserve relevance and trust as our business offering scales?

Thank you to our contributors: Yasmine Dabbous & Lidiia Tulenkova


End Notes
Princeton University. “Variance reduction in output from Generative AI,” published March 2, 2025.