Universal design has long been regarded as a way to address the needs of the majority. Whether applied to physical spaces or digital CX, the idea of maximum usability was intended to remove the need for adjustments.
As an idea it held for decades, but in recent years we've started to realize that the more we optimize for the majority, the more the edge cases get left behind.
Now the question becomes: How do you build one platform that works equitably for all customers?
The answer is simple; designing for the hardest case rather than the median, improves both the median and edge case. For example, reducing friction for high-friction users improves outcomes for all users. The mindset change this requires, however, is far from simple.
Kapil Poreddy, senior software engineering manager for Walmart Global Tech, has dedicated his career to researching and addressing this point. He has two decades of experience across retail, healthcare, telecom and aviation and has conducted extensive research into cognitive load reduction and equitable access, concluding that populations under extreme financial stress and those who experience language barriers or have low digital fluency experience a higher friction in identical flows compared to the median user.
"Personalization AI ignores this reality. Equity-focused AI detects it and intervenes. That's the game changer," he says.
Based on Poreddy's interview at All Access: AI + Data in CX this article dives into his research in detail, explains how equity-focused AI reaches the underserved, and talks about how to design systems that help overcome the underlying challenges.
You can watch the full interview with senior event
producer Chloe Chappell on CX+
How can a single platform work for the majority?
There's a fundamental question that underpins Poreddy's work: Who are we accidentally leaving out?
Earlier in this career, Poreddy was part of the team building the online booking engine for Emirates. The system processes millions of transactions across dozens of languages for customers in almost every country in the world. "What you quickly learn at that scale is that the customer is never homogeneous," Poreddy says.
During his time with Castlight Health, he built a benefits navigation and population health platform that served millions of American workers. The idea was to support all users to find in-network care, compare the costs across providers and make decisions that worked for their situation.
Similarly to the Emirates example, the customer was never homogeneous, but in the healthcare industry, the stakes of inequity were more than commercial.
When user data was analyzed across income brackets, language preferences and health literacy levels, stark gaps in engagement emerged. The people who needed the platform most – those with more complex chronic conditions, lower health literacy, or less fluent English language skills – were least likely to engage with it.
"At a population level, even a one to two percent gap in this equitable engagement across millions of covered lives translates to a material health outcome difference and a significant downstream benefit to the healthcare system," Poreddy says.
"Inequitable CX has a dollar figure and a clinical consequence."
AI for personalization vs AI for equitable access
In light of these insights, Poreddy wants to see systemic change in experience design.
"Personalization AI optimizes for conversion. That's useful, but it's the platform serving its own interest using the customer's data," he explains. "There is a lateral shift in equity-focused AI, which has a different question: what does this person need from the system where it is currently failing to deliver? That's a fundamentally different design intent."
As Poreddy's research consistently confirmed, populations under extreme financial stress, those who experience language barriers, and people with low digital fluency experience a higher friction in identical flows compared to the median user.
It is here where equity-focused artificial intelligence (AI) becomes "the game changer".
Explaining how it works in production, Poreddy cites his research findings on health crisis detection. This analyzed behavior signals and combined them with cardiac health scoring. It identified a consistent correlation between behaviors and health outcomes that emerge weeks before an event and therefore, could act as a clinical signal.
He says of the outcomes: "The design question becomes how do you build a system that surfaces those signals and then routes people to care without being paternalistic or violating trust?"
The answer is a layered AI system, in which behavioral signal detection feeds into the population segmentation, which feeds proactive outreach design, which again feeds care navigation.
"Each of these layers is independently valuable, but the compound effect is a system that reaches the person before they are in crisis," Poreddy says. "That's how I see equitable AI and access in production."
Embedding equitable access in experience design
The ability to identify when a system or experience is not equitable is the first step to fixing it. This requires organizations to disaggregate performance metrics by population to understand how the experience works across all users.
Poreddy highlights three warning signs that could be hidden in the data:
- Aggregate metrics look great, but a closer look reveals the support volume for specific segments is elevated. "This shows the system is succeeding for most people, but failing a subgroup quietly," Poreddy says.
- Another signal could be high opt-out or disengagement rates for users in a specific language, geography or access category.
- Third is model scores are systematically lower for populations that are underrepresented in the training data. "This one is the most underappreciated signal, in my opinion," he says.
While the problem has wide-ranging impacts, Poreddy says it can be solved in only a few steps.
The first is to select a problem where the data already exists and the cost of inequity is visible to stakeholders. This doesn't require new data collection, but segmentation analysis of existing data.
The metrics used should drill down into language, device type, geography and income proxy.
"In my experience, this analysis alone surfaces two or three places where a specific population is experiencing that inequity and in a meaningfully worse outcome. That becomes your proof of concept," Poreddy says.
"You're not making a case or asking for permission to pursue equity as a principle. You're showing a concrete gap, a proposed intervention, and a measurable outcome. That's the language that gets budget approval," he continues.
The measurement of outcomes must reach beyond conversion. Poreddy tracks equity metrics or the engagement equity index, a ratio of the lowest access segment versus the highest access segment. Specifically, this examines if engagement with an experience predicts a downstream positive event and a retained relationship.
"The case to efficiency focused stakeholders is that equitable engagement is a leading indicator of long-term value. The cost of inequity shows up as your customer care calls, a high support volume and high churn in underserved segments where you'll see a lot of populations unsubscribing. All these flow back as cost to the system. It's not a value argument, I would say, it's a cost structure argument," he explains.
Is AI the problem or the solution?
New laws around accessibility in the EU and US are intended to address the inequality that exists for people who have difficulty accessing physical and digital experiences. They were introduced to revolutionize access for the populations they are designed for, but they fall short on two areas: equity and experiences that are driven by AI.
Research published by ArvatoConnect in June 2026 concluded that the use of AI – particularly in financial services – "risks worsening access for those who need it most". In fact, among 1,000 business leaders, 90 percent said they believe AI has the potential to exacerbate bias and digital exclusion for customers who need the most support.
Solving these challenges requires action at an individual organization level, but rather than making unrealistic demands to reduce the growing dependency on AI, the lasting solution is to build systems that are capable of detecting and addressing inequity when it occurs.
Poreddy says: "Stop asking the question, what can AI do? I think that ship has long passed. Start asking this question, 'hey, where is my system failing and how can AI fix it?'
"This first question will lead you to pilot projects and proofs of concept. The second leads to production systems that actually matter. In my 20 years of building platforms, the constant is this: the people who are most affected by these decisions are almost always the people who aren't making them," he continues.
"AI gives us that for the first time. It's a computational capacity to design for the full distribution of humanity that touches our systems. It's not just about the median user. And that's actually the opportunity."
The task now is to ensure we are using it that way.
Watch the full interview with Kapil Poreddy
from All Access: AI + Data in CX via CX+
Quick links
- How accessible is CX in 2026?
- Why accessibility still needs an internal advocate: Lessons from Deutsche Bank Singapore
- The Back Market story: Building customer trust in a low-frequency purchase category