AI Doesn’t Fail in Healthcare. Workflows Do

Why Value-Based Care Needs Product Thinking Before AI

By Kiana Padash, Sr. Product Manager at WellStack

AI has become table stakes in value-based care. Nearly every organization is racing to deploy it, yet most AI initiatives stall quietly in pilots and never become part of daily operations.

From a product perspective, this failure is rarely about the model. Value-based care is a data and workflow problem long before it is an AI problem. When AI gets introduced without understanding how people actually work, what data they can trust, and how decisions are actually made on the ground, even the most accurate models fail to deliver real impact.

At its core, AI stalls in value-based care not because the technology fails, but because data, workflows, and people are rarely designed together.

Start With the Workflow, Not the Model

One of the most common mistakes I see teams make is starting with the model instead of the workflow. Product decisions begin with what AI can do rather than how people actually work.

Care teams are already juggling multiple systems every day. If a care coordinator is bouncing between an EMR, a care management platform, and three reporting tools, adding yet another standalone AI interface is almost guaranteed to fail. Even the most advanced analytics lose their value the moment they require someone to step outside their existing workflow to act on them.

This is why analytics that live only in dashboards or leadership views rarely change outcomes. For AI to matter, insights have to reach providers and care teams in a way that fits how decisions are actually made in real time, at the point of care.

A common example: quality care gap detection. AI models might surface hundreds of patients who are overdue for screenings or follow-ups, but without prioritization aligned to real staffing capacity, those insights overwhelm teams rather than enable them. Until AI is embedded directly into how people work and helps them decide what to do next, it stays theoretical instead of operational.¹

Data Foundations Are Not Optional

Value-based care is a data infrastructure problem before it is a clinical one. Strong workflows alone cannot compensate for weak data foundations.

Patient matching, claims alignment, incomplete historical data, inconsistent sources across systems. These things quietly undermine trust in AI outputs if they are not addressed first. I have seen this play out firsthand: when users encounter discrepancies, missing context, or unexplained results, confidence erodes fast and adoption stops.

Cleaning up the foundational data layer is not glamorous work. It does not feel as exciting as deploying new models, but it is what allows dashboards, analytics, and AI-driven insights to actually move faster downstream. Once the data foundation is solid, teams spend less time questioning numbers and more time acting on them.

And the research backs this up. Poor data quality is one of the most common reasons AI initiatives fail, particularly in healthcare’s fragmented data environment where claims definitions do not match Vizient, do not match Clarity, and do not match clinical definitions.²

Accuracy Is Not the Same as Impact

Many AI solutions over-optimize for sensitivity and confidence scores while overlooking whether the tool actually changes behavior. Accuracy is important, but it is not enough.

In value-based care, impact shows up in whether teams can identify risk earlier, prioritize outreach effectively, and change patient trajectories. Not in how precise a model looks on paper. A highly accurate model that nobody understands, trusts, or uses does not improve outcomes. Trust plays a central role here, and so does accessibility. When AI outputs are only understandable to technical users, adoption stays limited. The more clearly insights are presented to non-technical users alongside analysts and data scientists, the more likely those insights are to influence daily decision-making.

Research consistently shows that perceived usefulness and trust influence adoption more than model accuracy alone. Success has to be measured by usage and outcomes, not technical performance in isolation.³

The Human Element Is the Make-or-Break Factor

Fear, mistrust, and job insecurity block AI adoption more than technology does. In healthcare especially, people can never be removed from the loop.

Clinicians, care coordinators, and analysts need to trust that AI supports their work rather than threatens it. When AI tools are introduced without transparency or context, teams worry about losing autonomy or relevance. That leads to resistance before value is ever realized.

When done well, AI shifts work rather than replaces it. Data analysts spend less time wrangling data and more time generating insights that were previously out of reach. Care teams spend less time searching across systems and more time engaging the patients who need attention most.

Human-centered design and early clinician involvement are not nice-to-haves. They are foundational to trust, adoption, and long-term impact.⁴

What “Doing It Right” Actually Looks Like

When AI is done well, it does more than surface insights. It clarifies priorities.

Care teams gain a unified view of the patient journey across systems, allowing them to see which cohorts carry the highest cost and quality risk. Leaders move from reactive reporting to proactive planning, acting earlier before performance or outcomes suffer.

Rather than replacing workflows, effective AI optimizes them. It brings analytics back into the hands of providers and care teams at the moments when decisions are being made, enabling action instead of observation.

The data supports this too. Workflow-aligned tools consistently see higher adoption and sustained use, reinforcing the importance of designing AI around people rather than around models alone.⁵

How to Make AI Stick

Several patterns consistently emerge across successful implementations:

Start with workflows, not models

Understand how people actually work before introducing new technology.

Invest early in data foundations

You cannot build trust on top of unreliable data.

Design feedback loops from day one

If users cannot tell you what is working & what is not, you are flying blind.

Measure adoption and behavior change, not just accuracy

A model that nobody uses is a model that does not matter.

Keep humans in the loop at every stage.

AI should support clinical judgment, not bypass it.

These principles are reinforced across research on trust, usability, and human-centered design in healthcare AI. They serve as practical guideposts for moving beyond pilots and into real operations.³⁴⁵

Closing Thought

“The future of AI in value-based care will be defined by whether it makes patient care better. The technology matters, but people, trust, & thoughtful design matter more.”

“Success will not belong to the organizations that adopt AI first. It will belong to those that build it carefully around data, workflows, and the people who rely on it every day.”

ABOUT KIANA PADASH

Sr. Product Manager, WellStack

Senior Product Manager who focuses on building data and AI-driven solutions for value-based care. She specializes in translating complex healthcare data challenges into scalable, workflow-aligned products that drive measurable outcomes. With a strong focus on product thinking, human-centered design, and operational impact, Kiana works at the intersection of data, analytics, and care delivery to ensure technology is not just built—but adopted and trusted in real-world healthcare settings

References

Workflow misalignment and usability as barriers to clinical AI adoption. ScienceDirect
Estimates that up to 85% of AI model failure stems from poor data quality. Orion Health
Trust and perceived usefulness as drivers of AI adoption. JMIR
Human-centered design and clinician involvement in healthcare AI. Federation of American Scientists.
Sustained adoption of workflow-aligned clinical decision support systems. ScienceDirect
Execution gaps in value-based care driven by technology and data challenges. Mathematica
Surveys of clinical AI success rates in real-world health system deployment. PMC

Patient / Member 360

Decision Hubs

Population Health management

Accountable Care Organizations

Value Based Care

AI/ML Enablement

WellStack Labs

Company

Insights

News

Partnerships

Case Studies

Thought Leadership

Careers

AI Doesn’t Fail in Healthcare. Workflows Do

Why Value-Based Care Needs Product Thinking Before AI

By Kiana Padash, Sr. Product Manager at WellStack

Start With the Workflow, Not the Model

Data Foundations Are Not Optional

Accuracy Is Not the Same as Impact

The Human Element Is the Make-or-Break Factor

What “Doing It Right” Actually Looks Like

How to Make AI Stick

Start with workflows, not models

Invest early in data foundations

Design feedback loops from day one

Measure adoption and behavior change, not just accuracy

Keep humans in the loop at every stage.

Closing Thought

ABOUT KIANA PADASH

References

AI Doesn’t Fail in Healthcare. Workflows Do