Danny Stout on human-centered AI teams, the myth of bigger models, and why communication skills trump technical prowess

Listen now on YouTube | Spotify | Apple Podcasts

*The Data Faces Podcast with Danny Stout, Product Lead of the Intelligence Layer at EY*

It’s no secret that a large percentage of AI projects, some estimates range from 70% to 85%, never make it to production or fail to deliver on their promises. While headlines focus on large language models (LLMs) like how many billions of parameters, training times, and novel model architectures, the true determinants of AI success often lie with the people implementing these applications. I recently spoke with Dr. Danny Stout, Product Lead of the Intelligence Layer at Ernst & Young (EY), for the Data Faces podcast to uncover actionable strategies for improving AI project outcomes through better team dynamics.

About Danny Stout

Danny Stout is the Product Lead of the Intelligence Layer at EY and a self-described “data scientist by accident.” With a PhD in educational psychology, Danny brings a human-centered perspective to AI development. His career spans roles at STATISTICA, Dell, and TIBCO, where he focused on developing predictive analytics systems for enterprise customers. Before joining EY, he served as global head of forecasting, pricing, and analytics at Takeda. In our conversation, we discuss:

Why bigger AI models aren’t always better
The three critical roles needed for AI success
How team alignment determines project outcomes
Real-world lessons from a diamond mine theft prediction case
Why communication skills matter more than technical prowess
The evolution of technical skills in AI
Practical recommendations for staying relevant in AI

This blog offers frameworks, real-world examples, and specific methodologies to help data scientists and AI practitioners, and their leaders, build more effective teams, achieve stakeholder alignment, and deliver AI solutions that create measurable business value.

The myth of bigger models: Effectiveness over size

When every AI company seems to be competing on model size, Danny offers a refreshing counterpoint backed by his decades of experience.

The bigger model is not better, the bigger application is not better, the bigger LLM is not better. – Danny Stout, Product Lead of the Intelligence Layer at EY

Danny points to a consistent pattern in analytics history supporting this view. Ensemble models, which combine multiple smaller models, have consistently outperformed single large models in predictive analytics. This pattern continues with today’s language models, where specialized small language models (SLMs) often deliver better results than their larger LLM cousins for specific tasks.

He shared a concrete example from his recent work. His team built a specialized language model for a financial services application, assuming it would outperform popular models like Azure’s offerings or Mistral’s 7B parameter models. Despite investing significant resources in custom training, their specialized model performed 15% worse than the general-purpose models in accuracy and response quality.

Model Selection Framework

For AI practitioners and their teams, Danny recommends this evaluation process when choosing between custom and existing models:

Define success metrics clearly tied to your use case
Test existing models against these metrics before building custom solutions
Consider the total cost of ownership, including maintenance and updates
Run small-scale pilots with business users before full implementation

This methodical approach prevents wasting resources on unnecessary custom development. It echoes what I have always advocated for: start with the business problem and work backward to find the right technical solution.

However, selecting the right technology is only part of the challenge. Without the right team composition, even the perfect model will fail to deliver business value.

Building the optimal AI team: Three critical roles

Successful AI implementation depends on having the right mix of skills and perspectives on your team. Danny identifies three essential roles that form the foundation of high-performing AI teams.

I think of diversity as kind of like the base of a pyramid. The top of the pyramid is only going to be as high as the base will support. – Danny Stout, Product Lead of the Intelligence Layer at EY

1. The Technical Lead (Technology Specialist → Generalist)

In organizational terms, this might be a Senior Data Scientist or ML Engineer. Today’s ideal technical lead is a versatile generalist rather than a narrow specialist. They need:

Broad knowledge across multiple AI frameworks and approaches
Ability to evaluate which technologies fit specific use cases
Practical implementation skills for rapid prototyping
Understanding of integration requirements with existing systems

This marks a change from past requirements. “That first person is typically a generalist and not a specialist, whereas in the past, it may have been a specialist,” Danny explains. Instead of needing deep expertise in a single algorithm type, today’s technical leads must understand the strengths and limitations of various approaches.

2. The Business-Technology Translator (Product Manager)

Often holding titles like AI Product Manager or Business Analyst, this person bridges the technical and business worlds. They typically originate from business roles but develop technological literacy over time. Key responsibilities include:

Translating business requirements into technical specifications
Communicating complex concepts in business-friendly language
Identifying metrics that matter to stakeholders
Balancing technical possibilities with business priorities

“Unless you have that ability to translate between the two, you’re not going to get anything done,” Danny emphasizes. This role prevents the common problem of building technically impressive solutions that fail to address actual business needs.

3. The Leadership Bridge (Relationship Builder)

This person, often in a Director or VP role, creates organizational alignment and support. They understand power dynamics and build relationships with key stakeholders, particularly at the executive level. Their responsibilities include:

Identifying and cultivating executive champions
Navigating organizational politics
Securing resources and removing obstacles
Linking AI initiatives to strategic business priorities

Team diversity strengthens the entire project. Different perspectives lead to more robust solutions, better risk identification, and greater innovation. The pyramid rises only as high as its base allows—the more diverse your team, the more successful your AI implementation will be. For business leaders, the primary action is to consciously structure teams to include, or identify individuals who can fulfill, these three distinct roles within their AI initiatives, ensuring these individuals are empowered to collaborate effectively.

Alignment: The foundation of AI success

Before a single line of code is written or a model is trained, team alignment must be established. Danny identifies this as the most common reason AI projects fail.

It’s absolutely vital, if the team is not aligned beforehand, there’s no way that whatever model you choose is going to be successful. – Danny Stout, Product Lead of the Intelligence Layer at EY

Many organizations rush to technological implementation without ensuring agreement on the fundamental problem they’re trying to solve. This misalignment leads to wasted resources, frustrated teams, and solutions that don’t address business needs.

Measuring Alignment: 5 Essential Questions

Danny recommends teams answer these questions (and business leaders should ensure this occurs) before proceeding with AI projects:

What specific business problem are we solving?
How will we measure success?
Who are the key stakeholders and what are their priorities?
What skills and resources do we need to succeed?
What are the potential obstacles or concerns from each team member?

Documenting answers to these questions creates a shared reference point and prevents mid-project confusion.

Alignment in Practice

Danny shared experiences from his time at Dell, where his team replaced established predictive analytics platforms. “When we invested the time and energy going in beforehand and establishing that alignment, it went really, really smooth,” he explained. In contrast, implementations rushed without alignment consistently faced resistance and delays.

He warns against punitive approaches some organizations take when forcing new technology adoption. When leadership mandates change without addressing team concerns or skill gaps, resistance grows and projects falter. Signs of poor alignment include:

Stakeholders disagreeing about project priorities
Technical and business teams using different terminology for the same concepts
Frequent scope changes or “requirement creep”
Team members unable to clearly articulate the project’s purpose

Successful alignment requires input from product teams, engineering, architecture, and leadership. Creating this foundation takes time upfront, but dramatically increases the probability of successful implementation and adoption. Leaders foster this by making these discussions a non-negotiable prerequisite for project initiation, not merely a checkbox exercise.

Communication: The career-defining AI skill

Of all the capabilities that drive success in AI, Danny places communication at the top of the list – both for project outcomes and individual career advancement.

Without those presentation skills, they’re going to be lost. They’re going to be stuck in the position that they’re in, and they’re not going to be able to grow personally and professionally. – Danny Stout, Product Lead of the Intelligence Layer at EY

Characteristics of Effective AI Communication

What distinguishes effective communicators in the AI field? Danny identifies several key traits:

Audience adaptation: Tailoring technical complexity to the listener’s background
Visualization skills: Converting abstract concepts into accessible visuals
Narrative ability: Framing AI solutions within relevant business stories
Translation: Converting technical jargon into business terminology and vice versa
Focusing on outcomes: Emphasizing business impact over technical details

The most common communication failure he observes is technical experts who communicate at their own level of understanding rather than adapting to their audience’s needs. “I’ve been able to go in, look at what was being done, present it differently, not change any of the technology, and automatically get stakeholder buy-in,” Danny explains.

Practical Communication Development

For professionals looking to strengthen this skill, Danny recommends these specific approaches:

Co-presentation opportunities: Ask to present alongside experienced communicators and observe their techniques. Leaders can facilitate this by pairing junior and senior team members.
Structured practice: Join Toastmasters or take formal communication courses.
Internal presentations: Volunteer to present project updates to stakeholders at varying technical levels.
AI-specific communication exercises: Practice explaining the same AI concept to different audiences (e.g., an engineer, a marketer, an executive).

Dual Communication Challenge

Today’s AI professionals face two distinct communication challenges:

They must communicate effectively with humans across the organization, from fellow technical experts to business leaders with limited technical knowledge.
They must communicate precisely with AI systems through prompts and instructions, which requires understanding how these systems interpret language and instructions.

By developing strong communication skills in both domains, AI practitioners not only improve project outcomes but also distinguish themselves for leadership roles and career advancement. Business leaders who invest in communication training and provide opportunities for such practice often see a direct improvement in project clarity and stakeholder satisfaction.

The evolution of technical skills in AI

The technical skill requirements for AI professionals are changing rapidly.

The rising importance of guardrails

Danny highlights industry shifts that practitioners should understand to remain relevant. ne of the most critical is the implementation of robust operational guardrails.

We did not build the guard rails into that in order to make sure you’re not accidentally spinning up something that you don’t need… it spun up GPU that over the course of three events incurred about a quarter of a million dollars have cost. – Danny Stout, Product Lead of the Intelligence Layer at EY

This expensive lesson highlights why every AI implementation needs effective guardrails:

Resource limits: Caps on computational resources, API calls, and storage.
Cost thresholds: Automatic alerts when spending reaches certain levels.
Approval workflows: Human verification for high-impact actions.
Monitoring systems: Real-time tracking of system behavior and outputs.

Without these guardrails, even well-designed AI systems can create significant unexpected costs or unintended consequences. Business leaders must ensure that establishing and monitoring these guardrails is a standard operational procedure, not an afterthought, to protect against such financial and operational risks.

Technical Skills for 2025-2026

Looking forward, Danny sees these emerging technical skills becoming increasingly valuable:

Prompt engineering: The ability to craft effective instructions for generative AI systems.
Systems integration: Connecting AI capabilities with existing business processes and tools.
Data quality management: Ensuring training data is representative, unbiased, and appropriate.
Technical risk assessment: Identifying potential failure modes and mitigation strategies in AI systems.
Multi-modal development: Working with combinations of text, image, voice, and numerical data.

The Soft Skills Advantage

Perhaps counterintuitively, soft skills are becoming more important for technical professionals as AI tools become more accessible. Danny observes that teams who excel at interpreting human needs and translating them into effective solutions consistently outperform those focused solely on technical sophistication.

Practical recommendations for AI practitioners

Danny offers concrete, actionable advice for AI professionals looking to stay relevant and advance their careers in a rapidly changing field. Business leaders can also foster these practices within their teams to cultivate a more adaptable and effective workforce.

The technology we have now makes it accessible to anyone. Previously, you had to be able to code in a particular language and understand stochastic gradient boosting in order to develop a good application. – Danny Stout, Product Lead of the Intelligence Layer at EY

1. Schedule exploratory “play time” with AI applications

Danny allocates approximately one hour weekly to explore AI applications outside of work requirements. This isn’t casual browsing—it’s structured exploration:

Explore multiple environments: Test both production systems and those in user acceptance testing (UAT).
Document discoveries: Keep notes on unexpected capabilities or limitations.
Try edge cases: Intentionally push systems beyond their normal operating parameters.
Cross-pollinate ideas: Apply techniques from one domain to entirely different problems.

“I dedicate maybe an hour a week to go out and just play with the application, not just what is in production, but also what’s in UAT,” Danny explains. “I make sure to talk to people about what I’m finding when I’m playing with it, and it’s helping me to develop skills and also contribute to the roadmap.”

This systematic exploration helps identify capabilities others miss and contributes valuable insights to product development. Leaders can encourage this by allocating time for such exploration and creating forums for sharing these discoveries.

2. Develop a communication improvement plan

Communication skills require deliberate practice. Create a concrete improvement plan:

Month 1: Volunteer for at least two presentation opportunities within your team
Month 2: Request feedback from a mentor on your communication style
Month 3: Take a formal course or join a speaking group like Toastmasters
Month 4: Present to cross-functional teams or leadership. Consistent practice, rather than relying on natural talent, leads to communication mastery.

Track your progress using metrics like audience feedback, questions asked, and implementation of your recommendations. Communication mastery comes through consistent practice rather than natural talent.

3. Position yourself at the technology-business intersection

The democratization of AI creates new opportunities. As Danny explains, “The technology we have now makes it accessible to anyone.” This shift means:

Technical experts need business understanding to remain relevant
Business professionals with basic AI literacy can contribute more directly
The most valuable professionals bridge both domains

Build your career at this intersection by:

Learning the language and priorities of the business side
Staying current with AI capabilities without getting lost in technical details
Focusing on outcomes and value rather than the technology itself

4. Apply the Occam’s Razor principle

When evaluating approaches, start with the simplest viable solution:

Test basic regression models before complex neural networks
Use existing models before building custom solutions
Deploy minimum viable products before perfect systems

To evaluate your progress in implementing these recommendations, consider tracking metrics such as:

Project completion rates
Stakeholder satisfaction scores
Career advancement milestones
Implementation of your ideas across the organization
Expansion of your responsibilities and influence

Final Thoughts

Throughout our conversation, Danny repeatedly returned to a simple but powerful principle that should guide AI practitioners.

Occam’s Razor is real. The simplest solution is typically the one that’s going to fit, not just for the problem, but for the business as well. – Danny Stout, Product Lead of the Intelligence Layer at EY

This insight runs counter to many data scientists’ natural instincts. Technical professionals often gravitate toward complex, novel solutions, particularly in AI where new models and techniques emerge weekly. But Danny’s experience shows that simpler approaches frequently deliver better business outcomes. A basic multiple regression might outperform a sophisticated large language model for specific use cases.

Key Takeaways

Team composition over model sophistication: Build teams with complementary skills covering technical, business translation, and leadership connection roles.
Alignment as the foundation: Use the five essential questions to ensure all stakeholders agree on the problem, approach, and success metrics before beginning implementation.
Communication drives career advancement: Deliberately practice presenting to different audiences and adapting your message to their needs and background.
Implement proper guardrails: Prevent costly mistakes by establishing clear limits and monitoring systems for AI applications.
Schedule structured exploration time: Dedicate regular time to explore AI applications beyond immediate work requirements to discover unexpected capabilities.

About David Sweenor

David Sweenor is an AI, Generative AI, and Product Marketing Expert. He brings this expertise to the forefront as founder of TinyTechGuides and host of the Data Faces podcast. A recognized top 25 analytics thought leader and international speaker, David specializes in practical business applications of artificial intelligence and advanced analytics.

Books

With over 25 years of hands-on experience implementing AI and analytics solutions, David has supported organizations including Alation, Alteryx, TIBCO, SAS, IBM, Dell, and Quest. His work spans marketing leadership, analytics implementation, and specialized expertise in AI, machine learning, data science, IoT, and business intelligence.

David holds several patents and consistently delivers insights that bridge technical capabilities with business value.

Follow David on Twitter @DavidSweenor and connect with him on LinkedIn.

Podcast Highlights – Timestamps

Podcast Highlights – Timestamps

0:05 – Introduction
David Sweenor introduces the Data Faces podcast and today’s discussion about the human impact of AI and analytics. Introduces guest Dr. Danny Stout, Product Lead of the Intelligence Layer at EY.

0:45 – Danny’s Background
“I call myself a data scientist by accident. I really thought that I would hate statistics and anything related to the math part of humanity.” Danny explains his journey from educational psychology to data science, including his dissertation on measuring medical knowledge and epistemology.

3:16 – Beyond Model Size
Danny challenges the tech obsession with model size: “The bigger model is not better, the bigger application is not better, the bigger LLM is not better.” He explains how smaller ensemble models often outperform larger ones.

6:00 – The Importance of Executive Alignment
“It’s absolutely vital, if the team is not aligned beforehand, there’s no way that whatever model you choose is going to be successful.” Danny discusses how alignment issues often doom AI projects from the start.

7:29 – Alignment Best Practices
“A lot of people want to go right for the technology and right to go for the solution, without making sure to look and see where people are at before they even join the team.” Danny shares his approach to gaining alignment across stakeholders.

10:22 – The Three Critical Team Roles
Danny outlines his first three hires for an AI team: a technologically gifted person, a business-technology translator, and someone who understands leadership dynamics and power structures.

14:52 – The Evolution of Technical Skills
“That first person is typically a generalist and not a specialist, whereas in the past, it may have been a specialist.” Danny discusses how AI is changing the required skillsets for technical team members.

17:56 – The Growing Importance of Soft Skills
Danny explains why soft skills are becoming increasingly important in AI work as natural language capabilities expand.

21:54 – The Diamond Mine Case Study
Danny shares a fascinating real-world example of using predictive analytics to detect diamond theft, demonstrating how human insight and technical solutions work together.

24:35 – AI Project Blind Spots
“Occam’s Razor is real. The simplest solution is typically the one that’s going to fit, not just for the problem, but for the business as well.” Danny discusses common misconceptions in AI projects.

27:22 – The Importance of Guardrails
Danny shares a cautionary tale about an AI system that incurred a quarter-million dollars in unexpected costs due to lack of proper guardrails.

30:19 – Communication as a Career Differentiator
“Without those presentation skills, they’re going to be lost. They’re going to be stuck in the position that they’re in, and they’re not going to be able to grow personally and professionally.” Danny emphasizes communication as the most important skill for advancement.

33:57 – Practical Advice for AI Practitioners
“What I’d recommend them do is actually dedicate some time to play with the applications.” Danny offers actionable recommendations for staying relevant in a rapidly changing field.35:58 – Conclusion
Danny summarizes his approach to human-centered AI development and the growing accessibility of AI tools to non-technical users.