AI in process excellence: Laying the foundation for enterprise AI success

The 4th Annual All Access: AI PEX webinar series brought together thought leaders and experts from Google, SAP, Camunda, Strategy, and CGS to address the critical challenge facing organizations today: moving AI initiatives from pilot to production. Across multiple sessions, speakers emphasized that successful AI adoption requires foundational work in data quality, process excellence, and organizational readiness before technology deployment.

You can watch all sessions on-demand here.

Google’s approach to fixing data quality 

The day opened with Google's Rahul Chawla and Puneet Thakkar reinforcing findings from PEX report 2025/26 which states that 52% of leaders say data quality and availability are the top barrier to AI adoption.

Google's approach centers on establishing a semantic layer that acts as a universal translator between legacy systems and modern AI tools. Rather than waiting years for cloud migration, this architecture allows AI agents to access on-prem data through a governed abstraction layer. The semantic layer maintains business definitions and data lineage, ensuring AI outputs remain traceable and trustworthy.

Shifting the focus from potential to realized value

Seth Lippincott and Syeda Noor Zehra Naqvi from SAP discussed how generative AI has moved quickly from experimentation to real-world application. As AI agents become more common in the enterprise, organizations need to identify where AI can create value, track, and realize it.

That shift is driving a more disciplined approach to transformation. Instead of starting with technology, leading teams are starting with value potential and identifying where impact is possible, what metrics matter, and how success will be measured.

As Zehra says in the AI-driven business transformation report: “AI-driven business transformation represents a fundamental leap beyond traditional automation... By leveraging natural language and agentic assistants, this new partnership moves expertise from a siloed resource to an on-demand capability. The result is the democratization of critical insights, empowering a broader range of employees to solve complex problems and execute sophisticated tasks with greater autonomy.”

Reimagining workflows for agentic AI

Camunda's Lana Stawowski and Leon Strauch introduced the concept of "agentic orchestration" - blending deterministic processes with dynamic AI execution while maintaining human oversight. They warned against the "fragmented enterprise" trap where organizations add AI point solutions to disconnected systems, creating what they called "supercharged fragmentation" rather than seamless end-to-end experiences.

Their framework addresses this through four phases: discovering where agents can unlock value, building agents with proper business context, mining agent execution data to ensure compliance, and managing agents at scale.

Semantic layers enable trusted AI

Lauren O'Connor and Erika Moreno from Strategy demonstrated how AI-powered semantic layers address the integration nightmare many organizations face.

Strategy’s platform uses AI to accelerate data model creation and can achieve coding 10x faster development than traditional hand-coding, while ensuring consistency across all downstream applications.

This capability addresses a critical governance gap by detecting anomalous data access patterns and agents can access governed data to build applications in minutes, knowing the underlying numbers are trustworthy.

Moving from pilots to measurable ROI

Doug Stephen, CGS Immersive CEO tackled the statistic haunting every AI initiative: 95% of pilots fail to reach production. His framework focuses on three non-negotiables: clear business ownership with committed KPIs, workflow redesign that accounts for AI's impact on roles, and a repeatable operating model that can scale.

He demonstrated this with a knowledge capture use case where retiring experts are interviewed by AI agents trained on SOPs and job descriptions. The system automatically generates documentation, learning content, and performance support, reducing time-to-competency by 45% and first-time-right execution by 22%.

Doug's checklist emphasizes data access, human-in-the-loop validation, daily workflow integration, and usage measurement. Organizations scoring 8 out of 10 on these criteria have a strong likelihood of production success. Equally important: starting with focused use cases rather than ambitious transformations that create ammunition for skeptics when they inevitably stumble.

AI governance without stifling innovation

The expert panel featuring Andreas Welsch, Wayne Butterfield, and Lucas Root tackled the governance paradox: how to enable innovation while maintaining control. Their consensus: AI naturally converges rather than diverges, making it less prone to drift than feared, but organizations must still establish frameworks for approved tools, training, and accountability.

The panelists noted that adoption, not ROI, should be the primary focus for foundational AI initiatives and emphasized measuring beyond productivity toward process performance indicators, as productivity gains alone often get absorbed by non-work activities.

The panel agreed that system thinking helps leaders see the "machine" and identify where AI creates leverage versus where it fragments operations. They recommended finding internal champions who experiment with AI on weekends, evaluating existing vendor capabilities before adding new tools, and always starting with the problem rather than the solution.

Essential capabilities for process leaders

Doug Shannon's closing conversation emphasized returning to systems thinking fundamentals. He argued that most "processes" are actually disconnected workflows performed differently by each team member. True processes account for software, human touchpoints, data flows, platforms, and upstream/downstream dependencies.

The future belongs to organizations that build "centers of intelligence" rather than traditional centers of excellence, federating AI capabilities while maintaining centralized governance over APIs, LLMs, and data access. Process leaders must develop infrastructure awareness, understanding how hardware and architecture enable (or constrain) AI possibilities.

Key Takeaways for AI Implementation

• Start with data governance: Without trusted, governed data layers, AI agents amplify existing problems. Invest in semantic layers that provide consistent definitions across all consumption points.
• Design for hybrid workflows: The most effective implementations blend deterministic orchestration with dynamic agent execution and strategic human oversight—not full autonomy.
• Measure what matters: Establish baseline KPIs before implementation, track usage and outcomes, and focus on process performance rather than just productivity gains.
• Enable through governance: Create frameworks that allow innovation within guardrails rather than either stifling experimentation or allowing ungoverned proliferation.
• Prioritize change management: The people on the ground know where processes break. Involve them in discovery, provide training on new tools, and communicate why changes benefit both the business and employees.

You can watch All Access: AI in PEX on-demand here.