Artificial intelligence (AI) has become the most sought-after capability in business today. Across boardrooms and business units, leaders are eager to deploy it, teams are experimenting enthusiastically, and every function from operations to finance to customer service believes AI will unlock the next wave of efficiency and competitive advantage.
The excitement is well deserved. AI feels powerful, fast, intelligent and, at times, almost magical, but beneath this optimism lies a more sobering reality: most organizations are not yet realizing meaningful, scalable value from AI.
Outputs often require verification. Models behave inconsistently. Data quality exposes limitations. The promised productivity gains quietly dilute under layers of manual checks, exceptions, and rework. The truth is not that AI is disappointing; it’s that organizations are not structurally ready for AI. This is precisely where the ‘new process excellence’ becomes indispensable.
This article explores why AI still behaves like a shiny new toy in many organizations, full of potential but limited in performance and what leaders must do to convert that potential into transformation.
The illusion of AI efficiency
To give AI its due, it excels at language tasks. Summaries, drafts, sentiment analysis, document structuring, and pattern detection are areas where generative models perform impressively well. These ‘micro-wins’ genuinely help teams work faster.
However, the moment AI is asked to operate with precision, judgement, or multi-source data, its limitations become obvious. Inconsistencies appear. Numbers drift. Hidden assumptions surface. The human review effort increases.
Where AI works today:
- Summaries and meeting notes
- First-draft narratives
- Risk flags and anomaly detection
- Data extraction from documents
- Organizing large unstructured inputs
- Pattern identification in operational workflows
Where AI still struggles:
- Financial projections
- Creditworthiness analysis
- Multi-year trend analysis
- Highly regulated decision steps
- Data reconciliation
- Cross-functional, complex workflows
These high-value processes demand accuracy, traceability, and explainability, qualities that current generative models cannot reliably guarantee without robust human oversight.
Register for All Access: AI in PEX 2026!
What real-world AI adoption reveals
In my current role at a small but rapidly scaling financial institution, we have been testing AI across a wide spectrum of real operational use cases, from preparing underwriting packs and credit memos, to automated data extraction, anomaly detection, and portfolio review support. Some of these use cases perform exceptionally well.
AI summarizes dense credit files in seconds. It surfaces discrepancies that analysts might otherwise miss. It structures narrative sections with speed and consistency. However, when the task shifts from language to logic, cracks appear. We’ve seen AI generate coherent financial summaries accompanied by numbers that still require manual verification. Multi-source reconciliations and complex trend analyses continue to challenge most models.
When every generated output must be checked line by line, the time savings shrink rapidly. Instead of reducing effort, AI creates a new layer of activity: validating the machine. This is not a failure of AI – it’s a diagnostic signal. It tells us exactly where our processes need redesign, where data foundations must be strengthened, and where human-in-the-loop controls are essential. It also reminds us that AI cannot compensate for upstream process variability or inconsistent data quality.
Why AI fails: The structural reality
Most AI failures are not technology failures, they are process, data, and governance failures. Five structural issues surface consistently across organizations.
- Unstable processes: AI requires repeatable patterns. If every team executes a process differently, the model learns inconsistency and performs unpredictably.
- Weak data foundations: AI magnifies both the quality and the flaws of your data. Duplicated, incomplete, siloed, or ungoverned data leads to unreliable outputs.
- Undefined governance and ownership: When AI ownership is unclear (split between business, tech, risk, and data) no one owns accuracy, safety, or sustained value.
- AI layered on top of broken processes: Inserting AI into a fundamentally flawed process simply accelerates the flaws. AI must be part of a redesigned workflow, not a patch.
- Leadership expectations misaligned with readiness: Large organizations move cautiously but expect scale; smaller organizations move fast but risk making impulsive decisions. Both are challenged if foundational readiness is lacking.
Join us at All Access: AI in Business Transformation 2026!
The new process excellence: Enabling AI to scale
This is where the conversation shifts. AI does not replace process excellence. AI elevates the importance of process excellence.
An organization cannot scale AI without stable processes, trusted data, clear governance, and measured outcomes. Process excellence becomes the scaffolding on which AI can operate safely and reliably.
Here are the five pillars of the new process excellence that enable AI to succeed:
- Stabilize before you automate: AI cannot learn accurately from unstable workflows. Lean and Six Sigma principles – standardization, reduction of variation, control plans, and capability baselines – are prerequisites for any meaningful AI adoption.
- Redesign work using an AI-first mindset: Instead of asking, “Where can we add AI?” we must ask: “How would this process operate if AI were native to it?” This redesign must define which tasks AI performs, which decisions stay human, what triggers escalation, what accuracy thresholds are acceptable, how exceptions are routed, and how outputs are validated. Transformation requires intentional design, not patchwork integration.
- Build practical, responsible AI governance: Governance must be business-led, not technology-imposed. A workable model includes clear RACI across business, data, technology, and risk, accuracy, transparency, and explainability thresholds, continuous model monitoring, risk-based use-case classification, and human-in-the-loop checkpoints for high-stakes decisions. Governance is what turns AI into a safe, scalable asset.
- Strengthen data foundations: Data is the fuel, and most organizations are still cleaning the tank. Process excellence must partner with data functions to improve data quality and completeness, lineage and traceability, integration layers, metadata standards, master data governance, and process mining insights. Better data means better decisions, human or AI.
- Build a scalable value engine: AI success should follow a disciplined pipeline: identify the right use cases, assess risk and ROI, fix the underlying process, pilot with real data, measure throughput + accuracy + rework, scale intentionally, and sustain through monitoring. This rhythm produces repeatable, compounding value.
Large versus small organizations: Different contexts, same lessons
Having worked in both a global enterprise and a smaller agile financial institution, I’ve seen the contrasting realities of AI adoption.
Large organizations
Strengths: Robust governance, structured data, strong risk frameworks.
Challenges: Complex legacy systems, slow decision-making, high coordination needs.
Small organizations
Strengths: Agility, speed, faster experimentation.
Challenges: Immature governance, thinner data disciplines, risk of abrupt decisions.
Despite the differences, the success formula is identical: AI works only when the process, data, and governance are ready for it.
The evolving role of process excellence leaders
AI is not diminishing the relevance of process excellence – it is expanding it. Leaders are becoming:
- Orchestrators of human–AI collaboration
- Designers of AI-enabled workflows
- Guardians of high-quality data flows
- Responsible AI advocates
- Translators between business and technology
- Value realization leaders
This is a pivotal moment for the profession.
A 90-day roadmap: From experimentation to impact
Here is a practical approach for organizations seeking real value from AI:
Days 1 – 30: Prepare
- Pick high-value, low-risk use cases
- Stabilize the underlying process
- Resolve immediate data issues
- Map human–AI decision boundaries
Days 31 – 60: Pilot
- Introduce AI with defined guardrails
- Test accuracy and exceptions
- Measure throughput and effort saved or added
- Gather lessons
Days 61 – 90: Scale
- Implement model and process monitoring
- Update SOPs
- Train teams
- Expand to similar processes
This approach balances ambition with responsibility.
Turning AI’s promise into performance
AI is not a shortcut to transformation. It is a powerful capability that requires mature processes, clear governance, strong data, and thoughtful adoption. The organizations that succeed with AI will not be the ones that experiment the fastest, they will be the ones that redesign the foundations that allow AI to perform consistently and responsibly at scale.
AI is not replacing process excellence. AI is the next frontier of process excellence. The leaders who understand this will shape the future of how organizations think, work, and transform.