Putting AI to Work: The Operational Excellence Imperative

Businesses report feeling squeezed and reach for a combination of the evergreen (supplier changes, cost cutting, restructuring) and the new (digital solutions and AI). McKinsey questions whether either works. The track record of cost-focused transformation is poor, and AI remains overwhelmingly stuck at the experimentation stage. The survey of 1,000 senior and midlevel executives worldwide is framed around finding what separates the companies where both actually pay.

• Nearly 90% of organizations are at least experimenting with AI; 7% report scaling it across the enterprise.
• At most a quarter of organizations are experimenting with AI agents in any function.
• Mastering both AI and operational excellence is difficult, and adoption correlates closely with revenue, so scale and resources still shape who can do it.

The headline result, shown in Exhibit 1. Companies with AI embedded across multiple functions generate nearly double the profit margins of peers using it in only a few departments, and more than five times the three-year ROIC. Companies with high operational-excellence maturity show the same shape independently: comparable margins and more than triple the three-year ROIC. McKinsey is explicit that these are correlations, but argues the consistency across deployment, operations, productivity, and financial performance points to stronger operating systems being what lets AI investment convert into results.

• Operational excellence becomes visible through everyday practice: clear KPIs, reallocating resources on real-time data, continuously incorporating customer feedback.
• The two forces form a reinforcing loop rather than two separate programs.
• Companies scoring higher on operational excellence are further along in deploying AI, echoing earlier research that high performers innovate faster.

Exhibit 2 traces the mechanism. Productivity gains rise steadily with operational-excellence maturity, with high-maturity companies clustered in the highest improvement bands and low-maturity peers lagging well behind. Companies that embed real-time data, digital workflows, and advanced analytics into their operating models achieve the largest improvements, and those gains free up capacity and capital for faster AI deployment.

• Higher throughput, lower variability, and more efficient asset use are what protect margin under disruption.
• One of the world's largest lithium producers built on more than a decade of operational-excellence investment to run AI on drone and sensor data, reducing water use while raising output and quality, and is now designing agents for complex maintenance work.

Exhibit 3 shows a clear relationship between the breadth of AI deployment and productivity improvement. Impact comes from integration into core operational processes, not from experimentation. The global automaker case makes the sequencing explicit: the company mapped processes end to end, found extensive technical debt and manual work, and simplified and redesigned workflows before deploying AI capabilities.

• Software development productivity rose by up to 44%.
• AI-supported sales tools helped build tailored customer profiles and contributed to roughly 40% more incoming orders.
• The organization identified 20% to 25% long-term efficiency potential as AI adoption spread across the business.

Exhibit 4 isolates the practices that higher-margin companies consistently share: clear KPIs cascading from leadership to the front line, enterprise-wide performance measures tied to purpose, and disciplined use of real-time data to prioritize resources and decisions. The definition has shifted from standardized processes and periodic improvement programs to a living system that combines execution discipline, digital enablement, and continual learning.

• Performance management becomes faster and more transparent, decision rights get clearer, and data flows let teams adapt in real time.
• Companies treating operational excellence as an enterprise-wide capability absorb disruption and integrate new technology without fragmenting operations.
• Service-oriented sectors such as finance, technology, and pharmaceuticals score highest; asset-intensive, variable, or fragmented sectors lag.
• Midsize organizations of 500 to 5,000 employees are squeezed in the middle on maturity and scale.

The cases are the strongest part of the argument because each one shows a site deliberately declining to scale technology until the operating foundation could support it. In every case the constraint addressed first was organizational, not technical.

• Siemens Nanjing: resisted scaling digital twins until a manufacturing operations management system governed data flows, plan-do-check-act routines validated simulations, and decision rights defined when human confirmation was required. Units per hour rose 50%, delivery lead time fell 83%, cycle time fell 12%.
• Qatar Shell Ras Laffan: concentrated on one high-value intervention, asset reliability through real-time monitoring and predictive analytics on physics-based models. Engineering costs fell 90%, response time to system upsets fell 98%, capex fell 64%, and targeted maintenance extended critical equipment life by six years.
• Midea Si Racha: rebuilt the entire complaint-resolution process around AI, with engineers reviewing and confirming each step. Time from quality issue to action plan went from beyond 60 days to one day, defect rates fell 43%, complaint rates fell 32%.
• CITIC Dicastal Morocco: machine learning on casting and machining parameters plus AI vision inspection, with engineers reviewing outputs. Defect rates fell more than 30%, OEE rose 17%, labor productivity rose 27%, direct energy-related emissions fell more than 50%.

Exhibit 5 isolates the capabilities that separate performers among companies already using AI at scale: well-defined processes for defining AI business cases, deployment across multiple functions, sufficient access to people, data, and systems to build high-value applications, and an agreed adoption roadmap. Leading companies link use cases directly to operational and financial outcomes and invest in the data foundations, redesigned processes, and hybrid technical-operational talent required to scale.

• Where these elements come together, AI becomes a repeatable performance lever; where they do not, gains stay uneven.
• The pragmatic implication for laggards is to focus first on clarity, integration, and capability building, so that deployed AI contributes to performance rather than staying confined to pilots.

Five moves, illustrated through Schneider Electric's Evreux distribution center, which redesigned its operating model so returned products could be refurbished and reenter the supply chain at scale. The closing argument: use operational excellence as the stabilizer and technology as the accelerant, and the gains compound.

• Start with a small number of operational priorities such as throughput, yield, service levels, or asset utilization, instead of spreading investment across dozens of pilots.
• Design for scale from the first use cases, choosing initiatives that cut across functions and align data, ownership, and decision rights. One energy company ran an AI negotiation platform against more than 2,000 tail-spend suppliers simultaneously, delivering about 2.5% savings on the addressed spend in two months.
• Build the operating backbone alongside the technology, treating data foundations, performance management, and governance as core disciplines.
• Embed capability where decisions are made, placing technical expertise alongside operators and business leaders.
• Sequence ambition to absorption capacity, avoiding both waiting for perfect conditions and scaling faster than the organization can absorb.
• Within two years, circular products reached 38% of the Evreux product range while Scope 3 emissions fell 43%.