Rewiring Operations: Embedding AI and Decision-Making into Core Processes

Most AI implementations are layered on top of existing operating models. These models were designed around functional execution rather than shared decision-making.

As a result, analytical capability and operational execution evolve separately. AI systems generate forecasts, scenarios, and recommendations, while actual decisions continue to be made in spreadsheets, local meetings, or within individual functions.

This creates structural fragmentation:
‍

• data is updated in analytical systems, but interpreted differently across functions;
• procurement, production, and logistics operate on different versions of the same reality;
• decisions move through a sequence of approvals rather than a single coordinated flow.

The difference can be illustrated across key operational dimensions.

This gap explains why improvements in forecasting, optimisation or analytics often remain local. In supply chain contexts, this becomes visible when demand signals, inventory decisions and supplier actions are not aligned in time, even when underlying data is accurate.

These patterns become visible in day-to-day supply chain execution.

Demand forecasts are updated in planning systems, while procurement continues working with earlier assumptions. Production adjusts schedules based on internal capacity constraints. Logistics adapts execution after decisions have already been made.

Each function operates rationally within its own context. However, the overall system reflects accumulated misalignment: inventory builds up in some nodes while shortages appear in others, and continuous adjustments replace coordinated planning.

‍Boston Consulting Group describes similar dynamics in end-to-end transformation work, where improvements at the functional level fail to translate into system-wide performance without a unified operational design.

This misalignment directly affects core supply chain metrics. Stockouts lead to lost sales and reduced service levels, while excess inventory ties up working capital and increases holding costs. Supplier coordination becomes less predictable, extending lead times and forcing reactive adjustments across production and distribution. As a result, operational performance becomes less stable, even when underlying data and forecasts are accurate.

A different approach can be seen in companies that connect AI directly to operational decisions.

A relevant example comes from Toshiba Tec, working with McKinsey & Company and NVIDIA.

The company originally operated as a hardware provider for retail point-of-sale systems, generating large volumes of transactional data. This data was analysed retrospectively, limiting its impact on operational decisions.

The transformation addressed both model performance and the way decisions were made using that data.

With GPU-accelerated infrastructure and transformer-based models, the system shifted toward real-time processing of transaction data. Customer behaviour signals began to feed directly into pricing, promotion, and product recommendations.

Retail decisions started to reflect live data rather than historical reports.

The outcome included a sevenfold improvement in personalization performance and a significant reduction in processing time. More importantly, analytics became part of the operational cycle rather than a separate analytical function.

When AI is embedded into operations, decision-making structures begin to shift.Instead of sequential handovers between functions, operational processes begin to operate as continuous decision flows where data, constraints, and execution are linked within the same system.This shift is visible in organisations scaling AI across service and operational workflows.

A relevant example is PwC’s work with Wyndham Hotels & Resorts.

AI agents were integrated into customer and franchise operations:

• routine service requests are handled in real time;
• operational queries are resolved through automated systems;
• human teams focus on exception handling and higher-value decisions.

‍
The key change lies in how work is distributed within a single operational system rather than across separate functions.

Most organisations today already use AI across multiple functions. The main constraint lies in how these systems are designed and coordinated across functions.

Companies that achieve scale tend to shift three elements at the same time:

• AI becomes part of operational workflows rather than a standalone analytics layer;
• decisions are made within integrated flows instead of moving between functions;
• responsibility is attached to outcomes rather than individual steps in the process.

‍
This combination determines whether AI remains a set of local improvements or becomes part of the operational system itself.

AI is increasingly present across operational environments, yet its impact depends on how it is integrated into decision-making structures. In many organisations, data, analytics and execution continue to operate as separate layers, which limits the effect of even advanced models.

Organisations that achieve consistent results connect planning and execution within a single operational logic. In supply chain terms, this links demand planning, procurement, production and distribution into one continuous flow of decisions.

Planning signals are directly connected to replenishment and production decisions, while inventory and capacity constraints are reflected immediately in execution. This reduces delays and improves coordination across the supply chain.