AI and the Retail Sector

The value pool is real, but it is not evenly captured

McKinsey has estimated that generative AI could create US$240 billion to US$390 billion in economic value for retailers, equivalent to a 1.2 to 1.9 percentage point margin improvement across the industry. For executives, the implication is significant. In a low margin sector, even small improvements in forecast accuracy, stock availability, pricing, fulfilment and labour productivity can materially change performance.

The strongest opportunities are not confined to customer facing experiences. They sit across the retail operating model.

Demand forecasting can reduce forecast error, improve availability, reduce waste and lower inventory holdings. Pricing and promotions can improve gross margin, reduce ineffective discounting and sharpen commercial decisions. Inventory optimisation can release working capital and reduce markdowns. Fulfilment and route optimisation can lower cost to serve. Store operations and workforce planning can improve service performance and productivity.

Personalisation still matters, but it is too narrow as the dominant retail AI story. A better product recommendation is useful. A better product recommendation that also understands stock availability, margin, delivery cost, customer value and promotion strategy is far more valuable.

Retail AI value compounds when demand, pricing, promotions, inventory, fulfilment, cost to serve and margin decisions are connected.

The scaling gap is an architecture problem

The most useful retail statistic is also the most confronting: 90 percent of Fortune 500 retailers are experimenting with AI, but only around 4 percent have scaled it enterprise wide. The difference is architecture.

This aligns with what we see across Australian enterprise environments. Retailers often have capable teams and sensible use cases, but the data landscape works against them. Commerce platforms, point of sale systems, loyalty platforms, ERP, warehouse management, marketing systems, finance platforms, supplier data and store operations systems all carry parts of the truth. Each has its own definitions, timing, ownership and quality issues.

For Australian retailers, the business consequence is practical. Forecasts are not trusted. Promotions are measured too late. Inventory decisions are made with incomplete signals. Pricing decisions do not always reflect fulfilment cost or stock position. Customer engagement is disconnected from availability. Digital growth can increase revenue while reducing margin if cost to serve is not visible.

The retailer that connects its data estate will outperform the retailer that deploys the best model on fragmented data.

Point solutions do not change the operating model

Many retail AI programs start with sensible point solutions. A demand model. A recommendation engine. A service assistant. A pricing tool. A workforce planning prototype. Each may deliver a local improvement.

The issue is that retail performance is not local. A promotion that lifts demand but creates stockouts can damage customer experience. A pricing model that improves conversion but ignores margin can weaken profitability. A fulfilment model that reduces delivery time but increases cost to serve may not improve enterprise performance. A customer model that improves engagement but cannot activate consistently across channels will not scale.

What we see in production is that AI starts to matter when it is embedded into operating decisions, not when it sits beside them. That means the forecast affects replenishment. Pricing reflects stock, margin and demand. Promotions are planned with supply chain input. Customer offers reflect availability and fulfilment economics. Store labour planning reflects demand, channel mix and service expectations.

This is why the path to Return on AI in retail is not more point solutions. It is creating the operating model and data foundation that makes AI decisions usable in daily operations.

Domain transformation is the better unit of change

Retailers should move from isolated use cases to domain level transformation. That does not mean trying to rebuild the whole business at once. It means choosing domains where connected decisions can create measurable commercial value.

Demand and inventory is a natural starting point. AI can improve forecasting, allocation, replenishment, markdown decisions and supplier planning. The business outcome is not better forecast accuracy for its own sake. It is improved availability, lower inventory, less waste, fewer markdowns and better working capital.

Pricing and promotions is another high value domain. AI can help teams understand elasticity, competitor movement, promotional effectiveness and margin trade offs. But value depends on governance. Retailers need clear commercial rules around customer fairness, brand position, regulatory expectations, supplier funding, stock levels and margin thresholds.

Customer engagement is broader than personalisation. The higher value opportunity is connecting loyalty, lifecycle management, churn prevention, basket growth, service recovery and channel experience. That requires trusted customer data, consent management, offer governance and reliable execution across channels.

Supply chain and fulfilment is equally important. AI can improve route planning, stock positioning, warehouse productivity and delivery performance. The commercial impact is margin protection, better service levels and lower cost to serve.

The common thread is simple: AI creates more value when it improves connected decisions, not isolated tasks.

The operating model determines whether value scales

Retail AI cannot be owned by an innovation team alone. The highest value opportunities cut across merchandising, marketing, supply chain, digital, store operations, finance, technology and data. Without a clear operating model, every use case becomes a cross functional negotiation.

A practical model balances central governance with domain ownership. The central AI and data capability should set architecture, security, governance, reusable delivery patterns, data standards and platform controls. Business domains should own prioritisation, process change, adoption and benefits realisation.

This matters because value is often lost after deployment. A model may improve forecast accuracy, but planners still need to trust it and change their decisions. A pricing tool may identify margin opportunities, but commercial teams need clear rules for when to act. A service assistant may reduce handling time, but workforce plans need to convert that time into capacity or service improvement.

AI value is realised only when the business changes how work gets done.

The executive provocation

Retailers do not need more disconnected AI activity. They need a clearer view of where AI will improve the economics of the business.

The executive agenda should be direct. Choose the domains where better decisions will improve margin, revenue, working capital, cost to serve, service performance or customer experience. Connect the data estate around those domains. Build governance into the platform layer. Redesign workflows so AI outputs become operational decisions. Measure realised value, not activity.

The gap between experimentation and enterprise scale is not a technology story. It is a leadership, architecture and operating model story.

For Australian retailers, the question is simple: are you investing in more point solutions, or in the foundation that makes every AI investment more valuable?

Read more of Adrian’s ‘Return on AI’ series here.