Procurement and demand: from reactive ordering to planning

More pressure, same resources

A recent McKinsey analysis of procurement, reviewed by Industry Today, notes that 55% of procurement leaders reported flat or shrinking budgets, while all of them saw their savings targets rise. The same analysis estimates that AI tools applied to procurement tasks can raise the function’s productivity by between 25% and 40%, although AI adoption in procurement is roughly six times lower than in marketing and sales (Industry Today).

In the large global company, the conversation is about copilots and agents. In the mid-sized Latin American company, the gap usually comes earlier: procurement still lacks the basic information needed to decide. Before automating, you have to put things in order.

The reactive cycle

In our work with a food services business group, the supply chain diagnosis revealed a common pattern. The monthly raw material estimate was done manually during the first days of each month. Deviations between what was estimated and what was consumed were corrected with urgent purchases, which were more expensive, and with surpluses that ended up obsolete. Administrative processes (planning, procurement, customer service) and operational ones (supply, inventory, distribution) operated in isolation.

The result was a reactive cycle: inaccurate order, urgent purchase, poorly rotated inventory, shortages at the point of service and, once again, an inaccurate order. Nobody was doing their job badly; the system was designed to react.

Three levers, in order

The first lever is demand management. A forecast based on consumption history, special events and seasonality reduces the variability that feeds everything else. Getting started doesn’t require complex models: it requires clean data and a person responsible for reviewing it.

The second lever is inventory. In the diagnosis we found no policies differentiated by product type, no systematic coding, which prevented effective FIFO or FEFO rotation, and control kept in spreadsheets without barcode readers. The prioritized response was a redesign by product type, a standard coding system and automated recording in the ERP. These are basic operating decisions, but without them the buyer works blind.

The third lever is logistics. Unoptimized routes and the absence of a make-or-buy criterion for in-house versus outsourced transport had a direct impact on the margin. The lever here is not negotiating the rate with the carrier, but redesigning the model.

S&OP as the backbone

The piece that connects the three levers is a sales and operations planning (S&OP) process. Its function is simple: for sales, operations, procurement and finance to agree once a month on a single demand and supply plan, and for deviations to be discussed around a table rather than in urgent emails.

In our experience, the main barrier to S&OP is not technical, but one of governance. Each area has its own indicators and incentives. Procurement is measured on price, operations on fulfillment and sales on revenue. S&OP forces the company to measure what matters to it: service level, total cost and working capital.

End-to-end visibility

A cross-cutting lesson was the need for visibility. Without consolidated information on demand, inventory, production and delivery, the chain is managed by putting out fires. An integrated executive dashboard, even a simple one, changes the conversation: problems are seen before they reach the client.

Visibility also reduces risks that are not financial. In food supply chains, poor traceability exposes companies to food safety risks and reputational damage that no procurement savings can offset.

When AI makes sense in procurement

With the foundations in order, AI stops being a promise and becomes a tool. There are three use cases with the best effort-to-value ratio in the mid-sized company. The first is reconciliation between invoices, purchase orders and contracts, where value leaks hide in unagreed prices, duplicated quantities or conditions not applied. The second is spend analysis: automatically classifying what is bought, from whom and how often, to identify supplier consolidation and off-contract purchasing. The third is support for demand forecasting, combining historical data with external variables.

The condition for all three is the same: consistent supplier and product master data, and an ERP that records every transaction. In our work, end-to-end configuration of the ERP’s procurement module and multi-warehouse management emerged as enablers of almost every supply initiative. Skipping that step to go straight to AI produces eye-catching pilots that don’t scale.

The procurement organization changes too

A procurement function that moves from transactional to strategic needs different profiles. The McKinsey analysis cited above notes that chief procurement officers consider strategic thinking the most critical skill for category managers in the future. In the mid-sized company, the first step is usually more modest: separating the role of the person who places orders from the role of the person who manages categories and suppliers, and giving the latter total-cost targets, not price targets.

Where procurement sits in the structure also matters. When it reports to operations, it tends to prioritize availability; when it reports to finance, price. Neither is wrong, but S&OP must balance them.