Choose a Demand-Aggregation Window for Volatile Categories

What is a demand-aggregation window?

It is the time a category team lets requirements accumulate before acting on them together: a weekly purchase-order run for consumables, a monthly sourcing wave for a regional category, or a quarterly tender for a commodity. The window decides how many orders you place, which volume tiers you reach and how long a site waits for what it asked for.

Two public procurement systems formalize the same idea. OMB's consolidation memo describes category management as buying as one organized entity rather than as thousands of separate buyers in common spending categories. World Bank guidance on aggregation ties the right approach to how much requirements have in common, including their timelines and the buyers' ability to work together or rely on a central team.

What does pooling demand save, and what does waiting cost?

Pooling saves on fixed ordering costs and can unlock discount tiers, as a joint-replenishment study sets out for multi-item ordering. Grouping demand over time can also make it steadier: a spare-parts simulation found that aggregation reduced demand variability and the number of periods with zero demand.

The same simulation shows the limit: Hušková and Dyntar report that once the benefit of aggregation is used up, holding and ordering costs climb quickly as average inventory grows. Aggregation has organizational costs too. The World Bank fact sheet lists dependence on fewer suppliers and delays while buyers seek consensus among themselves as risks of aggregated buying.

How long should a demand-aggregation window be?

No study reviewed here sets a window length for every category. In the simulation of 12,374 spare parts, 5, 10 and 15 day windows lowered total holding and ordering costs, to a low of about 6.8 million euros against 7.2 million euros unaggregated; 20 and 25 day windows stayed below that baseline, while 30 days or more produced a steep increase. That study grouped one distributor's sporadic spare-parts demand over time; it did not pool requests across sites or test a volatile-price category, and its authors call it a single dataset.

In this guide's view, the practical method is to replay the last year of each category's requisitions under a few candidate windows, count orders, average stock and late deliveries for each, and choose the shortest window that captures most of the ordering and tier savings.

Then bound it. The window plus the supplier's lead time should not exceed the longest delay a critical requester can tolerate, the window should not exceed the shelf life an item will have left on delivery, and the stock it builds should fit each site's storage in days of cover, using the lower of physical space and any permitted quantity for hazardous materials. Minimum order quantities, pack sizes and full-load freight set a floor under the window, and critical items are often better kept on reorder points than in a pooled window.

How can sites with different rhythms share one buying calendar?

Nest the windows. Under the power-of-two policies studied by Federgruen and Zheng, each item is reordered at constant intervals that are power-of-two multiples of a base planning period, so a weekly item, a fortnightly item and a four-weekly item still coincide on a regular wave. The same paper shows that the best such policy comes within 6% or 2% of a lower bound on the minimum cost, depending on whether the base period is fixed or variable. Its model assumes constant, known demand, so the bound is a design argument rather than a promise for volatile categories.

In this guide's view, a multi-site category can set one base period, such as a week, assign each site and item group to a multiple of it, and publish the resulting calendar so that sites know when their requests join a wave. Items with very different demand or shelf life belong in their own groups rather than on the slowest common cycle, and waves only need to coincide where items share a fixed cost, such as one supplier or one delivery. For the sourcing event itself, a structured RFQ workflow helps keep pooled requirements comparable across suppliers.

When do volume discounts stop paying for the extra stock?

Quantity discounts can tempt buyers to order more than a window needs: Ai and co-authors note that suppliers offer such discounts to stimulate demand, but that extra purchasing can lead to extra loss of products that deteriorate. In their model, the optimal replenishment policy and basic replenishment cycle were highly sensitive to changes in the system parameters, which is a reason to recheck the window when costs or demand move.

Buying ahead behaves the same way. In a working paper on one distributor's plywood buying, forward buys increased profits over not buying ahead, but the benefit shrank or disappeared as holding costs rose and profit margins fell.

In this guide's view, compare totals rather than unit discounts: under an all-units discount the lower price applies to the whole order, so count ordering and freight costs avoided as savings, and count holding, spoilage, financing, obsolescence and the risk that prices fall below the tier price as costs, with storage space as a hard limit. Volume can also be pooled in the agreement itself, with tiers on cumulative volume and deliveries released on a schedule, which can reach a tier without holding the stock.

How should price volatility change the window?

Using market-wide physical inventory data on 21 commodities from 1993-2011, Symeonidis and co-authors found price volatility to be a decreasing function of inventory for most commodities, with a stronger link for energy and agricultural goods than for metals; that describes markets, not a buyer's own stock. Volatility does not translate into a simple rule, however: in a model of sourcing under fluctuating procurement costs, the best forward-contract order quantity was generally not monotone in the current spot purchasing cost.

In this guide's view, volatile categories need shorter review intervals rather than automatically longer windows, and the contract's pricing terms, such as a firm price, index-linked resets or short quote validity, decide how much purchase timing matters at all. Keep the base cadence, recheck the inputs (price, demand, holding cost and supplier capacity) more often, and let the trigger rules below decide when to buy ahead or behind the calendar.

What should trigger an off-cycle or spot purchase?

Write thresholds before they are needed. Mahapatra, Kar and Levental find it cost-effective to keep an existing mix of contract and spot buying until the contract-to-spot price ratio reaches a threshold, then shift the mix, and their model widens that price band when more of the planning horizon remains. Their figures come from a hypothetical dataset, so the band must be recalculated with a buyer's own prices and switching costs. Threshold rules also appear in forward-buying research, which adds that a forward purchase should include safety stock based on holding costs and the gap between the current and expected price.

In this guide's view, each category plan should name three kinds of trigger: a price band that justifies buying ahead or delaying, a stock or service level below which a site may buy outside the wave, and a safety or regulatory need that overrides the calendar.

Price bands belong to the category owner for the whole pool, while a site's own off-cycle authority should come from the stock, service and safety triggers, which override any price-based delay. Set a maximum forward cover for each category, agree it with finance, separate off-cycle releases on an existing contract from true spot buys, and give each trigger an approver and a record, so that off-cycle buying stays visible rather than becoming a habit.

Who should own the window and approve exceptions?

Central ownership has limits. OMB's memo notes that requirements involving specialized mission knowledge, urgent response or continuous program involvement can be less suited to buying by a central organization. The memo says FAR 8.004 will be amended so that, when a commercial product or service meeting an agency's requirements is available on an existing government-wide contract, the agency must use it unless the agency head grants an exception.

The same memo calls for accountability metrics in its consolidation workstreams so that buying strategy can be adjusted in time. The World Bank suggests agreeing the process and documents up front and delegating some decisions to a central team to limit consensus delays.

In this guide's view, the category manager should own the window and its triggers, sites should own their service levels and storage limits, and exceptions should go to a named approver with a reason recorded. Because pooling concentrates volume, check the resulting supplier exposure with a tiered supplier risk assessment before each wave is awarded.

What should a demand-aggregation register record?

The register below turns these trade-offs into fields a category team can fill in and review. It is an authored planning aid, not a standard.

How do AI agents change demand-aggregation decisions?