Author: James Synder is a Certified Master Anaplanner and Associate Director at Accenture.
Bridging the gap
Where we left off
In Part I of this series, we explored three gaps that emerge when inventory constraints are not fully represented in the Anaplan Inventory Planning application: (1) the customer-facing shortage can be hidden, (2) external capacity is not automatically a planning constraint, and (3) demand and inventory may live at different levels. This article examines strategies that have been deployed in practice to close each of those gaps.
Before diving in, two important caveats apply. First, one of the approaches described below is not feasible in a Polaris workspace and requires a Classic workspace; this distinction is explained where relevant. Second, for scenarios where weekly forecasting and external processes are operationally practical, the out-of-the-box application can partially address some of these gaps without custom development. That option is covered as well.
We will move from the most straightforward approach to the most complex.
1️⃣ Allocating inventory to customer-facing locations
Recall the first gap: a customer-facing distribution center assumes it can receive inventory from an upstream DC, but if that upstream DC is constrained, the shortage appears there rather than at the customer-facing location. The customer-facing DC continues to plan against expected receipts that may not materialize.
To resolve this, allocation logic can be applied across the multi-tier structure, drawing on the same framework that IVP200 uses to roll up inventory need across tiers out of the box. Using the native Tier 1 version as the starting point, a custom allocation layer is built to distribute available inventory based on documented need.
The recommended allocation priority order is:
- Planned transfers: deliberate, system-configured movements that should be honored first.
- Simulated network replenishment: the rolled-up need from customer-facing DCs.
- Customer demand: direct demand to be satisfied from the DC's remaining available inventory.
For each allocation tier, the logic determines whether a shortage exists and distributes what is available proportionally. Because this is a planning projection, a simple pro-rata reduction is recommended for clarity and consistency. For example: if 1,000 units are needed across three planned transfers and only 800 units are available, each transfer receives 80% of its request. Which transfers are fulfilled completely versus partially is an execution decision that will be resolved when the actual movement occurs.
The key technical mechanism is the PREVIOUSVERSION() formula. As documented in Anaplan's formula reference, PREVIOUSVERSION() returns the value of an expression from the version that precedes the current one in the model's version ordering. In the context of the supply chain application, versions are used structurally, not as planning scenarios, to sequence calculation passes without creating circular references.
The practical application works as follows: allocation outputs calculated in the "Tier 1" version are staged there. The "Tier 1 & 2" version then uses PREVIOUSVERSION() to pull those values and incorporate the allocated inventory as receipts at the receiving DC, mirroring how IVP200 handles the network replenishment roll-up natively. A DC receiving an allocated transfer in the Tier 1 & 2 pass treats that quantity as available inventory when calculating its own downstream obligations.
Mastering this pattern, stage in Tier 1 and retrieve via PREVIOUSVERSION() in Tier 1 & 2, is foundational. It underpins the allocation logic across all three gap scenarios discussed in this article.
2️⃣ Applying constraints with the out-of-the-box configuration
Before building custom allocation logic, it is worth considering whether the standard application configuration can address the second gap: unconstrained capacity from a central purchasing authority or contract manufacturer.
The Inventory Planning application calculates suggested receipts based on inventory policy and lead time parameters, treating future supply as unconstrained. However, the application can be configured in a scenario where suggested receipts are suppressed, limiting the inventory flow calculation to what is actually on order in the transactional system. When coupled with a weekly forecasting and ordering cadence, where supply transactions are refreshed weekly, this approach gives planners a near-real-time view of constrained inventory position in the short term.
The trade-off is visibility beyond the ordering horizon. Once the horizon is exceeded, shortages will accumulate because the model is no longer projecting future receipts. This approach works well for near-term constraint visibility, but planners should be aware that it does not provide a full long-range supply picture. Applied appropriately, this configuration is a practical mitigation for Gap 2, particularly for organizations with the operational discipline to maintain weekly order placement.
3️⃣ Allocating constrained supply from a central authority or contract manufacturer
When the out-of-the-box configuration is not sufficient, for example when a central purchasing team or contract manufacturer is allocating finite supply across multiple international regions, a more structured allocation model is required.
In one client engagement, multiple regional entities were competing for supply from a limited set of contract manufacturers. The standard Rough-Cut Capacity Planning (RCCP) and Production Planning applications were not a good operational fit given how the business managed its central buying process. Instead, demand was aggregated from region-level SKUs up to a global SKU level to assess total need against manufacturer capacity. The central authority then reviewed the aggregate need and returned an allocation of available supply back to each region.
Because the central review process operated on a monthly cadence, the model needed to project where inventory was likely to land in each region after the allocation was applied. The allocation output was then fed back into the regional inventory model using the same tier-based approach described in the first section. The returned allocation was treated as a constrained receipt and processed through the Tier 1 / Tier 1 & 2 version structure.
4️⃣ Allocating demand to the level of inventory: A note on Polaris vs. Classic
The most complex gap involves a mismatch between the level at which demand is planned and the level at which inventory is held. In the engagement described here, sales were planned at an aggregate main SKU level, while inventory existed at a more granular variant SKU level. This created two distinct challenges: determining which variant SKU to replenish, and simulating the depletion of variant-level inventory in age order.
The complexity was compounded by the need to account for inventory transfers and routine network replenishment simultaneously. The solution was approached in order:
- Calculate simulated receipts, routine network replenishment, and planned transfers at the main SKU level.
- Establish rules for which variant SKU would be replenished (in full or split across variants).
However, allocating across several variant SKUs proved to be a barrier in a Polaris workspace. While Polaris uses the same formula syntax and modeling interface as the Classic Engine, certain calculation patterns behave differently. Specifically, Polaris has limitations with non-winding iterative calculations; the sequential, pass-by-pass depletion logic required for variant-level allocation could not be resolved within the Polaris engine despite multiple implementation attempts.
The decision was made to stand up a dedicated allocation model in a Classic workspace. Data and lists were synchronized to the Classic model, where concatenated lists and allocation rules were built to manage inventory across transfers, network replenishment, and customer demand. The inventory aging and depletion order logic for variant SKUs was also housed in the Classic model.
The allocation logic in the Classic model relied on three sequential passes using the RANKCUMULATE function to deplete inventory in priority order. Each pass followed the allocation sequence used throughout this series: (1) Planned Network Transfers, (2) Routine Network Replenishment, and (3) Customer Demand.
The Tier 1 / Tier 1 & 2 version structure was applied here as well. Tier 1 handled the allocation of variant SKUs for inventory transfers, both planned and routine network replenishment. The Tier 1 & 2 version then received those values and treated the allocated transfer quantity as available inventory for downstream obligation calculations. Only customer demand at the DC level would surface any shortage; the allocation logic ensured that upstream transfers were withheld when inventory was unavailable, and the shortage was recorded at the DC supporting the end customer.
In closing
The four approaches described in this article, multi-tier allocation at customer-facing DCs, out-of-the-box constrained receipt configuration, central authority allocation, and variant-level depletion modeling, demonstrate that the gaps identified in Part I are addressable through deliberate design. None of these solutions requires replacing the core inventory planning logic. Each extends it in a targeted way to represent the constraints that the standard application does not automatically capture.
Two principles carry through every approach. First, the Tier 1 / Tier 1 & 2 version structure, combined with the PREVIOUSVERSION() formula, is the foundational pattern for sequencing multi-pass calculations without circular references. Second, implementation decisions about engine selection (Polaris versus Classic) matter: some allocation patterns that are natural to implement in Classic require architectural workarounds or separate models in Polaris.
This is not an exhaustive list of strategies. Supply chain networks vary widely, and the right design will depend on the specific constraints, planning cadences, and data structures of each organization. But the patterns here offer a starting point and a demonstration that visible, actionable constraint planning is achievable within the Anaplan platform.
In supply chain planning, the most dangerous constraint is not the one you can see. It is the one your planning system does not know exists, until you build the model that makes it visible.