The 3D bin packing problem behind mixed pallets, the algorithm families that solve it, and the constraints that decide whether a pallet arrives intact.
Mixed-SKU pallet building means stacking cartons of different sizes, weights and handling rules onto one stable pallet. Palletization software treats the pallet as a three-dimensional bin packing problem: it searches for positions and orientations that keep every carton inside the pallet’s footprint and height limit, respect weight and stacking rules, and use as much space as possible. The problem is strongly NP-hard, so real systems rely on heuristics instead of checking every option.
| NP-hard STRONGLY: THE 3D BIN PACKING PROBLEM BEHIND MIXED PALLETS (MARTELLO, PISINGER & VIGO, 2000) | 2.4 × 10¹⁸ WAYS TO ORDER JUST 20 CARTONS, BEFORE COUNTING ROTATIONS AND POSITIONS | 25% AVERAGE COMPRESSION STRENGTH LOST BY EMPTY BOXES STACKED THREE HIGH ON A PALLET IN LAB TESTS (CAL POLY / MSU) |
Contents
- 1 Single-SKU vs. Mixed-SKU Palletization: Why Complexity Explodes
- 2 The Mathematical Foundation: The 3D Bin Packing Problem
- 3 Algorithmic Approaches to Mixed-SKU Pallet Building
- 4 Standard Uniform Pallet Stacking Patterns
- 5 Constraint Engine Framework for Mixed-SKU Software
- 6 Financial and Operational ROI of Algorithmic Palletization
- 7 Frequently Asked Questions (FAQ)
Single-SKU vs. Mixed-SKU Palletization: Why Complexity Explodes
A single-SKU pallet is a solved problem. When every carton has the same dimensions and weight, the pallet follows a fixed pattern, such as block, brick or pinwheel, that can be taken from a pattern chart and repeated layer after layer.
Mixed-SKU pallets are a different job. Distribution centers, retail replenishment and 3PL operations build pallets from dozens of cartons that differ in size, weight, permitted orientation and fragility. There is no chart for that, and every order is a new puzzle.
| Why brute force fails Take a standard footprint such as the 1200 × 800 mm EUR pallet or the 48 × 40 in GMA pallet common in North America. The order in which 20 different cartons are placed can alone be chosen in 20! ≈ 2.4 × 10¹⁸ ways, and each carton can also be rotated and moved to many positions. Checking every combination is out of the question for a system that has to answer while a packer waits. |
The Mathematical Foundation: The 3D Bin Packing Problem
Mixed-SKU pallet loading is a practical variant of the three-dimensional bin packing problem (3D-BPP): rectangular items have to be packed orthogonally into the smallest number of rectangular bins, without overlapping and without crossing the bin’s walls. Martello, Pisinger and Vigo describe it as strongly NP-hard and “extremely difficult to solve in practice” (Operations Research, 2000). In plain terms, no known method guarantees the best layout quickly for large orders, which is why commercial software uses heuristics that find very good layouts in a fraction of the time.
Physical and Operational Constraints in Palletization
Textbook 3D-BPP only asks whether the boxes fit. A real pallet also has to survive a forklift, a truck and a warehouse rack. A widely cited review of container loading makes the same point: many proposed approaches have been criticized as having limited practical value because they ignore constraints met in practice (Bortfeldt & Wäscher, 2013). The ones that matter most on a pallet:
- Base support: every carton needs enough support underneath, from the pallet deck or the tops of the cartons below, so it cannot tip or sag into a gap.
- Load bearing and stackability: heavy, strong cartons go low; light or crushable ones go on top.
- Center of gravity: the combined center of gravity should stay low and close to the middle of the footprint, so the pallet does not tip during forklift handling or hard braking.
- Orientation (“this side up”): liquids, glass and some electronics may only be rotated around the vertical axis.
- No overhang: cartons should not stick out past the pallet edge. Overhang snags on conveyors and racks, and it weakens the boxes themselves: in lab tests by packaging researchers at Cal Poly and Michigan State University, a 25.4 mm (1 in) overhang cut box compression strength by up to 19% (study).
Algorithmic Approaches to Mixed-SKU Pallet Building
Pallet-building software usually combines several of the families below. Speed and fill rate depend heavily on the item mix and the rules switched on, so read the table as a map of trade-offs rather than a benchmark.
| Algorithm family | How it works | Strengths | Typical use |
| Greedy heuristics (e.g. first fit decreasing, best fit decreasing) | Sorts cartons, usually by volume or weight, and places each one in the first or best feasible position | Very fast and predictable; easy to add rules | Real-time requests from a WMS at the packing station |
| Layer building | Groups cartons of similar height into flat layers and stacks the layers | Flat, stable layers that are easy to wrap and to build by robot | Robotic palletizing; orders with many cartons of similar height |
| Guillotine and space-splitting methods | Recursively divides the free space into smaller rectangular blocks with edge-to-edge cuts | Simple geometry that is easy to follow by hand | Cartonization and standard box packing |
| Metaheuristics (genetic algorithms, simulated annealing, tabu search) | Repeatedly modifies complete layouts and keeps the better ones | Higher fill on complex mixes | Batch planning before dispatch, when there is time to compute |
| Deep reinforcement learning | A neural network learns placement decisions from many simulated packing episodes | Can handle online packing, where each box is placed as it arrives | Research and robotic packing cells |
Deep reinforcement learning is the newest of these families. Zhao and colleagues trained an agent for the online version of the problem, where each box must be placed immediately on arrival and the placement has to stay physically stable, and reported that it outperformed the state-of-the-art methods they compared against (AAAI 2021). For a warehouse, though, the practical question is less which family is newest and more whether the plan respects your rules and can be built by a person or a robot.
In 3DBinPacking, this is what Stack Pallets is for: you enter the pallets you have, the quantities and dimensions of the items and, optionally, a maximum height and weight. It returns the dimensions of each packed pallet, the list of items on it and step-by-step visual instructions for the packer.
Standard Uniform Pallet Stacking Patterns
Mixed pallets often contain blocks of identical cartons, and for those, software falls back on uniform patterns. Our guide to palletizing patterns covers them in detail; here is what matters for pallet-building algorithms.
Column Stacking (Block Pattern)
Cartons sit directly on top of each other, corner over corner, in every layer. The load runs down through the strong corners of each box, so column stacks carry the most weight. The weakness is that nothing ties the columns together, so the load depends on stretch wrap, tie sheets or corner boards to stay in one piece.
Brick / Interlocking Pattern
Each layer is rotated or offset against the one below, like bricks in a wall. The overlap ties the load together and resists shifting in transit, but it puts box corners on top of the weaker side panels below. In the same Cal Poly and Michigan State tests, interlocked stacks were weaker in compression than column stacks, by up to about a third for some boxes, depending on the box and the pallet.
Pinwheel / Windmill Pattern
Cartons are arranged around a central square, each turned 90° to its neighbor like the blades of a windmill. The pattern interlocks square or nearly square cartons that do not form a good brick pattern, at the cost of a void in the middle that software can fill with smaller items.
Constraint Engine Framework for Mixed-SKU Software
| Constraint | How the algorithm handles it | Why it matters |
| Weight per tier / load bearing | Adds up the weight resting on each carton and blocks placements that exceed what the cartons below can carry | Prevents crushed bottom cartons and leaning stacks |
| Center of gravity | Checks the combined center of gravity after each placement and rejects layouts that drift too far from the middle | Lower risk of tipping during forklift handling and transit |
| Orientation locks | Allows rotation only around the vertical axis for flagged items | No leaking liquids, broken glass or upside-down products |
| Delivery sequence | Places cartons for the earliest stops where they can be reached first | Multi-drop deliveries without rebuilding the pallet at each stop |
| Item separation | Keeps incompatible goods apart, for example food and household chemicals | Avoids contamination and damage claims |
In the 3DBinPacking API, several of these rules are explicit parameters: item weight and a maximum weight for each pallet or box, whether an item may be rotated vertically, groups that keep items together and apart from other groups (the documentation’s example separates food from detergents), and items that must not share a space with anything else. With coordinates switched on, the response gives the x, y and z position of every packed item, along with step-by-step images for the packer.
Financial and Operational ROI of Algorithmic Palletization
The arithmetic is simple: for the same freight volume, the number of pallets falls in proportion to how well each pallet is filled.
| Pallets needed = current pallets × current average fill ÷ new average fill |
Example. Suppose a distributor ships 1,000 pallets a month at an average cube fill of 70%. If better building lifted the average to 80%, the same volume would travel on 875 pallets (1,000 × 70 ÷ 80): 125 fewer pallets to wrap, stage and pay for. Whether that reaches the freight invoice depends on how the carrier charges (per pallet or pallet space, by weight or by freight class), and a pallet often reaches its weight or height limit before it runs out of volume.
Time is the other lever. AW Latvia, a maker of wall panels, uses 3DBinPacking to calculate the number and height of pallets for orders on its B2B platform and passes the shipment size via API to FedEx and TNT to get a price. The team puts the reduction in order-processing work at roughly ten to one (AW Latvia success story). Fuller pallets are half of the job; the next step is how they load onto the truck, which we cover in our guide to truck load planning.
Frequently Asked Questions (FAQ)
What is mixed-SKU palletization?
Mixed-SKU palletization is building one stable pallet from cartons with different dimensions, weights and handling rules. Software decides where each carton goes and in which orientation, so the pallet stays within its limits and holds together in transit.
Why is mixed-SKU pallet building so hard to optimize?
Because it is a version of the three-dimensional bin packing problem, which is strongly NP-hard. The number of possible arrangements grows explosively with the number of cartons, so no known method guarantees the best layout quickly for large orders. Software uses heuristics that find good layouts fast.
What is the difference between column and interlocking pallet stacking?
Column stacking places cartons corner over corner, which gives the highest compression strength but little lateral tie. Interlocking rotates or offsets alternate layers to tie the load together, which improves stability but lowers compression strength; lab tests have measured losses of up to about a third.
How does pallet building software integrate with warehouse systems?
Usually through an API: the WMS or ERP sends item and pallet data and gets a packing plan back. The 3DBinPacking API returns JSON and, on request, the x, y and z coordinates of each item, which a WMS can turn into packing instructions.
| Try mixed-SKU pallet planning on your own orders Enter your pallet sizes and a real order in Stack Pallets and see the packed pallets with step-by-step instructions. Start a 14-day free trial, no credit card required. |