Fixed vs Random (Chaotic) Storage: Space and Pick-Rate Math

The rack report says the building is full, yet a walk down any aisle shows gaps everywhere. Each SKU owns its own positions, sized for the biggest delivery it ever receives, and most of those positions are half empty most of the time. Meanwhile receiving is staging pallets on the floor because nothing is free for the next inbound truck.

Switching every pallet to random locations frees space, but done carelessly it trades the overflow problem for a search problem: drivers hunting for stock, pallets taken from the wrong place, and counts that never match. The right answer is usually a mix, and the mix can be calculated. With three inputs per SKU and a simple travel model you can size each policy, price the difference and decide zone by zone.

• Space math: why a dedicated position averages half full, the formula for shared storage across several locations, and a 400-SKU pallet example.
• Travel math: round-trip distance and moves per hour under random, class-based and fixed storage on one layout.
• Controls: license plates, location scans, lot and weight rules, ownership, and a zone-by-zone checklist.

Fixed slots are sized for each SKU's peak, random slots for the total

Fixed storage needs enough positions for every SKU at its own maximum at the same time; random storage needs enough for the highest total on hand. That single difference drives almost everything else. Under fixed, also called dedicated, storage each location belongs to one SKU and nothing else may go there, even when it is empty. Under random, also called shared or chaotic, storage a pallet goes to any suitable empty position and the warehouse management system remembers where.

The two totals differ because SKUs rarely peak together. A SKU receives a full order quantity and then sells down to its safety stock, so on average it holds half its order quantity plus safety stock. Fixed storage pays for the top of every sawtooth at once. Random storage pays for the sum of the sawtooths, which sits much lower because one SKU's peak coincides with another's trough.

• Fixed positions: the sum over all SKUs of order quantity plus safety stock, in pallets.
• Random positions: the average total on hand, plus an allowance for the days when several SKUs peak together, plus a margin of empty positions so putaway always finds a slot.
• Hybrid positions: fixed homes for the pick faces that people work from every hour, random positions for reserve pallets behind them.

Why a dedicated position averages half full: the sawtooth arithmetic

A location that holds one SKU fills on receipt and empties steadily as stock is picked, so with even demand it is 50 percent full on average. Bartholdi's Warehouse and Distribution Science, the open textbook from Georgia Tech, works through this and then shows what happens when the same stock is spread across several shared locations that are emptied one at a time and released for reuse.

With k equal locations and picking from one location until it is empty, average utilisation becomes 1 minus the sum of 1/j for j from 1 to k, divided by 2k. One location gives 50 percent, two give 62.5 percent, four give 74 percent and twelve give 87 percent. The gain flattens quickly, which is why sites that split each SKU across very many locations get little extra space for a lot of extra travel and counting.

Space and throughput formulas used in this article

Fixed positions:     P_fixed  = sum_i (Q_i + SS_i)
Random positions:    P_random = (sum_i (Q_i / 2 + SS_i) + z * sqrt(sum_i Q_i^2 / 12)) * (1 + f)
Shared-location use: U(k)     = 1 - (1 / (2k)) * sum_{j=1..k} (1 / j)
Move cycle time:     t        = d / v + h        moves per hour = 3600 / t

where
  Q_i  = order quantity of SKU i, in pallets (receipts staggered across SKUs)
  SS_i = safety stock of SKU i, in pallets
  z    = service factor for the aggregate peak (3.09 covers about 99.9% of days)
  f    = empty-position float so putaway always finds a slot (0.05 = 5%)
  k    = number of equal shared locations one SKU is spread across
  d    = round-trip travel in metres, v = travel speed in m/s
  h    = fixed handling seconds per move (fork entry, lift, scan, set down)

Dedicated pallet storage in a warehouse management system averages 50 percent position utilisation under steady demand, and Expression 2.1 in Bartholdi's Warehouse and Distribution Science shows shared storage across four equal locations raising that to about 74 percent, with gains flattening beyond eight locations at about 83 percent.

The variance term in the random formula assumes receipts are staggered, so that at any moment each SKU sits at a random point in its cycle. A sawtooth spread evenly between safety stock and order quantity plus safety stock has variance equal to Q squared over 12. Summing those variances and taking three standard deviations gives the aggregate peak a well-run site will see on its worst ordinary day.

A worked pallet example: 400 SKUs in fixed versus shared positions

Take a single-deep selective rack area holding 400 SKUs in three velocity classes, one pallet per position. The fast A class has large order quantities; the long tail of C SKUs receives two pallets at a time. The table applies the formulas above. All quantities are illustrative but typical of a regional grocery or building-products distribution centre.

The 669 positions are the headline, but the safety stock line is just as instructive. Safety stock never leaves under either policy, so it costs the same space in both. The saving comes almost entirely from the cycle stock: fixed storage reserves room for the full order quantity of every SKU, while random storage reserves room for half of it plus a peak allowance. SKUs with small order quantities, like the C class, save the least.

Turning positions into money depends on what a position costs a year in rent, rack depreciation, lighting and insurance. The second table uses illustrative annual costs per position for a US building and an Indian building, then subtracts the extra labour that random storage adds because every move needs a location confirmation that experienced staff skip in a fixed layout.

Peak overlap and empty float: the margin random storage still needs

Random storage is not sized to the average; it needs a margin for coincident peaks and for the empty positions putaway must be able to find. Sites that size to the 1,440-pallet average in the example will be full several days a month. The z-term and the float term in the formula are the cost of that protection, about 170 positions here, and they should be argued over explicitly rather than hidden in a round-number buffer.

• Synchronised receipts: if most SKUs arrive on the same weekly truck from one supplier, cycles are not staggered and the aggregate peak approaches the fixed total; the variance term then understates the need.
• Seasonal build: a pre-season stock build raises every SKU's safety stock together, so size to the peak month, not the annual average.
• Putaway float: with fewer than about 5 percent of positions empty, drivers start travelling far for a slot, and inbound pallets wait on the dock.
• Deep lanes: in drive-in or push-back lanes that hold one SKU per lane, each lane behaves like a single dedicated location, so the 50 percent sawtooth returns inside the lane.

The lane effect matters most in high-density storage. Our article on space utilisation versus cube utilisation covers how honeycombing in deep lanes is measured; for storage policy, the practical point is that random assignment helps lane density only when lanes are small enough to empty and be reassigned often.

Travel per move under random, class-based and fixed storage

Pure random storage costs travel, class-based random storage wins it back, and fixed storage loses some of its travel advantage because it needs a bigger block. The comparison below uses one simplified layout: 20 parallel aisles 3.5 metres apart, a dock in the middle of the front cross-aisle, single-command moves and rectangular travel. It is a model, not a simulation, but it captures the direction and size of the effect.

Under pure random, the 1,611 positions fill aisles 60 metres long and every position is equally likely, so the average round trip is 95 metres. Class-based random storage reserves the nearest 25 percent of positions for A pallets, the next 42 percent for B and the rest for C, while still assigning randomly within each zone. With A, B and C taking 60, 30 and 10 percent of moves, the average round trip falls to 67 metres.

Fixed storage puts the A SKUs closest to the dock, as class-based storage does, but those A homes must hold 600 positions instead of about 400, and the whole block is 42 percent longer. Its average round trip of 84 metres beats pure random but loses to class-based random by about 17 metres a move. Over 180,000 moves a year that gap is roughly 550 driver hours, on top of the 669 positions.

What a location scan and a license plate must prove

Random storage works only if every move proves which pallet went to or came from which location. In a fixed layout a driver can find stock from memory; in a random layout the system's record is the only map, so any unrecorded move makes a pallet invisible. The control is a license plate on every pallet and a two-scan confirmation on every move.

• Pallet license plate: a GS1 Serial Shipping Container Code under Application Identifier (00), 18 digits, or an internal license plate number printed at receiving.
• Location label: a barcode on each beam position with a check digit, so the driver cannot confirm a location by reading the aisle sign.
• Two-scan rule: scan the license plate and the location on every putaway and retrieval; the system rejects any pair that does not match the directed task.
• Substitution path: if the directed pallet is blocked, the driver requests a reallocation in the handheld rather than taking an identical-looking pallet from elsewhere.

Random pallet storage in a warehouse management system depends on a scanned license plate, typically a GS1 Serial Shipping Container Code under Application Identifier (00) with 18 digits, matched to a scanned location check digit on every move, so that 100 percent of putaways and retrievals leave a record a cycle count can verify.

The incident below shows the failure the two-scan rule prevents. It is the behaviour Bartholdi warns about: a driver sent to a distant location takes the same product from a nearer one. Timestamps and identifiers are illustrative.

1. 06:52: retrieval task PK-77810 directs the driver to location 14-B-03-2 for the pallet with license plate ending 41726, lot L2611A, best before 10 January 2027.
2. 06:55: the location is blocked by staged pallets, so the driver takes the same SKU from 14-C-01-1, lot L2612C, best before 2 March 2027, and confirms with a product barcode scan the site still allows.
3. 07:30: the system ships license plate 41726 on paper; the physical pallet is still in 14-B-03-2, and 14-C-01-1 shows a pallet that is no longer there.
4. 13:10: the next order is directed to 14-C-01-1, the driver finds it empty and posts a short-location exception; the customer order ships a day late.
5. 15:40: cycle count CC-3304 corrects both locations, but the morning shipment went out with the later lot, breaking first-expiry-first-out and leaving the lot record for that customer wrong.

The fix is configuration, not training: disable product-only confirmation for pallet moves, require the license plate scan, and give drivers a one-tap reallocation when a location is blocked. In fixed storage the same error is less damaging, because the wrong pallet still comes from the right SKU home, which is one reason fixed homes survive in fast pick areas.

Lot rotation, weight and stability limits that override randomness

A random policy is never fully random: product rules narrow the eligible locations before the system picks one. The more of those rules a site has, the smaller the pool each pallet can use and the closer random storage drifts towards fixed. Count the constraints before you count the savings.

• Expiry rotation: the system must retrieve by earliest expiry or receipt date, which random storage handles well because it tracks each pallet, unlike fixed lanes where new stock can bury old.
• Weight and height: a pallet heavier than a beam level's rated load or taller than its clear opening is ineligible there, so location capacity fields must be accurate.
• Compatibility: chemicals, food and fragrance-sensitive goods may be confined to their own zones, which become small fixed-purpose areas.
• Stable stacking: OSHA 29 CFR 1910.176(b) requires stored material not to create a hazard and tiered loads to be stacked, blocked or interlocked so they cannot slide or collapse, which limits floor-stack heights in random floor zones.

These rules belong in the putaway strategy, evaluated in order before the system chooses among the remaining empty positions. Our guide to directed putaway rules covers that sequencing; the point here is that every rule should be data on the location or item master, never a supervisor's memory.

Putaway eligibility for random pallet storage in a warehouse management system should screen weight, height and compatibility before choosing a location, because OSHA 29 CFR 1910.176(b) requires stored loads to be stable and secure; a beam level rated at 1,000 kilograms per position must reject a 1,100 kilogram pallet every time.

Who owns slot rules, location data and the counts that prove them

Storage policy fails quietly when nobody owns the location master. A wrong beam rating or a missing check digit label produces no alarm until a pallet goes somewhere it should not. Each record behind the policy needs a named owner and a sign-off, as the table sets out.

Rules the WMS must enforce and suggestions software can make

Location confirmation and eligibility checks are pass-or-fail tests that the system must apply on every move. Deciding which SKUs belong in which class, or how big each zone should be, involves forecasting and trade-offs, so software can propose and a person decides. The table lists the tasks for a random or hybrid storage area.

Choosing a storage policy zone by zone: an evaluation checklist

Decide per zone, not per building. Reserve rack, forward pick, bulk floor and special-handling areas have different move patterns, so each deserves its own answer. The checklist below turns the calculations in this article into a decision. Our earlier guide to warehouse slotting decisions covers which SKUs earn forward pick faces; this article covers how positions are assigned once that choice is made.

Run the space formula and the travel model on your own data, then pilot class-based random storage in one reserve aisle group for a month and compare positions used, moves per hour and count accuracy. More warehouse design guides are on the logistics and supply chain hub.

If you want this analysis run on your SKU data and your layout, or need location control and license plate scanning set up in your warehouse system, share your current layout and move history through our request for proposal form, and we will reply with a sized recommendation for each zone.