The flooring sat in a warehouse for a year

Every warehouse decision depends on seven numbers most businesses have never pulled.

This is how to get them, and what each one decides.

The confessionWhy this happens

We bought a house, and we bought the flooring for it at the same time. Both real decisions, both correct ones. Then it turned out the house needed a rebuild before any flooring could go anywhere near it, and we collected that flooring from the supplier a year later.

Nothing was wasted. The flooring was fine. It was simply a year early, and for that year it sat in somebody else's warehouse occupying space it had not earned — which, for a paper about warehouses, is a joke that writes itself.

The tiles were worse. Those were bought before the rooms had been contemplated at all — before anyone had decided which room was which, how many there would be, or what size. You cannot know how many square metres of tile you need until you know what you are tiling. We bought them anyway.

I laughed about both at the time. Then I noticed I do the same thing at work, and so does nearly every business I visit.

The finish is available to choose today. It is tangible, there is a showroom, somebody is enthusiastic about it, and choosing it feels like progress. The structural work is a drawing that takes weeks and makes nobody happy.

There is a general pattern under both, and it is not really about houses. Nobody argues about the damp course. Nobody has strong views on the load-bearing wall. Everybody has an opinion on the colour, the wallpaper, the worktop, the handles.

That is exactly backwards, and it is backwards for an understandable reason. Finish decisions are visible, reversible and cheap — which is precisely why everyone feels entitled to weigh in on them. Foundation decisions are invisible when they are right, catastrophic when they are wrong, and permanent either way. So attention flows toward the decisions that matter least, because those are the ones people can see and judge.

What this looks like in the field

I visit a lot of warehouses, and two versions of this come up constantly.

The second is quieter and more expensive: "we put in racking, because that is what you do." Standard selective racking, single deep, every pallet accessible. It is the default, nobody questions it, and for a range with many SKUs and one or two pallets each, it is correct.

But I have sat with businesses holding six or eight pallets of the same article and watched the realisation land that push-back would have given them two or three times the density in the same footprint. They gave up accessibility they did not need, on a range where LIFO was never a problem, and paid for it in floor area every year since.

The racking case is the one that bothers me most, because it never feels like a decision. Nobody chose selective racking over push-back. Selective racking is simply what arrives when nobody chooses.

It is part one of five, and this one is deliberately the least exciting, because everything in the four that follow is unsizeable without it.

The argumentYou cannot size what you have not measured

There are seven figures. None of them requires a consultant, a site visit or a survey. All of them are sitting in your ERP right now, and in most businesses nobody has ever pulled them.

Each one decides something specific and largely irreversible. That is the part worth holding on to — this is not a data-gathering exercise for its own sake. Each figure is the input to a decision you are going to make anyway, with or without it.

Figure oneOrder profile — and why the average lies

Lines per order, and units per line. Not the average — the distribution.

This matters more than any other single figure, because a warehouse where most orders are one line behaves nothing like one where most orders are twenty, and the average of the two tells you nothing about either.

How to get it. Count lines per order across twelve months of order headers, then plot the frequency of each line count. Twelve months, not three — you need the seasonal shape in there. Look at the shape, not the mean.

What to look for. A single-line dominant profile means the walk is your cost and batching is your answer. A clustered multi-line profile means consolidation is your cost and zoning is your answer. A genuinely bimodal profile — a lot of one-liners and a lot of twenty-liners — means you have two businesses in one building and should probably run two processes.

The fields

Derive: count of lines per order_id, then the frequency of each line count. Also units per line, as a separate distribution.

Exclude: cancelled orders, internal transfers, samples and staff orders. Keep them separate rather than deleting them — if internal movements are 15% of your line volume, that is a finding about the operation, not noise.

Split by channel if you have one. A wholesale channel and a webshop in the same data will average into a profile that describes neither.

Figure twoVolume, and the shape of the year

Orders per day is not the number. Peak-to-average is the number, and where the peaks land.

Capacity is sized for peak. Everything — dock doors, pick faces, staffing, despatch lanes — has to survive the worst week of the year, not the median one. A business that sizes on the average builds something that works for ten months and falls over in the two that matter most.

Two refinements worth making. Use the busiest week rather than the busiest day, because one freak day is noise and a sustained week is a design constraint. And check whether your peak is demand-driven or promotion-driven — a peak you create yourself is one you can move.

And the peak inside the day

There is a third peak underneath the seasonal one and the day-of-week one, and it is the one almost nobody measures: what time of day do orders actually arrive?

Take two businesses shipping 1,200 orders a day. One receives them evenly from eight in the morning. The other receives most of them between three and five in the afternoon, against a five o'clock cut-off. Identical daily volume. Completely different building.

The second one has to do the day's work in a window, and that window sizes headcount, handling equipment, pack benches, despatch lanes, staging area and dock doors all at once. Size any of them on the daily average and they will be wrong by a wide margin.

Two things to note while you have it. Whether the arrival curve is customer behaviour or a consequence of your own cut-off — because if it is the second, it is movable, and moving it is far cheaper than building for it. And whether carrier collection times leave you a realistic window at all, which is the outbound mirror of the dock arithmetic in figure five.

Sizing comes in part two. What matters in part one is that you have measured it, because it cannot be reconstructed later from data that only records the date.

The fields

Derive: orders per calendar day over 24 months; busiest rolling 7-day window ÷ annual daily average; and orders per hour of day across a typical week and a peak week. Also lines and units per day, because order count alone hides a peak where orders get bigger rather than more numerous.

Use ship date, not order date, for capacity sizing. The warehouse feels the day it ships, not the day it was sold. If the two diverge a lot, that gap is itself a capacity buffer you are already using.

Figure threeSKU movement — and how many pick faces you actually need

This is the figure that sizes the building, and it is the one I see guessed most often.

Not every SKU deserves a pick face. A pick face is expensive floor area at working height in your most valuable square metres. A SKU that sells half a unit a day and sits in a pick face holding forty units is occupying prime space for eighty days at a time.

The rule I would use for the cut-off

Forget percentages for a moment. The honest test is turn rate at the pick face.

If a SKU turns its pick face less than once a month, it does not belong there. It is using prime space as long-term storage. Put it in bulk and pick from the pallet.

At the other end, if a SKU turns its face more than once or twice a day, you have the opposite problem — you are paying for constant replenishment labour, and that SKU wants a deeper face, a full pallet position, or its own flow lane.

Getting to the number

Worked through, with illustrative figures:

That number, multiplied by the footprint of the location type your product profile requires, is a floor area figure. It is the first genuinely hard input to part two, and you now have it.

Why I am insistent about this one

I was once shown around a customer's brand new warehouse. Finished, occupied, operating. I walked the pick face, counted the locations, and compared it against the SKU count they had just told me on the way in.

I told them on the spot that they were about 15% short of picking locations — for the range they already carried. Not for the growth they were planning. For the stock sitting in the building that day.

That is the worst version of this problem, because a new building is the one moment when the constraint is genuinely fixable and cheap. By the time you are standing in it, the options are all bad ones: convert bulk into pick face and lose storage, go to double-deep and lose accessibility, add a mezzanine, or start planning the next move.

Note what was not wrong. The building was fine. The racking was fine. Somebody had sized the storage cube properly, and the pallets all fitted. The number nobody had calculated was how many SKUs needed a face — which is figure three, and which is the number that does not fall out of a cube calculation at all.

The fields

Derive two numbers per SKU, not one: pick count (number of times someone goes to that location) and pick quantity (units leaving it). The first drives how many faces you need; the second drives replenishment frequency at each face. A SKU with high count and low quantity wants a small face visited often. High quantity and low count wants a deep one.

Count pick lines, not order lines. If one order line is satisfied from two locations, that is two picks and two visits.

Filter to active SKUs — status, and something like at least one movement in the last twelve months. Dead range inflates the location count and is a separate conversation.

Figure fourPhysical characteristics — bigger or taller

Dimensions, weight, stackability, and anything needing temperature control or segregation.

What this decides is the one question that separates an expensive building from a very expensive one: do you need more floor, or more height?

My rule: screen sites on permitted height

My default is to build as high as I am allowed to, and I would go further — when you are choosing a location, permitted building height is one of the first things to check, not one of the last.

The economics are not close. Land, slab and roof are the expensive components per square metre. A taller building adds none of them — it adds wall and steel, which are comparatively cheap. A second level of pallet positions costs a fraction of a second hectare, and that stays true even when you cross into a more expensive sprinkler or equipment class, because the alternative is not two more metres. The alternative is another footprint, with its own slab, its own roof, its own land and probably its own dock.

There is an exit argument on top of the operating one. A taller building holds more usable volume on the same land, so it is worth more per square metre when you sell it and easier to let when you do not. Clear height is one of the first things any logistics tenant or buyer asks about, and you are not going to add it later.

That argument does have an edge, and it is worth knowing where it is. Height in the ordinary range widens your pool of future occupiers. Height that only works with a crane structure, a superflat floor and a specific automation layout narrows it — you have built an asset that suits one kind of operator. Generous clear height is liquid. A purpose-built automated high bay is not.

And an automated high bay is usually well above what any ordinary zoning allows, so it is not a design choice at all — it is a planning application. A different process, a different timeline, objections from people who have never been in a warehouse, and a real chance of refusal after you have spent money finding out.

This is why height belongs in part one rather than part two. By the time you are designing the shell, the ceiling has already been set by a decision someone made about a plot.

Where to stop

Within what you are permitted, there are bands where cost per position steps rather than glides. Knowing where they are means you stop just under one rather than just over it.

The expensive version is landing two metres above a band for no reason, paying the sprinkler class, the equipment class and the floor specification to gain almost nothing. That is a design detail, not an argument against height.

What height actually buys: options

The obvious use of height is pallet racking, and pallet racking is bulk storage. That is where the lazy version of this argument stops, and it is wrong.

Height also lets you build floors. A mezzanine, a two or three tier pick module, a pick tower. Each level is working height, and working height is where pick faces live. That is precisely how high-SKU-count operations get their location count without buying more land — they stack the pick face rather than the pallets.

That qualifier is doing real work. A tiered pick structure needs the floor loading to carry it, column spacing that does not cut the levels into unusable strips, fire detection and egress for the upper levels, and a way to get goods down — conveyor, goods lift, chute. Some of that can be retrofitted. Some of it cannot, and the parts that cannot are decided in the shell.

Which is the argument for height stated properly. It is not that a taller building stores more. It is that a taller building keeps decisions open. You can convert a bay to racking, or to a mezzanine, or to a pick module, and you can change your mind in five years when the range has grown. In a building that is too low, none of those decisions exist — there is one thing to do with the space and you are already doing it.

The one asymmetry to plan around

Cost per position falls as you go up. Cost per pick rises — vertical travel per retrieval, and VNA or crane cycles are slower than a reach truck at low level.

That does not argue against height. It tells you what to put where. Pallet storage and slow movers high, fast-moving pick faces low or on a tiered structure sized for them. Both economics work at once, and you only get to arrange it that way if the height is there in the first place.

And the calculation that still gets skipped

None of this rescues you from figure three. A building can be tall, capable of carrying a mezzanine, and still open short of picking locations — because nobody counted how many SKUs needed a face.

That is what had happened in the warehouse I described earlier. The cube was right. Whether the shell could have taken a pick module is a question I do not know the answer to, and neither did they, because it had never been asked.

Cube is the easy calculation and it is the one that gets done. It answers whether the stock fits. It says nothing about whether you can pick it.

The fields

Derive: pallet equivalents held at average and at peak stock; the distribution of case and pallet sizes; how much of the range is non-stackable, temperature-controlled or hazardous.

Measure completeness first. Run a percentage-populated check on every dimension field before you use any of it. In most businesses this comes back between 50% and 80%, and everything downstream is an estimate until it is fixed. That is a finding worth reporting on its own.

Figure fiveReceiving profile — the dock door arithmetic

Inbound vehicles per day at peak, average dwell time per vehicle, and the operating window. Three numbers, and almost nobody has them.

Dock doors are the purest example of an irreversible decision. Adding one to an existing building means cutting the wall, relocating the apron and probably losing yard circulation. It is the closest thing in a warehouse to moving a load-bearing wall.

The clustering factor is the whole point. A flat arrival profile needs two doors; the same volume arriving in a morning cluster needs three. Most businesses calculate the first and build the second badly.

Look up while you are standing in the receiving area

Receiving is the clearest example of the optionality argument, so it is worth making concrete here rather than leaving it abstract.

A receiving area needs floor. It needs apron, staging lanes, a check bench, somewhere to put a pallet that has failed inspection. What it does not need is height. Above a receiving area, in most buildings, is dead air — and it sits next to the dock, which is where power, data and the main entrance already are.

Three good things to put there:

• Offices. Near the entrance, services already run to that end of the building, and they overlook the operation — which is worth more than it sounds. A supervisor who can see the floor from their desk asks different questions than one who cannot.
• A mezzanine for value-added services. Labelling, kitting, repacking, customisation. These are bench activities that need floor area, not volume, and they sit naturally between goods arriving and goods going to stock.
• Returns processing. Returns flow into receiving, not out of despatch — inspect, decide, back to stock or onward. Putting them above receiving keeps that adjacency and stops them colonising the pick area, which is what happens when nobody allocates them a room.

The usual constraints apply and they are all shell decisions: floor loading, column positions that do not cut the mezzanine into awkward strips, fire detection and a second means of escape from the upper level, and a goods lift or conveyor if anything heavier than paperwork is going up there.

One caution on offices specifically. A dock is noisy and it vibrates, and visitors should not have to cross the operation to reach a meeting room. Both are solvable, but they are solved at design time rather than discovered afterwards.

The cheapest fix in this entire paper is not a door. It is booking slots. If you can flatten the arrival curve by asking carriers to book, you may save yourself a door and the wall it goes in. That is a process decision with a capital consequence, which is exactly the kind of thing that only becomes visible when you do the arithmetic first.

The fields

Derive: receipts per day at peak; dwell time as completion minus arrival; and the arrival distribution by hour of day, which is the clustering factor in the calculation above.

Arrival time is usually missing. Most systems record when receiving was booked in, not when the vehicle appeared on the yard. If that is your situation, say so and measure it manually for two weeks — it is the input the door count is most sensitive to, and two weeks of a clipboard beats a year of guessing.

Figure sixReturns — the room nobody allocates

Returns volume as a percentage of outbound, and what happens to a returned unit.

I include this as its own figure rather than folding it into volume, because returns are almost universally forgotten at design time and then colonise whatever space is nearest the despatch door — which is usually your pick area.

The question is not just how many. It is what the process is: inspect, repack, return to stock, refurbish, scrap. Each of those is a different amount of space and a different adjacency. A business returning 4% with a straight back-to-stock flow needs a bench. A business returning 25% with inspection and repack needs a room, and it needs it next to receiving rather than next to despatch.

The fields

Derive: return rate as returned lines ÷ outbound lines, by channel; and the disposition mix — straight back to stock, inspect first, repack, refurbish, scrap.

The disposition mix decides the space, not the rate. Ten percent going straight back to stock is a bench. Four percent needing inspection and repack is a room with people in it.

Figure sevenThe shape of the business — now and intended

The last figure is not a number. It is what the business is, and what it intends to become.

I have put it last because it is the one most often answered with a growth percentage, and a growth percentage is the least useful form of the answer. "We expect to double" tells you almost nothing. Doubling what? Order volume with the same profile is one building. Doubling SKU count at the same volume is a different one — more pick faces, more storage, identical throughput. Doubling average order size is a third. All three are "doubling" and all three size differently.

The change that catches people

The clearest example, and the one I see most: a wholesale business that intends to sell direct to consumers.

On paper it is the same product, the same stock, the same warehouse. In practice almost every figure in this paper inverts.

A wholesale distribution centre converted to B2C without being re-planned is a predictable failure: plenty of bulk storage, nowhere near enough pick faces, no packing area, no returns room, and a despatch dock designed for two lorries facing a carrier collection window it cannot meet.

None of that is fixed by floor area. It is fixed by configuration, and configuration is decided in parts two and three of this series — which is why the intention has to be on the table now, while it is still free to plan for.

Omnichannel is not the average of the two

The instinct is to build something in the middle. That is usually wrong.

Running both means running two order profiles through one building, with different pick methods, different packing, different despatch and different promises. Sometimes the right answer is two processes sharing a stock pool. Sometimes it is genuinely two areas. Rarely is it one compromise process that serves both badly.

What it always means is that the stock is contested — the same unit is wanted by a pallet order and a parcel order, and something has to decide which one gets it. That is a whole subject in itself, and it is not this series.

The questions to ask, and who answers them

None of these come out of the ERP. They come from the board, and they should be asked before a plot is chosen.

• What channels do we sell through today, and what share does each take?
• What channels do we intend to add, and roughly when?
• Is the range growing, shrinking or changing character?
• Are we moving up or down in order size — bigger customers or smaller ones?
• Are we going to hold stock for anyone else, or let anyone else hold ours?
• Is any part of this business likely to be sold, or acquired into?

That last one is not idle. A business that may acquire needs a building that can absorb somebody else's range. A business that may be sold needs one that is attractive to a buyer, which is the liquidity argument from figure four.

Write the answers down with dates and confidence levels against them. "B2C pilot within two years, uncertain" is a genuinely useful planning input. "We want to grow" is not.

My rule: plan for 50% minimum

Whatever the range says, I would not design for less than 50% headroom above where you are today. That is my floor, not my forecast.

The reason is not optimism about growth. It is the cost of being wrong in the other direction. I know of businesses that moved four times in thirteen years — a move roughly every three years, each one sized to the business as it was at the time of signing rather than as it would be.

People price a move as removal and fit-out. The real bill is longer:

• overlapping rent on two sites during transition
• racking dismantled and rebuilt, usually not quite fitting the new bay spacing
• re-slotting the entire operation, and the WMS configuration that goes with it
• a throughput dip that lasts weeks and lands on customers
• the staff who do not come with you — move twenty kilometres and you lose warehouse people you cannot replace in a tight labour market

Headroom does not have to be empty space

The objection to 50% is obvious: you are paying rent or capital on space you are not using, possibly for years.

Which is why headroom is better held as capability than as floor. You do not need 50% more building standing empty. You need 50% more accommodatable — and most of that is specification rather than square metres:

• clear height that allows a mezzanine or a tier to be added later
• floor loading that will carry one
• column spacing that does not make the addition awkward
• dock positions framed into the wall line even if the doors are not fitted
• power and data capacity above current draw, with routes to where equipment might go
• a yard that still works with more vehicles in it

All of those cost little at build time and are expensive or impossible to retrofit. That is the same argument as the height section, arriving from a different direction — the value is in keeping decisions open, not in owning space you do not yet need.

So the honest use of figure seven is to decide two things: what you build now, and what you leave room for. Get the second one wrong and you are looking at move number two.

Doing itPulling the data

All seven figures come from three tables you already have: order headers, order lines, and stock movements. Twenty-four months where you can get it, twelve as a minimum.

What you will actually find

Two things, in my experience, every time.

The item master is worse than anyone thinks. Dimensions missing on a third of the range, weights that are obviously placeholder values, stackability blank. That discovery is itself a finding — it tells you the cube calculation will be estimated rather than measured, and it tells you what to fix before anything else.

The distributions surprise people. The order profile is rarely the shape the sales team believes. The movement curve is usually steeper than expected, which is good news — it means fewer pick faces than feared. And the peak is usually sharper than the finance figures suggest, because revenue smooths what order count does not.

Either wayNew building, or the one you are standing in

Most readers of this are not building. They have a building, the columns are where they are, and the roof is the height it is.

The arithmetic is identical. The output is different: instead of a specification, you get a gap analysis. Here is what we need, here is what we have, here is where the shortfall is and whether it is fixable inside these walls.

That is a more useful document than it sounds, because it converts a vague sense that the building is not working into a list with numbers on it. And it answers the question underneath — whether the problem is the building or the way it is being used. Usually it is the second, and that is a far cheaper problem.

NextDemand is measured. Everything else is chosen.

There is a distinction running underneath this whole paper that is worth making explicit before part two, because it explains the order of the series better than the house metaphor does.

The seven figures are demand. They are facts about your business. You do not get to choose your order profile, your SKU count or your peak — you measure them. They are not negotiable and they do not care what you would prefer.

Everything after this is response. How many people, on what shifts, with what equipment, picking by what method, against what cut-off. Those are choices. There are several legitimate answers to the same demand, and different businesses reach different ones for good reasons.

Which is why part two is not the building

I had originally intended part two to be the shell — footprint, height, columns, docks. It is not, and the reason is that you cannot size a building for a process you have not chosen.

A business meeting its peak with two shifts needs materially less staging space, fewer despatch lanes and a smaller yard than the same business running one. A business that mechanises picking needs different aisle widths, different floor tolerances and power where the racking goes. A business relying on flexible labour needs the welfare, parking and training space to absorb forty agency staff in November.

Same demand. Three different buildings.

So part two is method: labour and its flexibility, mechanisation and where it earns its place, picking strategy, shift patterns, and the order cut-off that ties them together. It takes the seven figures from this paper as its input and produces the operating design that part three then puts a building around.

Before you read it

Pull two things. The order profile distribution, and the SKU movement curve with pick count and pick quantity separated.

Those two are an afternoon of work and they change more downstream decisions than anything else on the list.

And if you are already deep in a project — already looking at plots, already talking to equipment vendors — that is not a reason to skip this. It is the reason to do it this week, while the answers can still change something.

One question to take back with you

How many of your SKUs are sitting in a pick face they turn less than once a month? Nobody in your business knows, and it is an afternoon's work to find out. Whatever else you take from this, take that.

Part one of The hows in a warehouse. Willem ten Asbroek — BizBloqs Management Solutions B.V., Netherlands. September 2026. Figures in the worked examples are illustrative; the point is the method, not the numbers. Quotation permitted with attribution and a link.