Make-to-Order or Make-to-Stock: It Is Arithmetic, Not Preference

The sales manager promises three weeks because the customer said three weeks and the order was worth taking. The extrusion line is booked for five. The resin grade the part needs is on a vessel that has not left the port of origin, and the supplier's quoted lead time was six weeks when the purchase order went out eleven days ago.

Everybody in the building knows this order will be late. They knew on the day it was accepted. What happens instead of being late is a sequence: the planner pulls it forward, which pushes two other orders back, one of which belongs to a customer who will notice; the machine is changed over out of sequence, which costs purge and startup scrap; and somewhere in the following fortnight an air freight charge appears against a job that was quoted on sea freight.

None of that is a planning failure. It is the predictable consequence of a promise made without reference to a number that nobody in the room had.

The number

Every manufacturing business has two lead times, and the relationship between them determines almost everything about how the plant should be run.

The first is the cumulative lead time to make and deliver a product starting from nothing: procure the longest-lead raw material, receive it, run every operation in the routing, including any subcontracted step, pack it and ship it. Call it the production lead time.

The second is how long the customer will wait between placing an order and expecting delivery. Call it the demand lead time.

If the production lead time is shorter than the demand lead time, you can make to order. The customer's wait is long enough to contain your process, so there is no operational reason to hold finished stock and take the risk of holding the wrong thing.

If the production lead time is longer than the demand lead time — which is the normal case for anyone importing raw material into this region — then some part of the work has to be done before the order arrives. That is not a strategy. It is arithmetic. The only remaining question is which part, and where you stop.

The ratio between those two lead times is the single most useful figure a manufacturer can put on a whiteboard, and very few can produce it per product family without a week of work. Producing it is worth the week.

The decoupling point

The place where you stop working ahead and start waiting for a customer is the decoupling point. Upstream of it, everything is driven by forecast and held as stock; downstream of it, by a real order. The decision is not binary and it is not made once for the company. It is made per product family, and moving it is the most powerful lever a manufacturer has over both service and working capital.

Where you decouple	What you hold	What the customer waits for	Where the risk sits	Typical of
Finished goods in your warehouse	The complete product	Picking and delivery	Obsolescence and cash tied up in the wrong variants	Food, fasteners, standard cable drums, catalogue items
Finished goods, unlabelled or unpacked	Bulk or unbranded product	Packing, labelling, private label artwork	Modest, if the bulk is common to several customers	Food and beverage, cosmetics, private-label anything
Sub-assemblies or semi-finished	Components, extruded core, cut blanks	Final assembly and test	Low, if the sub-assembly serves many end items	Furniture, panels, assembled equipment
Raw material only	Sheets, resin, rod, fabric	The whole conversion process	Low on obsolescence, high on capacity	Fabrication, injection moulding to customer tooling
Nothing — buy on order	Nothing	Procurement plus conversion	Almost none on stock, all of it on the delivery promise	Engineer-to-order joinery and steelwork, project supply

Read that table with your own product families in mind and one question attached to each: what is the cheapest thing I can hold that removes the most weeks from the customer's wait? The answer is almost never the finished product, and it is almost never nothing.

The middle rows are where the money is, and they are the rows most factories never seriously consider. Holding a common semi-finished item that serves eleven end items pools the demand variability of eleven products into one stock position. Holding the eleven finished items instead requires more total inventory for the same service level, and half of it will be the wrong half.

Quoting a lead time you cannot hold

This is the most expensive habit in manufacturing and it does not appear as a line item anywhere.

A delivery promise is a claim on capacity. When it is made without checking whether the capacity exists, it does not fail politely by being late — it succeeds at the expense of every other order in the plant. The costs land in five places, none of them attributed to the order that caused them.

The changeover taken out of sequence, so the line runs light after dark or a material change that could have been avoided. The overtime. The premium freight, inbound to get the material and sometimes outbound to recover the date. The order that got pushed and became the next emergency, which is how a plant enters the state where every week is firefighting. And the erosion of the promise itself: once the factory learns that the published lead time is negotiable, nobody plans to it, and the plan stops being a plan.

The fix is not a rule that sales must consult production. That rule exists in every factory and is broken in every factory, because it depends on someone being available and willing to say no under commercial pressure. The fix is to make the available-to-promise date a fact the sales system produces rather than an opinion a person offers.

Two levels of that, and the difference matters when you are specifying a system.

Available to promise checks the stock position: what is on hand, less what is already reserved for other orders, plus what is on confirmed inbound supply within the horizon. It answers the question for anything sold from stock and it is standard in most mid-market products.

Capable to promise goes further and checks capacity: given the current order book on the constraining work centre, when could this actually be made? It is much less commonly implemented properly, it requires routings and work centre capacities that are true, and for a make-to-order plant it is the difference between a quoted date and a guess with a signature.

Most factories should not attempt capable-to-promise across the whole plant. They should implement it on the constraint — the press, the paint line, the extruder, the CNC bay — because the plant moves at the speed of its slowest machine and every other work centre has slack that hides the problem.

Planning under each regime

The two regimes need different planning machinery, different measures and different people looking at them. Running one set of habits across both is where most hybrids come apart.

	Make to stock	Make to order
What triggers production	A forecast, a reorder level, or a replenishment calculation	A confirmed customer order
The main planning input	Demand history and its variability	The order backlog and capacity
The main risk	Holding the wrong thing	Promising a date you cannot hold
What safety buffers	Stock	Time and capacity
The measure that matters	Service level from stock, and stock turns	On-time in full, and lead time adherence
Costing basis that fits	Standard cost, with variance analysis	Job or order costing, against the estimate
Failure mode	Slow-moving stock nobody will write off	A backlog that is really a queue of broken promises
Who has to be good at their job	The demand planner	The scheduler and whoever quotes dates

The safety buffer row is the one that gets missed. A make-to-stock line buffers uncertainty with inventory. A make-to-order line has no inventory to buffer with, so it must buffer with time — a realistic lead time that contains normal variability — or with reserved capacity. A plant that quotes its theoretical minimum lead time as its standard lead time has removed its only buffer and will be late whenever anything at all goes wrong, which is weekly.

Safety stock deserves one honest sentence, because it is set by feel in most factories. The right level depends on how much demand varies during the replenishment lead time and on the service level you have chosen to offer — and the service level is a commercial decision with a price, not a target to be set at 100 and forgotten. Doubling the buffer does not double the service level; the return diminishes sharply, and the last few percentage points of availability cost more than all the ones before them. Anybody setting safety stock without having decided the service level first is spending money without knowing what they are buying.

The hybrid, which is what you actually run

Almost no factory in this region is purely one or the other. The catalogue items run to stock, the project and bespoke work runs to order, and both go through the same machines and the same store. That shared capacity is where the design work is.

Five things change in the system when a plant runs both at once.

The regime is a property of the item, not of the company. Each item or item family carries its replenishment strategy: replenish on order, replenish on stock rule, or manufacture on the planning run. Get this wrong on a handful of items and the planning output becomes untrustworthy, at which point the planner starts overriding every line and the run becomes decoration.

Reservations have to be real. A make-to-order job that consumes material earmarked for a stock replenishment is an ordinary Tuesday in a plant where reservation is advisory. If the system allows the storeman to issue reserved material to whoever asks first, then both regimes are running on the same pile of unallocated stock and neither plan means anything. This is also the point at which an inaccurate stock figure stops being a reporting problem and becomes an operating one.

There must be a written rule for who yields. When the constraint is oversubscribed and one customer order and one stock replenishment want the same four hours, somebody decides. If the rule is unwritten, the decision goes to whoever shouts, and the stock replenishment always loses because no individual customer is waiting for it — until the day the catalogue item is out of stock and everyone is surprised. A simple, published rule, applied consistently, outperforms a clever rule applied inconsistently.

The two halves must be measured separately. A single on-time delivery percentage across both regimes tells you nothing, because failing a stock item and failing a bespoke item have different causes and different fixes. Report service level from stock for the stock items and on-time in full against the promised date for the order items, and report the promised date against the originally quoted date as a third figure — that gap is the size of the over-promising problem.

Costing splits. Standard costing suits the repetitive stock lines where volume justifies maintaining standards. Job costing suits the bespoke work where every order has its own material and labour. Running one basis across both produces either meaningless variances on the bespoke jobs or no visibility at all on the stock lines, and this decision is taken during configuration far more often than by a finance director.

The two ways it usually goes wrong

Everything to order, because we do not want stock. This is common and it sounds disciplined. What it actually does is transfer the entire lead time onto the customer and the entire volatility onto purchasing. If your production lead time exceeds what customers will wait, a pure make-to-order policy does not eliminate inventory — it converts inventory into expediting, air freight and missed dates, which are more expensive and appear in different accounts, so nobody connects them to the policy that caused them.

Everything to stock, on a catalogue with real variety. The mirror failure. Six colours and four sizes is twenty-four finished items, and forecasting twenty-four items well is much harder than forecasting the family and holding the common part. The result is a warehouse where a third of the value has not moved in a year and the item the customer wants is the one out of stock. In cut-to-size businesses the same disease shows up as offcuts and remainders — material below a saleable length that is held forever or given away, decided at the cutting table with no view of the alternative.

Choosing per item, in an afternoon

You do not need a modelling exercise. You need two axes and a list of your items sorted by revenue.

Volume and stability on one axis: how often is this ordered, and how much does the quantity vary? Lead time coverage on the other: does your production lead time fit inside what the customer will wait?

The classification that falls out is old and still the most useful one in the trade. Runners are ordered constantly in stable quantities — hold them, and hold them as finished goods if they are cheap and small. Repeaters come round regularly but not weekly — decouple at the semi-finished level where a common component serves several of them. Strangers are the ones ordered twice a year by one customer — make them to order, quote a real lead time, and stop apologising for it.

The revealing exercise is to run this classification and then compare it against what you currently hold. Most factories find they are holding finished stock of strangers, because one was ordered urgently in 2023 and somebody set a reorder level in response, and it has been replenishing quietly ever since.

Two industry-specific wrinkles are worth naming, because they move the answer.

Where the main raw material is an exchange-traded commodity, the decoupling point is also a pricing decision. Holding raw material means holding a position on the metal price, and quoting a fixed price with a long validity while buying later means taking that position on the sales side instead, usually without anyone calling it that.

Where the customer owns the tooling, the regime is partly theirs to choose. A moulder running customer-owned tools against a call-off schedule is effectively making to stock on the customer's forecast and carrying the risk if the call-offs stop, which is a contractual matter before it is a planning one.

What to fix first

Produce the two lead times per product family and write them down. That single exercise usually reveals that the published lead time was set years ago, has never been measured against actuals, and is shorter than the plant has achieved in any of the last twelve months.

Then set the decoupling point deliberately for the families that matter, mark the regime on the item master, and make the promise date come out of the system rather than out of a conversation. Only after that is it worth arguing about scheduling tools, because a scheduler working against dates that were never achievable is optimising the wrong problem.

If the two lead times have never been measured and the decoupling point has never been chosen, that is the work to do first, and it is exactly the ground the diagnosis stage of an ERP implementation is meant to cover before anybody configures anything.