In short: The standard cost you are comparing a quote against is fully absorbed, and the allocated fixed share of it stays in the building after the volume leaves, so the comparison has to be rebuilt on avoidable cost only. Doing that on a typical part turns a headline saving into a small loss before freight and inventory are counted, and the overhead left behind raises the standard cost of everything else the plant makes, which makes the next candidate look like an easier buy. The honest output of the exercise is a break-even volume, and where that volume sits inside the forecast interval the cost model has run out and the decision turns on how reversible the commitment is.
A quote lands at 4.60 a unit. The internal standard cost for the same part is 5.30. Someone builds a slide showing 0.70 a unit across 240,000 units a year, calls it 168,000 of annual saving, and the sourcing committee approves it in eleven minutes because the arithmetic is not in dispute.
Eighteen months later the part comes from a contract manufacturer, the saving is not visible anywhere in the plant's accounts, and the standard cost of everything else that plant makes has gone up. The execution was fine. The comparison underneath it was wrong before anybody started negotiating.
Avoidable cost is the only cost in the comparison
The 5.30 is a fully absorbed figure. Break it apart and it usually looks something like this: 2.40 of material, 1.10 of direct labour, 0.55 of variable overhead, and 1.25 of allocated fixed overhead. The first three follow the volume out of the door. The fourth mostly does not.
Allocated fixed overhead is the plant's rent, its depreciation, its quality lab, its planners and its site management, divided across whatever it produced that year by machine hours or labour hours or units. Moving one part to a supplier removes that part's share of the divisor. It does not remove the rent.
The test to apply to every line of the fixed allocation is whether the cash actually stops. Some of it does. If the part ran on two dedicated machines with a dedicated supervisor and a maintenance contract, and all of that goes when the part goes, that is genuinely avoidable fixed cost. Say it comes to 96,000 a year, which is 0.40 a unit at current volume. The other 0.85 a unit of allocation stays.
Now redo the comparison. The avoidable cost of making is 2.40 plus 1.10 plus 0.55 plus 0.40, which is 4.45. The quote is 4.60. Buying is 0.15 a unit more expensive, roughly 36,000 a year worse, before freight, duty, inventory or anything else has been counted. What looked like a 168,000 saving was 204,000 of unavoidable overhead being reclassified as somebody else's problem.
Two things make this error persistent. The standard cost is the number the system will hand you, and it takes a day of work with a plant controller to produce anything better. And the fully absorbed figure is the correct number for valuing inventory and reporting margin, so nobody in finance thinks of it as wrong.
The overhead you moved lands on something else
The 204,000 that stayed behind gets recovered from whatever the plant still makes. If the outsourced part was consuming twelve per cent of the machine hours the site absorbs against, removing it raises the recovery rate on every remaining product by around fourteen per cent.
This is the mechanism Cooper and Kaplan set out in Harvard Business Review in 1988 when they argued that allocated cost systematically misleads product-level decisions. Every remaining product now carries a higher standard cost. Apply the same make or buy template to the next candidate using the new numbers and it looks like a clearer buy than the first one did. The one after that looks clearer still. Each analysis in the sequence is arithmetically correct, and the sequence walks a plant down to a volume that cannot carry its own fixed base.
The tell is a site whose overhead recovery rate has risen three years running while its volume has fallen, with a procurement function reporting savings across the same period. Both statements are true and they are describing the same event.
Where the two cost lines cross
Once the comparison runs on avoidable cost, the useful output of it is a volume.
Making costs 96,000 of avoidable fixed cost a year plus 4.05 a unit. Buying costs 4.60 a unit. The two lines cross where 96,000 equals 0.55 times the volume, which is about 174,500 units a year. Above that, making is cheaper. Below it, buying is.
That single number is worth more than the recommendation, because everyone in the room has an opinion about volume and nobody has an opinion about overhead absorption. Put the crossover next to the three-year volume forecast and the conversation becomes concrete. If the plan is 240,000 and the realistic range is 140,000 to 260,000, then the break-even sits inside the range, and the cost model has genuinely stopped being able to decide. If the plan is 240,000 with a range of 215,000 to 265,000, the model has decided, and the answer is to keep making it.
The break-even also tells you what to watch afterwards. A make decision taken at 240,000 against a crossover of 174,500 has about twenty-seven per cent of headroom, and the review trigger is a volume forecast that drops through that line rather than an annual calendar slot.
Base load in, peak out
The binary framing is usually a false one anyway, because internal capacity is finite.
Suppose the line can produce 200,000 units a year and demand averages 240,000 with real seasonality. Some volume is going outside whatever anyone decides. The structure that tends to survive contact with reality is to make the base load and buy the peak, which keeps the internal line at high utilisation where its unit cost is lowest, puts the variability onto a supplier who is set up to absorb it, and preserves the process knowledge and a live cost benchmark for the next negotiation.
Sizing that internal capacity is a capacity-under-uncertainty problem with the same shape as the trade-off J11 works through, and the answer moves with the cost asymmetry between idle internal capacity and bought overflow.
The hybrid carries its own bill. Two qualified sources means two qualification programmes and two sets of tooling. Split volume weakens your position with both parties. And a supplier quoting for peak-only volume prices it as peak-only volume, so the 4.60 you were quoted for the whole requirement is not the number you will get for the top 40,000 units.
The costs that arrive after the decision
The quote covers the part at the supplier's door. Most of what changes afterwards sits outside it.
Lead time is the largest of them on anything sourced at distance. Move from a two-week internal cycle to eleven weeks from an overseas contract manufacturer and you are carrying nine extra weeks of pipeline. At 4,615 units a week that is about 41,500 units of additional inventory, roughly 191,000 of working capital at the buy price, and at a twenty per cent annual carrying charge that is 38,000 a year, or 0.16 a unit. Safety stock against the new lead time variance sits on top, and I3 covers why that second term is usually the bigger of the two.
Run the whole thing forward. The headline 0.70 saving became a 0.15 loss on avoidable cost, and the pipeline takes it to about 0.31 a unit against making, or 74,000 a year. Still to come are freight and duty, the amortisation of tooling, dual running during transition, the quality cost of a new process while it settles, minimum order quantities that raise cycle stock again, which SS5 treats as its own negotiation, and the coordination time of managing a supplier eight time zones away.
Buying may still be the right call. What these additions do is make the decision close, and close decisions should be taken on the reversibility rather than on the second decimal place.
Why coming back costs more than going out
The two directions are priced very differently, and only one of them gets quoted.
Exit costs are visible and negotiated: severance, equipment disposal, tooling transfer, a qualification programme with a date on it. Re-entry costs are none of those things. You need equipment you sold, operators who took other jobs, a supplier's process knowledge that you no longer hold, and a requalification cycle that customers in regulated categories will police. The practical horizon on undoing a make or buy decision in a machining or moulding category is twelve to twenty-four months and a capital request nobody wants to sign.
Monteverde and Teece looked at this empirically in the automobile industry in 1982 and found that the more engineering-specific a component was, the more likely it was produced in-house. That result is about switching cost and appropriable value rather than about unit cost, and it is the reason a three per cent quote advantage on a part with dedicated tooling is not a decision.
Dixit and Pindyck's 1994 treatment of investment under uncertainty gives the general form. An irreversible commitment made under uncertainty needs a larger margin than a reversible one, and the required margin grows with the volatility of the thing you are uncertain about. Applied here: the noisier your volume forecast and the more supplier-specific your tooling, the bigger the cost gap has to be before the move is worth making. A three per cent advantage on a stable, catalogue-standard part with second sources available is a reasonable bet. The same three per cent on a bespoke part with one qualified supplier and a wide volume range is inside the noise.
Fine's 1998 work on clockspeed adds the part that shows up years later. Whoever does the work learns from doing it, so a supplier three years into a contract understands the process better than the team that handed it over, and that shifts the balance in every subsequent price conversation.
Where this stops
The avoidable cost study is the hard part, and most finance systems cannot produce it. What they hold is an allocation. Getting to avoidability means walking the cost centre line by line with the plant controller and asking, for each one, what stops. It takes days, it produces a number with a range rather than a point, and part of the answer is a judgement about a future decision, because a supervisor is only an avoidable cost if somebody actually removes the role.
That judgement is where the analysis is most often gamed, in both directions. A team that wants the outsourcing to happen classifies generously. A plant defending its volume classifies nothing as avoidable. The protection is to write down, at decision time, the specific cost lines that are supposed to stop and the month they stop in, then check that list twelve months later. Very few organisations do this, which is why so few outsourcing savings can be found in the accounts afterwards.
The break-even framing has its own boundary. It assumes the supplier's price holds as your volume moves, and on a part where you are a meaningful share of their capacity it will not. It also treats the internal variable cost as constant, which is fair over a modest range and wrong once you are talking about halving the volume through a plant.
And the whole exercise is silent on the reasons that actually decide some of these cases. Intellectual property, regulatory exposure, a customer contract that names the manufacturing site, or a deliberate strategy to hold a capability. Those are legitimate grounds to override the cost answer in either direction, and they are better stated as overrides than smuggled into the model as an adjustment factor that nobody can audit.
Take the last three make or buy decisions your business signed, sit down with the plant controller, and ask which specific fixed cost lines actually stopped and when. If the answer is that the overhead recovery rate went up instead, you have found the pattern and you can price the next one properly.