Core insight
You can outsource anything whose interface you can specify, and nothing you cannot. Once that test is passed, the decision turns on one number nobody states: the probability that you end up having to buy the capability back.
Why the make-or-buy argument never settles
Most make-or-buy analyses compare two cost estimates: own it, or keep paying a supplier. The two numbers are almost always close enough to argue about for a year, because each rests on assumptions the other side can dispute.
The argument is also usually framed on an axis that does not sort good outcomes from bad ones: how much should we own? Companies that own very little do well and companies that own a great deal fail, and the reverse. What separates them is not the amount owned but whether what they bought could be specified.
What must you own, and what can you safely buy?
The specification test
Take what you hold today for the component or service: the drawings, tolerances, acceptance criteria, service levels and interfaces. Hand it to a supplier who has never seen the work and ask whether they can quote from it.
| They can quote from it | They cannot | |
|---|---|---|
| What you are doing | Buying a part | Delegating a design |
| Who can supply it | Anybody who meets the specification | Only the firm that built it |
| If the supplier fails | Re-source it | Re-design it, or buy the supplier |
The test takes about a month and is rarely run, because it only seems worth running on a supplier who is already performing. That is exactly when it should be run. A component no outside firm could quote from is not being bought; it is being delegated, and a cost comparison against buying it is comparing a real cost with an option that does not exist.
The number that settles it: the chance of buying it back
Once the specification test is passed, the cost comparison still needs one more input that is almost always left out: the probability that you will have to bring the capability back in-house later, at a price someone else sets, plus the delay while you do.
The break-even probability
Run both paths over ten years: owning now (the capital, the overlap while you build, the running cost) against buying (the supplier's price, plus the expected cost of a forced buy-back at a premium, plus the delay). The probability of a buy-back at which the two cost the same is the number to put in the paper. Nobody can defend a probability, but everybody is already assuming one.
An illustration, from the fictional company used in the Vertical Integration Strategy Pack's worked files: the two point estimates are $1,947 million and $2,033 million, close enough to argue about all year. The finding is not the $86 million between them. It is that the break-even probability is 40.1%: the decision rests entirely on whether the chance of ending up owning this anyway is above or below that, and no one had ever stated it.
Buy-backs happen. Boeing sold its Wichita fuselage operations in 2005 for about $900 million in cash plus the transfer of certain liabilities, to become a leaner systems integrator. In July 2024 it agreed to reacquire Spirit AeroSystems at an enterprise value of about $8.3 billion.
What these decisions show
Each case below is drawn from the Stratrix case library, checked against the filings and reporting listed under Sources. The last column says what each one does not prove.
| Company | What it decided | What happened | What it does not establish |
|---|---|---|---|
| Apple, 2008 | Bought P.A. Semi for $278 million to design its own chips | The A4, its first in-house chip, shipped in 2010; the A6 in 2012 used a fully custom design | That every company should design its own silicon |
| Amazon, 2013 | Built its own delivery network instead of renting carriers | Parcels rose from about 2 billion in 2019 to 5.9 billion in 2023, passing UPS and FedEx by volume | That the network is profitable on its own |
| Nucor, 1969 | Built its own steel mill rather than buy from a supplier raising prices | Earnings rose about 140% in 1971; it later became North America's largest recycler | That every buyer can become its supplier |
| ASML, 2012 to 2016 | Bought Cymer and took a 24.9% stake in Zeiss SMT for €1 billion, instead of arm's-length supply | EUV-made chips reached smartphones by 2018 | That the stakes, rather than the partnership, made EUV work |
| Boeing, 2005 | Sold its Wichita fuselage operations to become a lean integrator | Agreed in 2024 to reacquire Spirit AeroSystems at about $8.3 billion | What owning it throughout would have cost |
| Ericsson, 2000 | Single-sourced key radio chips from Philips | A fire at the Philips plant left no qualified backup; about 7 million phones short; handsets folded into Sony Ericsson in 2001 | That a second source alone would have saved the handset business |
| Shopify, 2022 | Paid $2.1 billion for Deliverr to build logistics in-house | Sold the logistics business to Flexport within a year for a 13% stake and cut about 20% of its workforce | That the capability could not have been built more slowly |
| Peloton, 2022 | Exited owned manufacturing for a contract manufacturer | Its Taiwan plant stayed idle into 2023 and a sale of Precor fell through | That owning factories was the cause of its losses |
| Spotify, 2019 | Bought Gimlet and Anchor for about €300 million to own podcast production | Its CEO called podcasting a drag on gross margin in 2023; about 200 podcast staff were laid off | That owning content was wrong rather than mispriced |
Running the decision in a month
A sequence for an operating or sourcing team
Assemble the specification package
Everything you would hand a new supplier: drawings, tolerances, acceptance tests, service levels, interfaces.
Send it to a supplier who has never seen the work
Ask for a quote. The clarifying questions they send back matter more than the price: each one is something you cannot yet specify.
Read the answer
If they can quote, you are buying a part and the decision is about cost. If they cannot, you are delegating a design, and buying is not a real option until the specification exists.
Run both paths over ten years
Owning (capital, overlap, running cost) against buying (price, plus the expected cost and delay of a forced buy-back).
State the break-even probability
The chance of a buy-back at which the paths cost the same. Then ask the room whether the real chance is above or below it.
Check the supplier's margin against the prize
The margin you would capture by owning can be well below the supplier's headline margin once the overlap and running costs are counted. In the pack's worked case the supplier's margin is 18% and the honest prize is 12 points.
What this method cannot tell you
It does not tell you how much of a supply chain to own. It argues that this question does not separate good outcomes from bad ones, and replaces it with a narrower one that has a testable answer: what you must be able to specify.
No layer transfers. Where the value and the risk sit is specific to one technology at one moment, and every case above might be answered differently a decade later. The cases are selected on scale and on failure, because a company that specified well and bought competently for thirty years leaves almost no record.
Questions
What is a make-or-buy decision?
A make-or-buy decision chooses whether to produce a component, product or service in-house or buy it from an outside supplier. It covers manufactured parts, software platforms, logistics, content and engineering teams alike.
How do you decide whether to make or buy?
First run the specification test: could a supplier who has never seen the work quote from what you hold today? If not, you cannot safely buy it yet. If so, compare both paths over ten years, including the chance of having to buy the capability back later.
What is the difference between outsourcing a part and delegating a design?
A part can be specified, so any qualified supplier can make it and you can switch. A delegated design cannot be specified from what you hold, so only the firm that built it can supply, re-price or repair it. The first is buying; the second is dependence.
When does insourcing make sense?
When the work cannot be specified well enough for anyone else to supply it, or when the probability of being forced to buy it back is above the break-even probability. Nucor built its own mill in 1969 when its supplier raised prices; Apple bought chip-design talent in 2008.
Does this apply to services, not just manufacturing?
Yes, wherever a capability is bought from someone else: a platform, a content library, a delivery network or an engineering team. Amazon's delivery network, Shopify's logistics and Spotify's podcasts are all service cases.
Continue learning
Sources & further reading
- Wikipedia (2024). P.A. Semi.Acquired by Apple for $278 million in April 2008.
- Wikipedia (2024). Apple silicon.
- Supply Chain 24/7 (citing the Pitney Bowes Parcel Shipping Index) (2024). Amazon Logistics' Growth Shakes Up Shipping Industry in 2023.
- Amazon Science (2023). Maximizing the efficiency of Amazon's own delivery networks.
- Nucor Corporation (2024). Nucor history.
- Encyclopedia.com (2024). Nucor Corp.
- ASML Holding NV (2012). ASML to acquire Cymer to accelerate development of EUV technology (Form 6-K).
- ASML Holding NV (2016). ZEISS and ASML Strengthen Partnership for Next Generation of EUV Lithography.
- The Boeing Company (2005). Boeing to Sell Commercial Airplanes Operations in Kansas and Oklahoma to Onex (Form 8-K).
- Spirit AeroSystems (2024). Spirit AeroSystems Announces Acquisition by Boeing in $8.3 Billion Transaction (Form 8-K).
- Ericsson (2024). The problem with mobile phones (company history).
- Ericssoners (company history) (2012). Problems with Mobile Phones.
- Shopify Inc. (2023). Shopify Announces First-Quarter 2023 Financial Results; Agrees to Sell Shopify Logistics to Flexport.
- CNBC (2023). Shopify offloads logistics business to Flexport.
- Peloton Interactive (2022). Peloton Exits Owned-Manufacturing and Expands Partnership With Rexon (Form 8-K).
- Supply Chain Dive (2023). Peloton struggles to offload manufacturing assets.
- Spotify Technology S.A. (2019). Form 6-K, Q1 2019 investments update.
- Digiday (2023). Spotify cancels six true crime podcasts amid layoffs, Gimlet-Parcast merger.
More to learn
Other foundational guides worth reading.
When to Discontinue a Product: How to Price a Sunset Before You Announce It
Every room knows what a product loses a year. Almost none knows what its customers spend with the company on everything else, or how many would leave with it. That second number decides whether retiring it saves money or costs it.
Growth & PortfolioCompany Reorganization: What It Costs and How Long It Must Last
A reorganization costs a year: the one-time charge plus the drag of months in which management attends to the company instead of its customers. It pays only if the new structure changes enough each year, for enough years. Name the line you are redrawing, put a number on what it will change, and compare the years it must stand with how long the last structure stood.
Pricing StrategyProduct Bundling Strategy: When to Sell Together or Apart
A bundle first hands its discount back to every customer already paying for both products, and earns only on the customers it adds and the extra years bundled customers stay. Match the two customer lists to size the handback, measure the retention gap, and solve for the deepest discount at which selling together still beats selling apart.
Pricing StrategyPricing Model Change: How to Move to Subscription or Usage Pricing
A pricing model change moves up to four layers at once: the unit you charge for, the shape of the cash, the buyer who signs, and the promise the customer relies on. Name which ones you are moving, size the trough against carrying on unchanged, and run the case again with the volume uplift set to zero.
Crisis ManagementCorporate Crisis Response: What One More Quarter of Delay Costs
A corporate crisis is priced on the gap between when somebody inside knew and when the company acted, not on the defect itself. Find the earliest internal date, cost what one more quarter of delay adds, and act before the expected penalty passes the cost of the fix.
M&ASerial Acquisition Strategy: How to Price the Growth You Buy
A serial acquirer's engine is the gap between what the market pays for its earnings and what it pays for somebody else's. Split the plan's growth into the part you will grow and the part you must buy, price each deal as a multiple of the earnings it brings, compare that with buying back your own shares, and state the year the balance sheet stops funding it.
See this framework in the wild.
Follow it across the companies and lenses where it actually shaped the strategy.