A pipeline business creates value by making something and passing it down a chain to a customer. A platform business creates value by enabling producers and consumers to find and transact with each other, so the company owns the matching, not the goods. Pipelines scale by adding resources. Platforms scale by adding participants, which is why they are harder to start and harder to govern.
Most leaders can recite that distinction. Fewer can say what it does to the way a company is run. A platform is not a pipeline with an app on top. It needs different decision rights, different metrics, and a different relationship with the people outside your payroll who create most of the value.
This post explains both models, shows where each wins, and uses Amazon to show why the answer is often “both.” It then covers what the operating model must change, which is where most platform ambitions quietly die. That part connects to the Leader Flow framework: principles first, then boundaries, then action.
How does a pipeline business create value?
A pipeline business controls a linear sequence of steps, from inputs to production to distribution to a customer, and earns the margin between what the sequence costs and what the customer pays.
Picture a bakery chain. It buys flour, bakes bread, ships it to stores, and sells it. Every step is owned or contracted by the company. Value moves in one direction, and the company’s job is to make each step cheaper, faster, or better than a competitor’s.
This is the model most management thinking was built for. Strategy, org charts, and budgets all assume a chain. Scale comes from internal resources: more factories, more engineers, more sales reps, more capital. If you want to double output, you roughly double inputs, with efficiencies along the way.
Three properties define a pipeline:
- The company produces the value. Employees and contractors make the product or service.
- The customer is the end of the line. Value flows one way.
- Advantage comes from efficiency and differentiation. Cost, quality, brand, and distribution are the levers.
Pipelines are not outdated. They are the right model whenever you can make something better than the market can, or when quality control matters more than variety.
How does a platform business create value?
A platform creates value by connecting two or more groups who need each other and by providing the infrastructure, rules, and trust that let them transact. The economists Andrei Hagiu and Julian Wright define a multi-sided platform as an organization that enables direct interactions between two or more distinct sides, each of which is affiliated with the platform (Hagiu & Wright, International Journal of Industrial Organization, 2015).
The key phrase is “direct interactions.” A rideshare driver and a rider deal with each other. A seller and a buyer on a marketplace deal with each other. A developer and a user on an app store deal with each other. The platform runs the stage; it does not write most of the play.
Three properties define a platform:
- Participants produce most of the value. Drivers, sellers, developers, creators, and hosts supply what the other side wants.
- Value flows in several directions. Consumers also create value for producers, through demand, reviews, and data.
- Advantage comes from network effects. Each new participant makes the platform more valuable to the others.
The 2016 book Platform Revolution (Parker, Van Alstyne, and Choudary) and the same authors’ Harvard Business Review article “Pipelines, Platforms, and the New Rules of Strategy” (April 2016) popularized this pipeline-versus-platform framing. They argue that platforms win when they can reduce the friction of matching and when outside producers can supply variety the company could never make alone.
Why do network effects change how a business scales?
Network effects mean each additional participant raises the value for existing participants, so growth compounds instead of adding linearly. There are two kinds, and the difference matters.
Direct network effects occur when users benefit from other users of the same type. A messaging app is worthless if nobody you know uses it (Katz & Shapiro, 1985, is the classic academic treatment).
Indirect (cross-side) network effects occur when one side benefits from the size of the other. More riders attract more drivers, and more drivers shorten rider waits, which attracts more riders. This loop is what most platforms mean by “flywheel.”
A pipeline cannot do this. Your hundredth customer for a loaf of bread does not make the bread better for your ninety-ninth. But the same compounding force that makes platforms powerful also makes them fragile in one specific way: it works in both directions. If sellers leave, buyers get less choice and also leave, and sellers leave faster. A pipeline loses revenue when it loses customers. A platform can lose its reason to exist.
What is the cold-start problem, and why do most platforms fail at it?
The cold-start problem is that a platform is useless until both sides are present, but neither side joins an empty platform. It is the chicken-and-egg problem in business form, and it is the main reason most platform attempts never reach scale. Andrew Chen’s book The Cold Start Problem (2021) is the most practical treatment of how teams get past it.
Common ways out include:
- Start with one side and a single-player utility. Offer a tool that is useful alone, then add the network later.
- Subsidize the scarcer side. Pay drivers, waive seller fees, or seed supply by hand.
- Start in a small, dense market. A platform that works in one city or one niche can be copied elsewhere.
- Seed the supply yourself. Run it like a pipeline at first, then open it up.
Note the last one. Many successful platforms began as pipelines. That is not a contradiction. It is a sequence, and it is the reason the question “platform or pipeline?” is often really “which, and when?”
Competition adds a second problem: multi-homing, where participants use several platforms at once. Rochet and Tirole’s 2003 paper on platform competition in two-sided markets shows why this matters: when drivers can run two ride apps at the same time, no single app can lock them in. Platforms with high multi-homing compete on price and service. Platforms with low multi-homing can build a durable position.
Platform vs. pipeline: how do the two models compare?
The two models differ in who creates value, how they scale, what they measure, and what can go wrong. The table below is the short version.
| Dimension | Pipeline | Platform |
|---|---|---|
| Best for | Products you can make better than the market | Markets where matching producers and consumers is hard |
| Who creates value | The company | Outside participants, enabled by the company |
| How it scales | Add internal resources | Add participants (network effects) |
| Core advantage | Efficiency, quality, brand | Network size, trust, data on matches |
| Key metrics | Margin, throughput, cost per unit | Liquidity, match rate, repeat participation, take rate |
| Main risk | Being out-competed on cost or quality | Cold start, disintermediation, governance failures |
| Governance question | “How do we run our own steps?” | “What may outsiders do, and who decides?” |
The last row is the one most companies underweight. A pipeline’s rules are internal. A platform’s rules are a product: they shape who joins, what they may do, and whether they trust the system.
Can one company run both? What does Amazon show?
Yes. Amazon is the clearest example of a company that runs a pipeline (selling its own goods) and a platform (a marketplace for third-party sellers) side by side, and the combination is deliberate. Amazon’s own shareholder letters report that third-party sellers account for more than half of the physical goods sold on its stores, a share that has been in the 50-to-60 percent range in recent years according to those letters. Amazon sells its own inventory as a retailer, and it also hosts millions of independent sellers.
Hagiu and Wright describe this as a reseller-versus-marketplace continuum: a company decides, category by category, whether to buy goods and resell them (pipeline) or let others sell directly (platform). Their research (“Marketplace or Reseller?”, Management Science, 2015) suggests the right choice depends on who has better information about demand and who is better placed to control quality and pricing.
Amazon also shows something less obvious: a platform needs an internal layer that is built like a product. Amazon is widely reported to have required its teams, in the early 2000s, to expose their capabilities through service interfaces rather than direct access to each other’s data. This account circulates mainly through former engineers’ recollections, and I would treat it as well-known but informally documented. The lesson holds regardless: the internal boundaries came first, and the autonomy came after.
That ordering is the whole argument of the next section.
What does the operating model have to change for a platform?
A platform changes three things in the operating model: who may make decisions, what you document, and what you measure. Companies that treat platform as a go-to-market choice and skip these changes usually discover the problem when something breaks in public.
If you want the vocabulary for operating models first, start with the complete guide to operating models and the post on operating model vs. business model vs. strategy. The short version: the business model is how you make money, the operating model is how the company actually runs to deliver it. Pipeline-to-platform is a business-model move. It only works if the operating model moves with it.
1. Decision rights now extend outside the company
In a pipeline, “who can change the price?” is an internal question. In a platform, you have to answer it for outsiders too. Who sets prices: the seller or the platform? Who handles a dispute? Who can remove a listing, and on what evidence? Who can change the rules, and how much notice must they give?
These are decision rights, the subject of Atlas Expectations. The test I use is simple: an undocumented boundary is a centralized boundary. If the rule is not written down, every edge case escalates to a person inside the company, and the platform cannot scale past that person’s calendar. Worse, participants cannot predict the system, so they stop trusting it.
2. Governance becomes a product
Platform governance covers access (who can join), conduct (what is allowed), and enforcement (what happens when rules are broken). Done well, it is the thing that makes strangers willing to transact. Done badly, it is the thing that makes your best participants leave.
Every governance rule is a decision-rights question with a customer on the other side. That makes it a candidate for the same discipline you would apply to a product: a named owner, a written standard, a review cadence, and a way to measure whether it works. This is Atlas territory, specifically standards, workflows, and service levels.
3. The metrics change
Pipelines measure throughput and margin. Platforms need to measure liquidity (do buyers find what they want and sellers find buyers?), match quality, time to first transaction for new participants, and repeat participation. If you only watch revenue, you will notice a failing marketplace after it has already failed.
The lesson from the KPI definitions in Atlas applies here: every metric needs one definition and one owner. “Active seller” means three different things in three different dashboards at most companies, and a platform cannot steer by that.
How should you decide between a platform and a pipeline?
Choose by asking three questions: can outside producers supply value you cannot, is matching the hard part, and can you govern what you do not control? If the answer to all three is yes, a platform is worth exploring. If any answer is no, a pipeline, or a hybrid, is the safer bet.
- Can outsiders supply more variety, speed, or specialization than you can? If you are the best maker of the thing, a platform adds risk with little gain.
- Is the core customer problem finding or trusting a counterpart? Platforms solve search and trust problems. If the problem is production cost, they do not help.
- Can you govern participants you do not employ? This is the question that gets skipped.
Then check the model against your actual ability to start. Can you solve the cold start through a single-player product, a dense niche, or seeding supply yourself? If not, begin as a pipeline and open the system later.
A hypothetical example
The following is an illustrative scenario, not a real company.
A regional home-services company employs 200 technicians and sells plumbing and electrical work directly. It is a clean pipeline: it controls quality, but growth is limited by hiring and scheduling. Leadership considers launching a marketplace that lets independent contractors take jobs.
Applying the three questions: outside contractors could add capacity in neighborhoods the company does not serve (yes to question one). Customers do worry about finding a trustworthy contractor (yes to question two). But the company has no written standards for contractor quality, no process for disputes, and no owner for pricing rules (no to question three).
The sensible move is a staged hybrid: keep the employed technicians as the quality benchmark, invite a small group of vetted contractors into one city, and write the governance rules before adding volume. The company keeps its pipeline advantage while testing whether it can run a platform. Expanding only happens when match rate and dispute rate hold up.
Where does Leader Flow fit when moving from pipeline to platform?
Leader Flow gives you the order to make the change in: principles, then boundaries, then action. The framework has three layers, and each one answers a different platform question.
- Compass is where you decide what you believe. What is the platform for? Whom does it serve first when sides conflict? Platforms constantly face trade-offs between buyer and seller interests, and without a stated principle each conflict becomes a political fight.
- Atlas is where the boundaries live: decision rights, KPI definitions, service levels, workflows, and standards. For a platform, this is the governance rulebook, written once and updated as a running record.
- Arcade is where the day-to-day happens: the huddles, the ops review, the experiments on supply and demand, and the roadmap. This is where you run the flywheel.
The full structure is on the Leader Flow page. The most common failure I see is starting in Arcade: launching features, recruiting sellers, and running promotions before the Atlas layer exists. It works until the first dispute nobody has authority to settle.
| Platform question | Layer | What to write down |
|---|---|---|
| Whom do we favor when buyers and sellers conflict? | Compass | A stated principle and order of priority |
| Who decides prices, listings, and removals? | Atlas | Decision rights by class, one owner per line |
| What does “active seller” mean? | Atlas | A single KPI definition and owner |
| How fast must a dispute be resolved? | Atlas | A service level and escalation path |
| How do we test a new supply category? | Arcade | An experiment with a reversible scope |
| How do we know the flywheel is turning? | Arcade | Weekly liquidity and match-rate review |
If you are weighing which structure suits a hybrid, functional vs. product operating models and centralized vs. decentralized operating models cover the structural choices, and the types of operating models lays out the wider menu. Platforms tend to need a centralized core (rules, data, standards) with decentralized edges (participants acting inside those rules). That is the same pattern the constraints create better leaders idea points to: clear limits are what let people move fast.
Frequently Asked Questions
Is a SaaS company a platform or a pipeline? Usually a pipeline: it builds a product and sells it to customers. It becomes a platform when third parties build and sell on top of it, as with an app marketplace or an API ecosystem that other businesses depend on.
Do platforms always beat pipelines? No. Platforms win in markets where matching and trust are the bottleneck. Where product quality or cost control is the bottleneck, a pipeline can outperform them, and many platforms fail before reaching critical mass.
What is a product platform, and is it the same thing? Not quite. A product platform is a shared set of components, such as a common engine or codebase, that internal teams build on. It is an internal efficiency tool. A multi-sided platform connects external groups. Companies often need the former to enable the latter.
Can a pipeline company add a marketplace later? Yes, and it is a common path, but it requires new decision rights, new metrics, and new governance. Treat it as an operating-model change, not a feature launch.
TL;DR
- A pipeline makes and sells; a platform enables others to make and sell, and wins through network effects.
- Platforms are harder to start (cold start) and harder to run (governance), and they can fail in reverse when one side leaves.
- Many companies, Amazon included, run both, deciding category by category.
- Moving to a platform changes decision rights, governance, and metrics, so it is an operating-model move, not just a business-model one.
- Use Leader Flow’s order: Compass for principles, Atlas for boundaries, Arcade for action.
Take the Next Step
If you are considering a platform move, start with one page: list every decision outsiders will touch, name one owner for each, and write down the rule. That page is your first Atlas. If you want to talk it through, get in touch.
About the author: Milen Vasilev builds operating models: the unglamorous machinery that makes ambitious things actually ship. He created Leader Flow, a framework that moves from principles (Compass) to boundaries (Atlas) to action (Arcade). Learn more about Milen.
Sources
- Hagiu, A. & Wright, J. (2015). “Multi-sided platforms.” International Journal of Industrial Organization, 43.
- Hagiu, A. & Wright, J. (2015). “Marketplace or Reseller?” Management Science, 61(1).
- Parker, G., Van Alstyne, M. & Choudary, S. (2016). Platform Revolution. W. W. Norton.
- Van Alstyne, M., Parker, G. & Choudary, S. (2016). “Pipelines, Platforms, and the New Rules of Strategy.” Harvard Business Review, April 2016.
- Rochet, J.-C. & Tirole, J. (2003). “Platform Competition in Two-Sided Markets.” Journal of the European Economic Association, 1(4).
- Katz, M. & Shapiro, C. (1985). “Network Externalities, Competition, and Compatibility.” American Economic Review, 75(3).
- Chen, A. (2021). The Cold Start Problem. Harper Business.
- Amazon.com, Inc. annual shareholder letters (third-party seller share of physical gross merchandise sales), 2018 to 2021. Coverage: Retail Dive, Modern Retail.
