Business Models Wiki
Platform vs. Pipeline
A pipeline creates and delivers value through a chain it controls; a platform supplies the rules, infrastructure, and trust that let third parties transact with each other.
Snapshot
What it is
A choice of business architecture. A pipeline acquires inputs, transforms them, and sells a controlled output — the firm earns the selling price and bears the delivery cost. A platform supplies rules, infrastructure, and trust mechanisms so that distinct participant groups can find one another and transact — the operator earns a fee for enabling the interaction.
Why it matters
The two architectures have different roadmaps, different go-to-market problems, different revenue presentation, different defensibility claims, and different regulatory perimeters. Choosing the label before doing the analysis is how founders end up building platform infrastructure for a pipeline business, or promising network effects they have no mechanism to produce.
What it is not
It is not a question of whether you have an app, an API, an integrations directory, or recurring revenue. A software company with 200 integrations can still be a pipeline. A "marketplace" that employs every provider and dictates every service term is economically a managed pipeline.
Key takeaways
The test is control, not interface. Who sets price, scope, quality, and fulfilment terms, and who bears failure risk?
Neither model is superior. Pipelines start with tighter quality control and legible unit economics; platforms can expand variety through third-party investment but must solve liquidity, governance, and multi-sided pricing.
A 20% take rate is not a 20% margin. Never compare pipeline gross revenue with platform net fees without normalising.
Hybrids are normal and must be declared. State which role the company performs in each transaction, and allocate costs accordingly.
This page is the architecture decision. If you have already decided you are a marketplace, the operating mechanics live in Marketplace Model.
On this page10 sections
What is the difference between a platform and a pipeline?#
In a pipeline, value moves through a sequence the firm controls: acquire inputs, produce or configure the offering, distribute it, support the buyer. Suppliers may contribute components, but the firm decides what is sold, at what price, to whom, and on what terms — and it normally recognises the full customer price as revenue.
In a platform, the operator connects two or more participant groups and provides the core services that make their interaction possible. The 2023 Merger Guidelines describe multi-sided platforms as providing different products or services to two or more groups who may benefit from each other's participation, with the operator controlling access and influencing how interactions occur. (DOJ/FTC, 2023 Merger Guidelines, Guideline 9)
Andrei Hagiu's merchant-versus-platform framework gives the sharpest test: a merchant buys from sellers and resells to buyers; a two-sided platform enables affiliated sellers to sell directly to affiliated buyers. Hagiu's model finds the merchant mode more profitable when the chicken-and-egg problem is severe and when seller products are strongly complementary or substitutable, and the platform mode preferable when seller investment incentives matter or when sellers hold private information about their own quality. (Hagiu, Merchant or Two-Sided Platform?, 2007)
Note what that framework does not turn on: the interface through which an order is placed. The economically important variable is control over key transaction terms and delivery.
The control map#
For each customer outcome, answer six questions. The distribution of the answers tells you which architecture you actually operate.
| Question | Pipeline answer | Platform answer |
|---|---|---|
Who creates the underlying good, service, or content? | The firm or its contract suppliers | Independent third parties |
Who sets price, scope, and quality? | The firm | The participant, within platform rules |
Who owns inventory or capacity? | The firm | The participant |
Who contracts with the end customer? | The firm | Usually the participant; the operator contracts for access |
Who bears refund, failure, and regulatory risk? | The firm | Allocated by policy; often shared |
Can participants transact directly through the system? | No — the firm is always in the middle as principal | Yes — that is the point |
If most answers fall in the left column, you are running a pipeline even if suppliers contribute inputs. If third parties retain meaningful control and transact under shared rules, the model is platform-like. Hagiu and Wright treat control over price, quality, and delivery as a spectrum, so real businesses legitimately sit in between — see the boundary test in Marketplace Model.
Why does the architecture choice matter to founders?#
It changes the roadmap. A pipeline optimises a product and an operating chain. A platform must additionally build participant onboarding, discovery, matching, identity, payments, reputation, dispute resolution, policy enforcement, and separate tooling for each side. That is a materially larger surface before the first transaction clears.
It changes the go-to-market problem. A pipeline can sell to one buyer at a time. A platform must recruit compatible supply and demand in the same market cell and time window. National sign-up totals do not solve local or category-specific liquidity.
It changes revenue presentation. A pipeline that controls the promised good or service reports the full customer price. A platform acting as an agent reports only its fee. Uber states that substantially all of its revenue comes from fees paid by drivers and merchants, reported net of provider earnings, because it concludes it acts as an agent for those transactions. (Uber 2025 Form 10-K) The determination is fact-specific and depends on control, not on which number looks larger.
It changes the defensibility claim. Third-party participation creates a network effect only when it improves value for other participants. More suppliers can improve choice and wait time — or worsen search quality and trust. Scale is evidence, not proof.
It changes the risk perimeter. Platforms inherit governance, fraud, safety, data, worker-classification, self-preferencing, and competition risk. Airbnb's filing describes the payments, fraud-prevention, support, regulatory, and technology systems required to operate a two-sided marketplace — infrastructure a pipeline of comparable revenue would not need. (Airbnb 2025 Form 10-K)
Key Facts
At scale, platform revenue is a small fraction of the value it intermediates
Uber reported $193bn of Gross Bookings and $52.0bn of revenue for 2025 — revenue is roughly 27% of transaction value, and most of it is presented net because Uber concludes it acts as an agent. (Uber Q4 & FY2025 results, 4 Feb 2026; )
Uber 2025 Form 10-KThe ratio differs sharply by platform
Airbnb reported $91.3bn of Gross Booking Value and $12.2bn of revenue for 2025 on 533 million nights and seats booked (+8%) — revenue of about 13% of GBV, roughly half Uber's ratio. Cross-platform "take rate" comparisons are usually invalid without normalising what each fee covers.
Airbnb 2025 Form 10-KThe merchant/platform choice has a formal economic answer
Hagiu shows the merchant (pipeline) mode dominates when the chicken-and-egg problem is more severe and when seller products are more complementary or substitutable, while the platform mode dominates when seller investment incentives matter or when sellers hold private information about quality.
Hagiu, *Merchant or Two-Sided Platform?*, 2007Gatekeeper regulation is now a designated, enumerated list — not a theoretical risk
The European Commission maintains a public register of designated gatekeepers and their core platform services under the Digital Markets Act, with obligations covering interoperability, data access, and business-user treatment. Check the register before assuming your category is unregulated. (, accessed 13 Aug 2026)
European Commission, DMA gatekeepers portalHow do you compare the two economically?#
The most common analytical error is comparing a pipeline's revenue with a platform's revenue. They are denominated differently. Normalise on the same underlying transaction, then compare contribution in currency and contribution as a percentage of both revenue and transaction value.
For a pipeline:
contribution per unit = realised price − direct input − fulfilment − variable support − transaction cost
For a platform:
platform revenue per interaction = commission + participant fees + ancillary revenue
platform contribution = platform revenue − payments − fraud and loss − support − incentives − variable trust and operations cost
Operating measures diverge too#
| Pipeline measures | Platform measures |
|---|---|
Yield, quality-adjusted unit cost | Serviceable supply, qualified demand |
Inventory turns, delivery time | Match rate, time to match |
Gross margin, repeat purchase | Completed interactions, repeat rate by side |
Capacity utilisation | Take rate, subsidy per completed interaction |
Defect and return rate | Dispute and loss rate, contribution by market cell |
Worked example: the same $100 order, two architectures#
A company can sell a $100 item as a reseller (pipeline) or enable a third-party seller to sell it (platform) at a 20% take rate. Payment processing is 2.9% + $0.30 on the amount the buyer pays.
Pipeline (principal). Revenue is the full $100.
$100 − $62 cost of goods − $9 fulfilment − $3.20 payments − $2.00 variable support = $23.80
Contribution is $23.80, or 23.8% of revenue and 23.8% of transaction value.
Platform (agent). The buyer pays $100, the seller keeps $80, and revenue is the $20 fee.
$20.00 − $3.20 payments − $0.60 fraud and chargeback reserve − $2.00 trust, support, and disputes − $3.00 participant incentives = $11.20
Contribution is $11.20, or 56% of revenue but only 11.2% of transaction value.
Read the two side by side:
| Pipeline | Platform | |
|---|---|---|
Reported revenue per order | $100.00 | $20.00 |
Contribution per order | $23.80 | $11.20 |
Contribution % of revenue | 23.8% | 56.0% |
Contribution % of transaction value | 23.8% | 11.2% |
The platform looks more than twice as profitable on a margin basis and roughly half as profitable per order. To match the pipeline's contribution the platform needs $23.80 / $11.20 = 2.13 orders. Whether that is easy or impossible is the actual strategic question — and it depends on whether third-party supply lets you serve variety and volume the pipeline could never carry.
Caveats this example hides: the payments cost is charged on gross in both cases even though the platform only recognises the fee; incentive spend is treated as variable when it is often a cold-start subsidy that should decay; and neither column carries fixed cost, which is usually much heavier for the platform in its first years. Run the comparison at your expected steady-state volume, not at unit one.
What are the common mistakes?#
- Calling software a platform because it has integrations. A platform requires meaningful third-party participation in the transaction, not an API catalogue.
- Assuming asset-light means operationally light. Trust, moderation, payments, disputes, incentives, and local liquidity are expensive, and most of that cost is invisible on a pitch-deck architecture diagram.
- Counting participants instead of completed interactions. Registered users establish neither network effects nor liquidity.
- Comparing gross pipeline revenue with net platform fees. Normalise transaction value, revenue presentation, and direct costs first — as the worked example shows, the ranking flips depending on which ratio you pick.
- Treating the classification as permanent. Taking inventory or employing providers moves you toward a pipeline; transferring control to third parties moves you toward a platform. Reclassify when the transaction changes, not when the marketing changes.
When does each model break?#
The platform model weakens when supply is scarce, highly standardised, or safety-critical; when customers demand a single accountable provider; when coordination is cheaper inside one firm than across a market; or when third parties cannot earn enough to invest in quality. It also weakens when participants multi-home freely and can transact off-platform after the first introduction.
The pipeline model weakens when customer needs are highly varied, when outside innovators can add value faster than you can build, or when owned capacity becomes the growth bottleneck.
Both are reshaped by concentration rules. Network effects push toward concentration, and concentration attracts regulation. The EU's Digital Markets Act imposes ex-ante obligations — interoperability, data access, and business-user treatment among them — on designated gatekeepers and their core platform services; the Commission maintains the current designation list publicly. (European Commission, DMA gatekeepers portal) A founder should not treat future gatekeeper power as a free option. Governance choices made early — self-preferencing, exclusivity, data use, ranking control — create later conflicts between the operator and the participants it depends on.
A deliberate hybrid is often the right answer. Many companies begin as a managed pipeline to learn the workflow, then open selected layers to third parties. Others run a marketplace and supply first-party inventory where liquidity is weak. Hybrids improve reliability and create channel conflict at the same time, and they make cost allocation genuinely hard. The discipline is to state which role the company performs in each transaction and to report the two economics separately.
Frequently asked questions
01Is every marketplace a platform?
Yes, but not every platform is a marketplace. A marketplace is a multi-sided platform organised around a transaction between buyers and sellers. Operating systems, app stores, and ad networks are platforms with different atomic interactions. The mechanics of transaction platforms are covered in Marketplace Model.
02Can I start as a pipeline and become a platform later?
Frequently, and it is often the lower-risk sequence — you learn the workflow, the quality bar, and the failure modes with control before delegating them. The hard part is the transition: opening a layer to third parties means giving up terms you previously set, and the first cohort of participants will test exactly the rules you have not written yet.
03Does reporting revenue net make my business look worse?
It makes it look smaller, not worse. Gross-versus-net presentation turns on whether you control the specified good or service before transfer, not on presentation preference. Report transaction value and revenue as separate line items so readers can compute both.
04How do I know if I really have a network effect?
State the causal mechanism, name the metric it should move, and check it in cohort data: do denser market cells show higher fill rates, faster matching, higher repeat, and lower incentive spend per interaction than sparser cells of the same age? If growth is uniform across densities, you have distribution, not a network effect.
05Should a platform ever supply its own inventory?
Sometimes — it is a standard cold-start tool and a reliability backstop. But it puts the operator in competition with the participants it also ranks, which is precisely the self-preferencing conflict the 2023 Merger Guidelines flag for platform operators that are also participants. If you do it, ring-fence the ranking logic and disclose the arrangement.
Related concepts#
- Marketplace Model — liquidity, take rate, trust, and transaction mechanics once you have chosen the platform architecture.
- Two-Sided Markets — cross-side demand and the allocation of price between sides.
- Network Effects — testing whether participation causally increases user value.
- Switching Costs — why participants stay after they have met.
- Managed Services — the controlled-delivery end of the control spectrum.
- Advertising Model — a platform architecture where one side is subsidised entirely.
Note: This page is educational and does not constitute legal, tax, accounting, or financial advice. Principal-versus-agent revenue presentation, platform liability, worker classification, and competition-law obligations are fact-specific and vary by jurisdiction. Consult qualified counsel and accountants before making structural decisions.
Sources#
- U.S. Department of Justice and Federal Trade Commission, 2023 Merger Guidelines, Guideline 9: mergers involving multi-sided platforms, December 2023. Source for the agencies' definition of a multi-sided platform and for the self-preferencing conflict when an operator is also a participant.
- Andrei Hagiu, Merchant or Two-Sided Platform?, Harvard NOM Working Paper (2006), published in Review of Network Economics 6, no. 2 (2007). Source for the merchant-versus-platform distinction and the conditions favouring each mode.
- Uber Technologies, Inc., 2025 Form 10-K, filed February 2026. Source for the statement that substantially all revenue is fees reported net of provider earnings on an agent basis.
- Uber Technologies, Inc., Uber Announces Results for Fourth Quarter and Full Year 2025, 4 February 2026. Source for FY2025 Gross Bookings of $193bn and revenue of $52.0bn.
- Airbnb, Inc., 2025 Form 10-K, filed February 2026. Source for the payments, trust, support, and regulatory infrastructure required to operate a two-sided marketplace.
- European Commission, Digital Markets Act — gatekeepers portal, accessed 13 August 2026. Source for the current list of designated gatekeepers and core platform services, and the obligations attaching to them.
Author
Dr. Sarah Zou
Independent economist · EconNova
Commercial strategy for technical products, with a focus on pricing, unit economics, and the operating choices behind the model.
About SarahTopics
Cite this page
Suggested citation
Zou, S. (2026). Platform vs. Pipeline: Choosing the Right Business Architecture. In Business Models. Pricing & Monetization Wiki. https://sarahzou.com/wiki/business-models/platform-business-model
Open license
Reuse with attribution
This content is available for reuse. When referencing or republishing it, please credit Dr. Sarah Zou and link back to the original source.
Licensed under Creative Commons Attribution 4.0 International. You may share and adapt the material with appropriate credit.