Subsidizing one platform side can be a network design decision

Consider a synthetic case. A B2B marketplace waives fees for one supplier group for a quarter. Supplier registrations rise, but repeat use by buyers does not, while support work and forgone fee revenue accumulate. At review, the team must decide whether to extend the subsidy, redesign it, or stop. The signup increase alone does not show whether buyers receive more value from the larger seller base.

Trace the test through the enterprise stack: put the authorized concession in the commercial discount-approval workflow, including the configure-price-quote system (CPQ) where it owns the quote; record the waiver or credit in the billing engine; and connect the supplier cohort to buyer-side activity in CRM or marketplace event records. Use one cohort definition and review window across those records. The system of record is company-specific. This is a proposed control design for the synthetic case, not a claim that every platform uses the same tools.

The useful question is not which side should always be free. It is: What cross-side response would make this allocation worth continuing, and what observation would make the team stop or re-price it? This article turns that question into a bounded decision card. It does not recommend a free side, set a price, or treat participation as proof of network value.

This is a different object from the marketplace governance article, which owns the platform’s roles as rule-maker, participant, and data-holder. It is also different from the channel ownership article, which owns route rights, service burden, margin, renewal, and incrementality. Pricing as positioning asks what a price communicates. This piece asks how a platform allocates price and support between interdependent sides.

What is a platform-side subsidy actually changing?

A platform-side subsidy is a deliberate price or support allocation to one participant group. It can be a lower fee, free access, a credit, development support, onboarding work, or another transfer. The label is useful only when the receiving side, the support unit, and the other side’s expected response are explicit.

Three objects must stay separate:

1. Side. Who receives the lower price or support, and what role does that group play?
2. Cross-side response. What should the other side do, buy, provide, or improve if the allocation is working?
3. Subsidy burden. What margin, capacity, service, moderation, risk, or opportunity cost does the platform carry while the test runs?

This is narrower than a buyer discount and broader than a promotion. A buyer discount may be a transaction-level concession. A penetration price may be a market-entry choice. A channel rebate may change partner economics. A platform-side subsidy is the part of the price structure that is expected to alter participation or value across connected sides.

Rochet and Tirole² write that platforms “must choose a price structure and not only a price level” (2003, p. 990). For a commercial team, because each side can affect the other side’s surplus, that implies documenting side allocation rather than only a single price point. This distinction does not tell a specific platform which side to subsidize. It tells the team to record the allocation and cross-side mechanism before treating the allocation as a network design decision.

Why should the lower price belong on one side rather than the other?

Parker and Van Alstyne¹ model cross-market network effects in which participation in one market can affect the value available in another. Their result makes the subsidy question conditional: the side receiving a discount depends on cross-price elasticities and on the relative sizes of the two-sided network effects in the model (Parker and Van Alstyne, 2005).

That is a mechanism, not a universal rule. It does not imply that the side with the most users should be free, that the side with the highest willingness to pay should pay, or that a current platform can copy an illustrated arrangement. The relevant comparison is the expected cross-side response relative to the burden of the allocation.

Rochet and Tirole² use business-model illustrations in which one side can be a loss leader or financially neutral while the other side supplies revenue (Rochet and Tirole, 2003). Those examples help explain price structure. They are not current benchmarks, target margins, or evidence that a named platform’s arrangement succeeded.

Before choosing the subsidised side, write the comparison in one sentence: “If side A receives support under condition B, side C should produce response D within horizon E, while burden F stays inside boundary G.” If the response or boundary cannot be written, the platform has a price intuition, not a reviewable subsidy test.

What does the literature support, and what does it leave open?

The source packet supports a bounded synthesis rather than a current recommendation.

Source	What it contributes	What it cannot establish
Parker and Van Alstyne (2005)	A model of cross-market network effects and the conditions that influence which side receives a discount.	The right subsidy for a named platform or a current price recommendation.
Rochet and Tirole (2003)	A price-structure distinction and theoretical or illustrative business-model arrangements.	A universal loss-leader design or evidence that an illustrated model succeeded.
Boudreau and Jeppesen (2015)	A bounded study of 85 online multiplayer game platforms with unpaid complementors.	A result for paid complementors, other platform types, or every current network.
Kaiser and Wright (2006)	An estimated model of nine German two-magazine groups using 1972–2003 data.	A transferable subsidy effect: the sample is small, the reader-side margin estimate is not significant, and the authors call the results illustrative.

Boudreau and Jeppesen³ report that, in their historical setting of 85 online multiplayer game platforms between 2002 and 2004, complementor development responded to platform growth, while attracting complementors had a net-zero effect on ongoing development activity and failed to stimulate network effects (Boudreau and Jeppesen, 2015). The setting matters: the complementors were unpaid crowd hobbyists writing full-conversion modifications for first-person-shooter game engines outside any price system, and the authors note that paid complementors can operate differently. The study examined development activity rather than user-perceived value or a price subsidy. By analogy, the finding serves as a caution against treating sheer participation as proof of useful network value, not as proof that commercial subsidies fail.

Kaiser and Wright⁴ provide a historically specific German case. Their data source is the IVW, Informationsgemeinschaft zur Feststellung der Verbreitung von Werbeträgern, for magazine circulation and related market measures.

The unbalanced panel narrows to nine two-magazine groups and 100 usable observations, 91 after first differencing; those groups account for only 0.4% to 5.3% of total German magazine circulation, depending on the year. Their estimated model implies that higher reader demand raises advertising rates, while higher advertising demand lowers cover prices. In their estimates, the model-implied cover price lies EUR 2.198 below marginal cost (price–cost margin −2.198) with a standard error of EUR 4.741, and the authors state this is not statistically significant. Their finding suggests the model likely overstates the two-sided pricing effect, and they treat the estimates as illustrative. This is a historical mechanism example, not evidence about what a current marketplace should charge or what a fee waiver would cause.

What can a subsidy attract without creating a useful network?

A lower price can change a participation count without proving that the connected exchange improved. The review should therefore separate at least five observations:

• Participation: Did the subsidised side join, remain active, or submit more supply?
• Interaction: Did the other side encounter, choose, transact with, or otherwise use that supply?
• Quality: Did the complement, listing, service, or interaction meet the declared standard?
• Burden: What capacity, support, moderation, risk, and margin did the allocation consume?
• Horizon: When should the expected response be visible, and what comparison makes absence interpretable?

This separation also protects against false comfort from a rising top-line count. A platform can attract participants who do not interact, interactions that do not produce useful outcomes, or supply that creates more review and service work than the allocation can carry. Multihoming, seasonality, changes in demand, and a different acquisition channel can make the observed count difficult to interpret. Those are reasons to declare the comparison and boundary, not reasons to invent a causal result.

The unpaid-complementor study supplies a specific transfer limit. It can inform a warning about participation and network-effect claims in that setting. It cannot establish what a paid complementor, a professional marketplace, or a current platform will do. The decision card keeps that boundary visible beside the operational fields.

How should a team build a subsidy decision card?

The card below is an author framework, not a validated scorecard. The foundational literature models standing equilibrium price structures. For an operational review, a team can treat a subsidy as a time-boxed test with an explicit baseline, response horizon, and exit criteria. The card below places those design choices beside the expected response, useful outcome, burden, comparison, and system-of-record owner. The rows are deliberately synthetic so a team can fill them with its own observations without mistaking the framework for platform data.

Field	Subsidised side	Other side	Evidence before continuation
Side and role	A declared participant group receives lower price or support	A declared participant group whose value or participation should respond	The roles are not only labels; the interaction is stated
Price or support baseline	Fee waiver, discount vs list/cost, or support resource is declared relative to baseline	Price and access conditions remain visible	The allocation baseline (cost vs fee) is separated from the outcome
Cross-side response	Expected response is written as a mechanism (or avoided loss vs counterfactual)	Response unit, lag threshold, and review window are declared	A comparison can identify the response, threshold lag, or its absence
Useful outcome	The side’s participation is observed separately	Interaction, transaction, or complement quality is observed	Participation is not used as a substitute for usefulness
Burden	Margin, capacity, service, moderation, or risk burden is recorded	Any burden shifted to the other side is visible	The cost boundary is stated
Governance and Decision	Continue, extend, re-price, or exit is assigned to a named owner	The other side’s evidence and system-of-record trail are included	The stop, extend, or redesign condition is dated and reviewable

The card is useful only if its fields are completed before the result is interpreted. Name the sides and roles first. Then name the price or support, the expected response unit, the horizon, the comparison, and the burden boundary. If the platform cannot observe the result or cannot assign an owner for a stop decision, the missing field is part of the decision risk.

This is where pricing architecture as a system matters. A side-specific price is not isolated from access, terms, authority, delivery burden, or review. The card is the cross-side view of that architecture.

How does the card change a review decision in a worked example?

Apply it to a constructed version of the opening case. Every count, amount, time allowance and observation below is invented for this application. These are neither platform data nor recommended subsidy levels or success thresholds. The pricing lead records the following plan before the test, including a scenario-specific continuation hurdle.

Field	Declared test	Constructed review record
Supplier cohorts	Twelve eligible supplier organizations, S01 to S12, receive the fee waiver. Twelve others, S13 to S24, meet the same stated eligibility criteria and keep the usual fee during the same window. Membership is fixed before either group registers.	Ten of twelve suppliers register in the waiver cohort; six of twelve register in the comparison cohort. This is a participation observation.
Other side and useful outcome	Ten existing buyer accounts, B01 to B10, can access both groups. Count a buyer once per group if it completes a second order from that group’s suppliers during the review window. The denominator remains these ten accounts. For this constructed decision, continue only if at least six of the ten buyers make a repeat order during the thirteen-week window. This invented operating hurdle is neither a sourced benchmark nor a recommended threshold.	Four of ten buyers complete a repeat order from the waiver cohort; four of ten do so from the comparison cohort. The cohort with more supplier registrations has the same observed repeat-order buyer count.
Comparison and horizon	Observe both groups during weeks 1 to 13. The synthetic plan allows four weeks for onboarding; it declares no further adoption lag or defensive avoided-loss objective.	The thirteen-week window is complete and the declared onboarding allowance has elapsed. There is no predeclared lag that would justify extending this test.
Fee baseline and burden	The billing record retains the usual fee that would have been charged. Authorization covers at most EUR 1,200 of waived fees and 24 support hours for the waiver cohort over the whole window. Record the two units separately.	Waived fees total EUR 900; support totals 18 hours. The comparison cohort uses 9 support hours and receives no waiver. The waiver cohort remains within both authorization limits.
Quality and record trail	Count completed orders under the same acceptance definition for both groups. Link supplier and buyer IDs, commercial authorization, waiver entries and order records to the same thirteen-week window.	The constructed record contains those joins and identifies no interruption that would make the comparison unusable. It reports no quality outcome. It does not establish random assignment or eliminate selection and cross-group spillovers.
Review owner and action	The marketplace pricing lead reviews participation, repeat orders, quality, burden and the remaining identification limits together.	The observed four-of-ten repeat-order count does not meet the example’s six-of-ten continuation hurdle. Stop offering this specific waiver for the next window to limit further exposure and opportunity cost, while honoring existing terms. This is a management response to a predeclared hurdle, not a conclusion about effectiveness.

The decision follows the predeclared hurdle, not a claim that four buyers is an inadequate market-wide result. With only ten accounts and no established random assignment, the equal observed counts are descriptive and do not show whether the waiver caused a change. This scenario stops a particular waiver to limit further risk and opportunity cost; it does not conclude that subsidies are ineffective. Registrations alone cannot justify continuation, and staying inside a spending authorization cannot substitute for the intended response. This application neither estimates an optimal price nor rules out a useful network effect under another design.

Change the facts and the action changes. If a predeclared adoption lag were still pending, a dated extension could be justified within the burden limits. If the objective were avoiding a loss, the record would need a defensible counterfactual rather than an expansion target. If a billing migration broke the cohort joins, the next action would be to pause and repair the comparison, not to declare the subsidy ineffective.

Which observations should trigger extension, repricing, or exit?

Continue only when the declared response appears within the stated horizon, the useful outcome is observable, and the subsidy burden remains inside the agreed boundary. In competitive or defensive subsidies, the response may be demonstrated as an avoided loss against a counterfactual rather than net expansion.

Extend the review window when participation has grown and an explicit, pre-declared threshold or adoption lag has not yet elapsed, provided the accumulated burden remains strictly within bounds. An extension requires a fixed, dated second horizon rather than an open-ended subsidy.

Re-price when a response appears but the allocation is not sustainable, when the burden differs from the declared boundary, when quality falls as participation rises, or when operating evidence suggests the subsidy level was miscalibrated. Re-pricing is not a confession that the mechanism was wrong. It is a change to the allocation after the system has learned what the first test could not establish.

Exit the test when participation rises without the defined cross-side response, when useful outcomes do not appear, when complement quality falls below the declared floor, or when service burden exceeds the agreed ceiling. When a major outside change confounds the comparison, pause or redesign the identification test before committing further resources, rather than assuming unidentifiable results automatically represent program failure. These are author decision rules for the synthetic card, not source-reported thresholds.

What does this framework not prove?

It does not prove that one side should be free. It does not turn the Parker and Van Alstyne¹ model into a current pricing recommendation. It does not turn a Rochet and Tirole² illustration into a benchmark. It does not transfer the Boudreau and Jeppesen³ finding beyond its unpaid-complementor setting. It does not claim that a platform’s participation count caused growth, establish a current platform benchmark, or make a legal conclusion.

The narrower conclusion is more useful: a platform-side subsidy is a reviewable network design decision only when the side, allocation, expected cross-side response, useful outcome, burden, horizon, comparison, and stop rule remain visible together. A lower price opens a test. It does not close the argument.

References

1. Parker, G. G., & Van Alstyne, M. W. (2005). Two-Sided Network Effects: A Theory of Information Product Design. Management Science, 51(10), 1494–1504. https://doi.org/10.1287/mnsc.1050.0400
2. Rochet, J.-C., & Tirole, J. (2003). Platform Competition in Two-Sided Markets. Journal of the European Economic Association, 1(4), 990–1029. https://doi.org/10.1162/154247603322493212
3. Boudreau, K. J., & Jeppesen, L. B. (2015). Unpaid Crowd Complementors: The Platform Network Effect Mirage. Strategic Management Journal, 36(12), 1761–1777. https://doi.org/10.1002/smj.2324
4. Kaiser, U., & Wright, J. (2006). Price structure in two-sided markets: Evidence from the magazine industry. International Journal of Industrial Organization, 24(1), 1–28. https://doi.org/10.1016/j.ijindorg.2005.06.002