Platform Strategy and Ecosystems

How open should the platform be, and at which layer?#

Openness is a ladder, not a switch. Each rung trades control for third-party investment, and different resources within the same company legitimately sit at different rungs.

The decision rule is comparative, not aspirational:

open the layer when
  expected value of partner innovation + partner-sourced distribution
> quality variance cost + security exposure + support load
  + coordination cost + value of the control surrendered

Two corrections to how founders usually run this. First, the two openness levers are different sizes. Boudreau's finding — roughly 5x from granting access versus roughly 20% from devolving control — says the large gain comes from letting others build, not from giving up governance. You can open the build surface and keep the rulebook. Second, openness is close to irreversible in practice. Re-closing an interface breaks businesses that were built on it, and the complementors who lose money become the loudest public critics of your platform. Treat each rung as a one-way door and open the narrowest surface that produces the learning you need.

How do you measure whether the ecosystem is real?#

Partner logo counts are the vanity metric of this category. Four measures separate an ecosystem from a directory.

1. Complementor investment, not complementor count. The economically meaningful number is what third parties spent: engineering months against your API, certifications obtained, headcount dedicated, capital deployed. A partner who spent two weeks integrating has told you the ecosystem is worth two weeks.

2. Cross-side response, measured on relevant participation.

cross-side response A←B = % change in active A / % change in relevant active B

"Relevant" is doing the work in that formula. Use participants who are available, qualified, and inside the same market boundary — geography, category, certification, time window. A global supplier count is meaningless to a buyer who needs one certified provider in one city tonight. Registered accounts are not participation.

3. The interaction funnel, segmented.

eligible → available → discovered → matched → accepted → completed → retained

Segment every step by market cell and cohort. The characteristic platform failure is adding participants while completed interactions fall, because discovery quality or trust degraded faster than supply grew.

4. Contribution per interaction, with subsidy separated out.

platform contribution = monetisation − payments − incentives − support
                        − trust and safety − variable infrastructure − expected loss

subsidy ROI = incremental cross-side contribution attributable to the subsidy
              / subsidy cost

The relevant counterfactual for subsidy ROI is what would have happened without it — which usually requires holding out a market cell, not comparing subsidised periods to unsubsidised ones. See Contribution Margin for the cost-boundary discipline and Cohort Analysis for reading the retention side.

Why is governance part of the product?#

Because every governance rule is a decision about who earns money.

Ranking determines which complementor gets demand. Eligibility determines who gets to compete at all. API deprecation policy determines whether a partner's investment has a usable life. Fee changes reallocate margin directly. Data access determines whether a complementor can build a business or only a feature. Enforcement consistency determines whether any of the above can be planned around.

Participants price this risk, whether or not you acknowledge it. The observable symptoms of governance risk are specific: complementors multi-home defensively, they build thin integrations rather than deep ones, they refuse to put your platform on their own roadmap, and they route customers off-platform after the first introduction. If you see those behaviours, the problem is rarely the fee level — it is the predictability of the rules.

Three design commitments are cheap early and expensive later:

• Publish the deprecation window before you need one. A committed notice period is the single most valuable thing you can give a complementor deciding whether to invest.
• Separate the ranking logic from the commercial team. If you also supply first-party inventory or first-party apps, the self-preferencing conflict is structural, and it is exactly what platform-focused competition analysis looks for. (DOJ/FTC, Guideline 9)
• Build export before you need to defend it. Data portability is arriving as a legal default in the EU on a fixed timetable. A platform whose retention depends on export friction is renting its retention from a regulator.

Worked example: pricing an openness decision#

Fieldpath (hypothetical) sells field-service software. It currently builds every vertical workflow itself. It is deciding whether to open a certified-partner build surface so third parties can ship vertical modules. Numbers demonstrate the method, not a benchmark.

Step 1 — What the closed path costs#

Fieldpath ships 4 verticals per year at a fully loaded cost of $310,000 each.

Annual first-party vertical cost = 4 × $310,000 = $1,240,000

Each vertical adds $180,000 of annual contribution once live.

Annual contribution added, closed path = 4 × $180,000 = $720,000

The closed path is currently contribution-negative in-year by $520,000 and only pays back across multiple years.

Step 2 — What the open path costs#

Opening a certified surface requires $540,000 of one-off platform work (docs, sandbox, certification tooling, review process) and $260,000 per year of ongoing cost (partner support, security review, certification maintenance, dispute handling).

Fieldpath expects 11 partner-built verticals in year one, each adding $180,000 of annual contribution, of which Fieldpath keeps 65% after partner revenue share.

Partner-sourced contribution to Fieldpath
  = 11 × $180,000 × 65% = $1,287,000

Year-one open-path cost
  = $540,000 + $260,000 = $800,000

Year-one net (open path only) = $1,287,000 − $800,000 = $487,000

Fieldpath also redirects two of its four first-party builds to core product, saving 2 × $310,000 = $620,000 while giving up 2 × $180,000 = $360,000 of wholly owned contribution — a net $260,000 gain on the redeployment.

Total year-one swing = $487,000 + $260,000 = $747,000

Step 3 — Stress the assumption that actually carries the result#

Every dollar above depends on 11 partners shipping, which is the number nobody can evidence before opening. Solve for the break-even partner count, holding the redeployment gain aside:

n × $180,000 × 65% = $800,000
n × $117,000 = $800,000
n = 6.84 → 7 partners

Below seven shipped verticals in year one, the openness investment loses money. And "shipped" is the right unit: a signed partner who never ships contributes nothing but support cost.

Now the downside case. Suppose only 5 partners ship, and certification and support cost runs 30% over at $338,000:

Contribution = 5 × $180,000 × 65% = $585,000
Cost         = $540,000 + $338,000 = $878,000
Year-one net = −$293,000

Step 4 — Read it correctly#

The base case is attractive and the downside is survivable, which is the right shape for a one-way door. But three things the arithmetic hides deserve to be written down next to it:

• The $540,000 is sunk and the ongoing cost is not. Partner support and security review scale with partner count, so a successful year one raises year-two fixed cost.
• Contribution per partner-built vertical is assumed equal to first-party. It usually is not — partner-built modules often convert worse and support worse, at least initially.
• The redeployment gain assumes core work is worth more than the two verticals given up. That is a product judgement, not an arithmetic result, and it is the assumption most likely to be wrong.

The honest framing for an investor or a board: the openness decision breaks even at seven shipped partner verticals, we have four in design-partner agreements, and here is what we will know by month five.

What are the common mistakes?#

• Opening into an empty ecosystem. Complementors invest where demand already exists. Opening an API before you have customers worth building for produces documentation nobody reads and a support obligation you now own.
• Counting logos instead of investment and activity. A partner page is a marketing asset. Engineering months committed, certifications held, and interactions completed are the ecosystem.
• Assuming every network effect is positive. More participants can mean congestion, spam, adverse selection, and worse discovery. Measure value per participant as density rises, not just density — see Network Effects.
• Subsidising both sides indefinitely. A subsidy with no expiry and no held-out control group is paid acquisition. Set a decay schedule at launch and hold a market cell out so you can measure what the subsidy bought.
• Treating governance as legal work. Ranking, eligibility, deprecation, and enforcement are product decisions with revenue consequences for other people's businesses. They belong to a product owner with a public changelog.

When does platform strategy break down?#

There are no genuinely distinct participant roles. If the "two sides" are the same buyers in different clothes, cross-side effects never appear and you have built two-sided infrastructure for a one-sided business.

Interactions are rare. Matching improves with participation only if participation is repeated often enough for density to accumulate. Low-frequency, high-value transactions frequently do better as a managed service or an integrated product.

Quality cannot be governed at acceptable cost. Safety-critical, licensed, or highly bespoke delivery may make third-party variance uninsurable. That is a legitimate reason to move toward the pipeline end of the spectrum, or to integrate the stage entirely — see Vertical Integration.

A conventional supplier delivers the same value more simply. If a customer wants one accountable provider and you can be that provider profitably, the platform surface is cost without benefit.

The rules stop being yours. At sufficient scale, interoperability, data access, and business-user treatment become obligations rather than choices. The DMA's first review kept the regime intact and named AI and cloud as the next enforcement frontier, so this perimeter is expanding rather than contracting. Governance choices made early — self-preferencing, exclusivity, data use, ranking control — are the ones that later create conflict with the participants you depend on.