The Infrastructure of Coordination: When Dedicated Coordinators Improve Collective Action—and When They Do Not (Report № 001)

Central finding

Dedicated coordination infrastructure is best understood not as a particular kind of organization, but as a way of repairing interfaces among interdependent actors. Backbone organizations, network administrative organizations, systems integrators, mission offices, secretariats, standards bodies, and boundary organizations repeatedly perform similar work: they specify how activities must fit together, create shared information, allocate or mobilize resources, monitor commitments, resolve disputes, supply capabilities no participant will fund alone, and represent the collective externally.

The best-supported available hypothesis is that such infrastructure improves coordinated action when its mandate, authority, resources, information access, legitimacy, and organizational position match a specific interdependence problem. When that match is absent, the same apparatus can generate meetings, plans, reports, and staff without improving handoffs or outcomes.

This conclusion is plausible, consistent with several well-documented successes and failures, and central to network-governance theory. It is not an established general law. A systematic review of 184 empirical studies published after the canonical theory of network governance found that none of its eight contingency propositions had been verified; the evidence remained “suggestive and circumstantial.” Roughly 85 percent of the reviewed studies were case-based and 77 percent were cross-sectional.¹ No study in the evidence reviewed directly compares alternative coordination architectures on the same substantive problem while holding the intervention itself constant.

Four narrower conclusions are better supported:

Coordination activity and outcome improvement frequently diverge. Independent evaluations have found improved joint working, reported service transformation, or claimed success alongside null or adverse downstream outcomes.
The function is separable from the dedicated entity. Coordination work can be distributed among member organizations, embedded in redesigned roles, performed by a lead organization, supplied by a market institution, or emerge informally.
Dedicated infrastructure fails in recurring ways: it is placed at the wrong organizational level; given responsibility without authority or information; allowed to monopolize expertise that the sponsoring principal needs for oversight; or extended into decisions governed by local knowledge.
The coordination layer can be a substantial part of total cost. In the only detailed decomposition found, coalition administration represented 37 percent of a five-year prevention programme’s cost—about the same as the interventions it selected and coordinated.² That figure is not generalizable, but it shows why coordination cannot be treated as costless background capacity.

The practical implication is therefore not “build a backbone” or “avoid centralization.” It is: identify the unresolved interface first, decide what functions it requires, and only then decide whether those functions need a dedicated institution.

1. What coordination infrastructure actually does

The institutional labels obscure a recurring functional repertoire. No comparative study has validated the following taxonomy across forms, so it should be treated as a descriptive synthesis rather than a settled classification.

Recurring function	Interface problem addressed	Examples of institutional carriers
Specify interdependence and interfaces	Components, services, or jurisdictions must fit together, but tolerances, responsibilities, or decision rights are unclear	Standards bodies, systems integrators, mission offices
Create shared information and measurement	Participants use incompatible data, cannot see the whole system, or cannot translate knowledge across boundaries	Boundary organizations, humanitarian clusters, backbone organizations
Mobilize and allocate shared resources	No member can—or wants to—finance system-level work; local allocations conflict with collective priorities	Network administrative organizations, programme offices, secretariats
Monitor commitments and performance	Participants can free-ride, implementation drifts, or joint outcomes have no accountable owner	Administrative organizations, standards bodies, programme offices
Resolve disputes and adapt the arrangement	Dependencies change or competing interests block decisions	Standards committees, secretariats, network governance bodies
Supply network-level capabilities	Integration, technical assistance, analysis, or system design is unavailable within any member	Systems integrators, technical secretariats, backbones
Represent and legitimate the collective	Funders and public authorities need a recognizable counterpart; members need buffering from outside interference	Secretariats, network administrative organizations, programme offices
Provide a stable point of contact and continuous communication	Outside principals or community participants cannot navigate a shifting network	Backbones, secretariats, administrative offices

These functions are not equally consequential. A useful distinction is between architectural and administrative coordination:

Architectural functions change what is specified, measured, funded, decided, or enforced.
Administrative functions change the amount and continuity of communication, convening, and reporting.

Administration can support architecture, but it is not a substitute for it. An office that convenes participants and produces reports but cannot alter decisions, incentives, resource flows, or interface rules has no clearly evidenced route to downstream improvement.

Different forms, similar work, different powers

The main institutional forms belong to the same functional family but occupy very different positions:

A network administrative organization exists specifically to govern a network. Theory predicts it is most useful when there are many participants and a high need for network-level competencies, but also warns that its formality creates bureaucratization and lock-in risks.³
A backbone organization concentrates agenda support, shared measurement, communication, outreach, and resource mobilization in a single entity.
A systems integrator coordinates component and supplier interfaces while maintaining an understanding of the system as a whole.
A mission or programme office combines technical direction, portfolio selection, resource allocation, and active project management.
A standards body converts decentralized compatibility problems into shared specifications and decision procedures.
A secretariat collects information, communicates with participants, coordinates activity, and may influence the allocation of attention and resources despite limited formal control.
A boundary organization mediates between distinct communities—such as science and policy—and is accountable to principals on both sides.
A project management office performs analogous work inside an organization, although the evidence on its effects is particularly weak.
Boundary-spanning roles, cross-cutting teams, and emergent coordinators can perform the same functions without creating a separate institution.

The last category is crucial. Relational-coordination research explicitly calls for redesigning rather than replacing formal structures, using cross-functional measures, protocols, conflict-resolution venues, and boundary-spanning jobs.⁴ In the World Trade Center response, most people who became actual coordinators did not hold formal coordination positions, although having such a position increased the probability of taking on that work.⁵ Dedicated infrastructure is therefore one carrier of coordination capacity, not its definition.

2. The decisive distinction: coordination architecture versus coordination activity

The strongest evidence in this literature does not show that one institutional form consistently works. It shows that process indicators are poor evidence of success.

Three warnings from public programmes

In England’s £5.3 billion Better Care Fund, more than 90 percent of local areas reported that the programme had improved joint working. Yet emergency admissions increased by 87,000, costing £311 million more than planned, while delayed-transfer days rose by 185,000 against a planned reduction of 293,000. The National Audit Office concluded that the fund had not achieved the expected value for money in savings, patient outcomes, or reduced hospital activity. Local partners did not monitor savings because they lacked a mandate to do so—a concrete gap between responsibility, authority, and information.⁶

The first phase of England’s Troubled Families Programme created local coordinators, data-matching capacity, keyworkers, and a payment-by-results system. Qualitative evidence indicated greater multiagency working, but the national evaluation found no consistent evidence of significant or systematic effects on employment, benefits, school attendance, safeguarding, or antisocial behaviour. A redesigned later phase reported narrower positive effects alongside numerous null results and some adverse early differences, illustrating that “coordination” is not a stable treatment and that results depend on programme design, population, and time horizon.⁷

In the US Comprehensive Primary Care Initiative, 497 practices received care-management payments, data feedback, learning support, and an opportunity to share savings. Practices reported substantial delivery transformation, but after two years the measured savings did not cover care-management fees—a median of $115,000 per clinician—and improvements in quality and patient experience were small.⁸

These are not proofs that coordination infrastructure was useless. Each programme contained multiple components, and some outcomes may have required longer to develop. They do establish a demanding minimum: claims of better collaboration, more complete plans, or perceived transformation cannot stand in for credible evidence of improved interfaces or beneficiary outcomes.

What more convincing positive evidence looks like

The strongest causal positive case is Communities That Care, a randomized trial across 24 US towns in seven states. Local coalitions used epidemiological data to identify risks, selected tested prevention programmes, received training and technical assistance, and monitored implementation fidelity. The package reduced the initiation of several youth problem behaviours and produced sustained changes in prevention systems. Its benefit-cost analysis estimated $4,477 in discounted benefits against $556 per youth over five years.⁹

But this is evidence for a complete implementation system, not for a coordinator or backbone in isolation. The trial bundled:

coalition infrastructure;
programme content;
grants;
epidemiological data;
training and continuing technical assistance;
intervention selection procedures; and
fidelity monitoring.

The cost decomposition is revealing. Of approximately $745,276 per community over five years, coalition activity cost $275,930, or 37 percent; intervention programmes cost 35 percent; and training, technical assistance, and implementation monitoring cost 23 percent.² The trial cannot identify how much of the effect came from the coordination layer. It demonstrates that an integrated architecture can work and that the architecture is a major intervention component.

A Medicare Accountable Care Organization study offers a qualified positive contrast. ACO contracts coupled measurement to shared-savings incentives; the 2012 entrants achieved an estimated 1.4 percent spending reduction with quality generally unchanged or improved. But 2013 entrants showed essentially no differential spending change. The study therefore suggests that aligned measurement and incentives can matter without establishing a reliable form effect.¹⁰

A better test of coordination improvement

A credible evaluation should follow the full chain:

Name the coordination failure. Examples include missed handoffs, incompatible standards, duplicated coverage, unresolved dependencies, jurisdictional externalities, or absent decision rights.
Identify the mechanism. Explain how the new infrastructure will change information, incentives, authority, resource allocation, or interface rules.
Observe changed decisions or behaviours. Shared data should be used, not merely produced; standards should change implementation; monitoring should trigger correction.
Measure interface performance. Relevant measures include handoff errors, coverage gaps, duplication, rework, waiting time, unresolved defects, conformance failures, and implementation fidelity.
Measure downstream outcomes against a credible counterfactual.
Price the full coordination burden. Include partner time, reporting and data-cleaning work, decision delay, compliance costs, displaced local coordination, and loss of discretion.
Examine distribution. Average improvement may conceal that some members bear most of the burden or lose agenda-setting power.

Most existing evaluations stop after steps two or three. Collaboration surveys are far more common than objective measures of interface quality.

3. When dedicated infrastructure is useful—and when a dedicated entity is not

Conditions favouring a dedicated layer

The evidence most strongly supports dedicated infrastructure when five conditions coincide.

1. There is an identifiable cross-boundary failure

A dedicated layer is easier to justify when the problem can be stated operationally: incompatible specifications, persistent coverage gaps, duplicated services, cross-jurisdictional externalities, or dependencies that no participant can resolve alone.

Humanitarian clusters illustrate the potential value. A six-country evaluation credited them with more predictable leadership, trained coordinators, improved gap identification, reduced duplication, better appeal planning, and improved coverage in some thematic areas.¹¹ These functions have plausible standalone value in a fast-moving environment with many organizations and incomplete information.

The same evaluation also shows why specificity matters. Coordination staff were often concentrated at national or capital level while operational coordination problems arose subnationally. The result was greater bureaucracy, process focus, and excessive meeting load. Some clusters serviced central reporting and financing demands rather than field problems, while others bypassed existing local arrangements and multiplied the number of coordination mechanisms. The evaluators concluded, on largely qualitative evidence, that benefits slightly outweighed costs and shortcomings—not that the design was uniformly effective.¹¹

2. Many participants need a capability none will fund alone

A separate body becomes more defensible as the number of participants and the need for network-wide competencies increase. These competencies may include system integration, common data, technical assistance, neutral dispute resolution, or a stable interface with external authorities.

Even here, the entity may be substitutable. In a configurational study of 39 Dutch Safety Houses, effective networks were centrally integrated, stable, and at least three years old, and had either substantial financial resources or a network administrative organization. The authors concluded that administrative and financial resources could substitute for one another.¹² The dedicated organization may therefore supply slack capacity rather than a categorically unique capability.

3. The work requires continuing rather than episodic integration

Standards, complex systems, and multi-agency services often need ongoing interface management. A one-time plan cannot handle changing components, new participants, disputes, or technical evolution.

Yet continuing need is not equivalent to permanent organization. Network-governance theory predicts that a network administrative organization is the least adaptable of the major governance forms once installed.³ The burden of proof should therefore increase when the arrangement is difficult to amend or dismantle.

4. The coordination body can act on what it learns

Information has little value if the body cannot alter resource flows, standards, incentives, or implementation. In a study of 25 collective-impact initiatives, the sites judged to have made a contribution not explained by rival accounts differed from weaker sites in their use of shared measurement to reallocate services and in strategies that changed funding flows. But the study selected initiatives already reporting change, lacked a comparator, and explicitly disclaimed causal inference. It identifies a plausible discriminator, not a rule.¹³

The general implication is still useful: shared measurement matters when it changes decisions, not when it merely creates dashboards.

5. The domain is sufficiently clear for concentrated authority

Mission offices make the domain requirement unusually explicit. The ARPA model is intended for missions that can be organized around technology-related goals with quantifiable subgoals, emerging technological opportunities, and frictions that markets will not resolve. It is not presented as a general solution for basic research or ill-defined social goals.¹⁴

An independent National Academies assessment found clear signs that ARPA-E was progressing toward its statutory mission six years after its creation, while also recommending that the agency develop an impact-measurement framework and noting that three-year projects were too short to expect concept-to-market transformation.¹⁵ The evidence supports the model’s organizational features and domain conditions more strongly than it supports claims of transformational impact.

Dedicated infrastructure is not shown to be necessary

No institutional form in the reviewed evidence is demonstrably necessary for coordination success.

Coordination functions can be embedded in existing organizations.
A lead organization can perform network-level work.
Administrative capacity can substitute for financial resources and vice versa.
Informal actors can become coordinators in crises.
Boundary work can be divided across organizations.
Market institutions can reduce transaction and interface costs.
Some effective networks operate with relatively decentralized governance.

This does not mean a separate entity is never necessary in practice. It means the literature has not established necessity. The design decision should therefore be split in two:

Does the interface problem require deliberate coordination?
Does that coordination require a new organization?

Evidence for the first proposition is much stronger than evidence for the second.

Conditions arguing against a dedicated or centralized entity

A separate body is less attractive where:

the work depends heavily on local knowledge and adaptation;
interdependence is reciprocal among autonomous producers rather than organized around a common unit;
goals are contested or cannot be expressed in operational terms;
existing organizations can perform the required functions with modest redesign;
the new body would duplicate established coordination arrangements;
no durable financing model exists;
the body would be responsible for results without authority or information;
members would lose critical capabilities needed to challenge or replace it; or
the principal costs are delay, reporting, and reduced discretion rather than missing integration.

A study of 26 US education networks found decentralized governance in two of three effective configurations and highly centralized governance in four of six pathways to ineffectiveness. The result is important but bounded: the study measured governance centralization, not the presence of a dedicated entity; one effective network had a backbone whose role was facilitative rather than directive. Moreover, the study’s “decentralized” condition partly included lack of coordination in planning or implementation, complicating interpretation.¹⁶

The unresolved question is not simply centralization versus decentralization. It is which decisions require system-wide integration and which require local discretion.

4. How dedicated coordination fails

Failure modes and evidential standing

Failure mode	Recurring mechanism	What the evidence establishes
Bureaucracy and coordination overload	Capacity sits where reporting occurs rather than where interdependence occurs; new meetings coexist with old ones	Directly documented in one multi-country humanitarian evaluation; prevalence across sectors is unknown
Responsibility without authority	The body can convene and report but cannot redirect resources, enforce commitments, or correct implementation	Visible in health and social-care integration cases
Over-centralization	Central decision-makers override local knowledge, impose uniform systems, or become the new bottleneck	Plausible and documented in cases, but the overall centralization question remains unresolved
Dependency	The principal delegates technical integration and loses the knowledge needed to specify, monitor, or replace the integrator	Strongly documented in two US defence-acquisition audits; no equally strong successful delegated-integration comparison was found
Information distortion	Actors control the data on which they are rewarded; narrow targets redirect effort or invite cosmetic compliance	Demonstrated in one policy setting; prevalence is unknown
Capture and exclusion	A coordination forum becomes a control point through which dominant actors set standards or agendas	Mechanisms are clear; frequency and distributional effects are poorly measured
Decision bottlenecks	Consensus becomes slower as distributional stakes rise	Quantified convincingly in one technical standards institution
Financial dependency or non-viability	Collective benefits are real but diffuse, leaving the intermediary without a defensible revenue stream	Documented in health-information-exchange surveys
Loss of local autonomy	Central coordination displaces member discretion or internal capability	Frequently asserted but not directly measured in the reviewed evidence
Institutional lock-in or self-preservation	The organization persists, expands, or redefines its mission after the original problem changes	Theoretically predicted; direct longitudinal evidence is scarce

Delegating integration without retaining the capacity to govern it

The clearest dependency mechanism appears in the US Coast Guard’s Deepwater programme. When the Coast Guard contracted with Integrated Coast Guard Systems in 2002, it lacked insight into how the contractor’s proposed system would meet mission needs, limiting its ability to trade off cost and capability. Its own cost baseline reflected anticipated contract costs determined by the integrator rather than an independent estimate. After significant problems, the Coast Guard brought the systems-integrator role back in-house, but rebuilding the capability took time and left staffing and acquisition-process gaps.¹⁷

The Army’s Future Combat Systems programme showed a related problem. The lead systems integrator was used because the Army lacked the agility, capability, and capacity to manage a cross-cutting system-of-systems. The Government Accountability Office warned that the relationship created long-term oversight risks, weak leverage over cost efficiency, and dependence on programme cost estimates that were lower than independent estimates. The audit establishes dependency and weak oversight; it does not prove that the integrator arrangement alone caused the programme’s later cancellation.¹⁸

The lesson is not “never outsource integration.” It is that the sponsor must retain enough technical, analytic, and commercial capacity to:

define system requirements;
maintain an independent cost and performance view;
assess trade-offs;
challenge the integrator;
change suppliers; and
re-assume the function if necessary.

Without that retained capacity, authority migrates toward the actor that controls information.

Over-centralization: the mirror-image risk

The UK National Programme for IT in the NHS is a case of excessive central authority rather than insufficient retained authority. A case history attributed its problems to haste and an overambitious centralized model imposed on semi-autonomous sites without adequate regard to user satisfaction. It distinguished electronic patient records as a potentially viable technology from the heavily centralized architecture used to deliver them. The cited £9.8 billion figure was a projected lifetime cost, paired in the National Audit Office analysis with £10.7 billion in projected benefits; the case history disputed whether those benefits would be realized and noted exclusions from the cost projection.¹⁹

This evidence is weaker than the GAO audits: the source is an unrefereed university case history, and part of its centralization diagnosis relies on trade reporting. It nevertheless illustrates the opposite design error. Coordination can fail because the centre lacks system knowledge, or because it extends control beyond what its knowledge justifies.

Consensus as a bottleneck

In the Internet Engineering Task Force from 1993 to 2003, a one-percentage-point increase in private-sector participation added an estimated 7.8 days to standards development at the mean. Internet commercialization was estimated to add roughly eight months of deliberation for a typical standard. The effect was concentrated in the upper part of the delay distribution, and academic and nonprofit authors faced delays three to six months longer. Non-standards-track documents followed the same publication process without the same distributional conflict, providing a useful within-institution comparison.²⁰

This is unusually strong evidence of a coordination body becoming slower as commercial stakes rise. It does not imply that delay is always wasteful—deliberation can improve technical quality. It shows that consensus has distributional costs and that the costs are not borne equally.

Standards institutions also create potential points of capture. A 2000 speech by an individual Federal Trade Commission staff member—not a Commission finding—described risks including collusion, exclusion, obsolete consensus standards, and capture by a subgroup.²¹ These mechanisms are credible, but the reviewed evidence does not establish how frequently they occur.

Authority and information distortion

China’s River Chief System illustrates both the potential and danger of strong accountability. A synthetic-control study of one interjurisdictional interface—14 monitoring sites across two neighbouring cities and one year after reform—estimated a 36 percent reduction in pollution at the boundary site.²² A separate national study found improvements in some pollutants, deterioration in others, and opposite conclusions depending on whether provincial or national monitoring data were used, interpreting the discrepancy as evidence of cosmetic governance.²³

The studies examine different units and can both be correct. The design lesson is not that target-based authority fails, but that it requires:

measures that cover multiple dimensions of performance;
information channels independent of the actors being rewarded;
third-party and public scrutiny; and
safeguards against shifting harm outside the measured boundary.

The evidence for information distortion is narrow—one country and policy regime—so its prevalence should not be inferred.

Capability substitution: dependency can be the benefit

Dependency is not inherently pathological. In seven modern patent pools, pool formation was associated with an average increase of about 14 percent in follow-on innovation around pooled standards. The reported mechanism was substitution: firms needed less internal integrative capability, enabling specialized, technology-focused organizations to innovate while vertically integrated firms shifted toward downstream applications.²⁴

This is the only measured substitution effect in the reviewed evidence and an important corrective to the usual dependency critique. The relevant question is not whether dedicated infrastructure replaces members’ capabilities. It is which capabilities it replaces and whether members or sponsors retain the ability to specify, challenge, and replace the institution.

In patent pools, reducing the need for every firm to integrate complementary rights may be the point. In Deepwater and Future Combat Systems, losing the sponsor’s ability to evaluate the integrator became an oversight failure.

5. Authority, resources, information, legitimacy, and position must work as a system

The recurring determinants of performance are interdependent. Possessing one cannot compensate indefinitely for lacking the others.

Authority

A coordinating body needs enough authority to change the decisions generating the problem. Convening authority alone may produce communication without implementation. But authority should be bounded by the information available at the level where decisions are made.

No evidence supplies a general rule for the optimal degree of authority. Mission offices rely on discretion but risk weak accountability; systems-integration cases show the danger of giving away too much sponsor authority; centralized public IT shows the danger of extending central authority too far.

Resources

Resources determine whether coordination functions can be performed continuously rather than ceremonially. Yet staffing is not sufficient. A five-country humanitarian WASH evaluation found that staffing was necessary but did not track perceived achievement consistently; the highest-staffed context scored lowest on reported outcomes. The study measured partner perceptions over six months, not beneficiary outcomes, and evaluated a framework developed with the Global WASH Cluster, so causal conclusions would be inappropriate.²⁵

The financing model also matters. In 2011, 50 of 75 operational US regional health-information organizations failed stated criteria for financial viability. By 2021, the operational population had fallen from 106 in 2014 to 89, and more than half still reported financial difficulty and competition from vendor-based exchange networks. Survivors tended to be larger and to offer services beyond basic connectivity.²⁶ This shows contraction and consolidation under competitive pressure, not proof that vendor networks displaced neutral infrastructure or that neutrality caused the revenue problem.

Information access

Information is what makes authority usable. A coordination body or sponsoring principal needs:

visibility across the relevant system;
access to disconfirming as well as favourable information;
independent cost and performance estimates;
measures that cannot be unilaterally controlled by rewarded parties; and
knowledge boundaries wider than immediate operational responsibilities.

The systems-integration audits and River Chief studies make the same point from opposite directions: authority without independent information invites dependency or distortion.

Legitimacy

Dedicated bodies face two constituencies:

Internal legitimacy: members and implementers accept joint action and regard the body’s procedures as fair and useful.
External legitimacy: funders, regulators, and public authorities treat the collective as a real and accountable entity.

Boundary-organization research expresses this as dual accountability. Useful information must be salient to users, technically credible, and procedurally legitimate; these qualities can trade off rather than rise together.²⁷ A body that performs well externally can still undermine internal legitimacy if it becomes the sole public face, displacing members’ relationships with funders and stakeholders.

The empirical basis for strong general claims is limited. A frequently cited matched comparison of one successful and one disbanded small-firm network could not be substantively inspected in the evidence reviewed, so the proposition that dedicated bodies systematically neglect one side of legitimacy should be treated cautiously.

Boundary position

Sitting between organizations can confer trust and system visibility. It does not confer neutrality automatically. Neutrality must be produced through transparent procedures, balanced representation, dual accountability, appeal rights, and independent information.

Boundary position also creates an economic problem: the intermediary may create value that accrues mainly to the actors it connects. Without public funding, member payments, or value-added services, the body can be valuable and financially unsustainable at the same time.

6. Transferable design principles

The following principles are supported to different degrees. The first group follows relatively directly from the evidence; the second consists of promising design hypotheses not yet tested as complete architectures.

Principles with comparatively strong support

1. Define the interface failure before selecting the institution

State the problem in observable terms:

Which handoff fails?
Which dependency remains unresolved?
Where is coverage duplicated or absent?
Which specifications are incompatible?
Which decision rights are missing or contradictory?
What cross-boundary cost is currently externalized?

“Improve collaboration” is not an adequate problem definition.

2. Separate the function decision from the organization decision

First decide what coordination functions are required. Then compare possible carriers:

member-led shared governance;
a lead organization;
redesigned roles inside existing organizations;
a rotating secretariat;
a dedicated administrative body;
a contractual integrator;
a standards or market institution; or
a layered combination.

A new organization should not be the default response to a coordination problem.

3. Use the minimum sufficient mandate

Give the coordinating layer the smallest scope and authority capable of repairing the interface. Preserve implementation discretion where local information is decisive. Expansion should require evidence that the new scope addresses a real unresolved dependency.

4. Couple responsibility to authority, information, resources, and consequences

A body should not be judged responsible for outcomes it cannot influence. Conversely, an empowered body should face transparent performance tests and consequences. Dashboards without decision rights and mandates without independent information are both incomplete designs.

When integration is delegated, the principal should retain technical, analytical, audit, procurement, and cost-estimation capability. It must be able to understand the system independently of the integrator.

6. Measure interface outcomes rather than administrative volume

Core measures should include:

missed or delayed handoffs;
unresolved dependencies;
rework and waiting time;
duplicated or uncovered services;
interface defects;
standard-conformance failures;
implementation fidelity;
beneficiary outcomes; and
net economic and organizational cost.

Attendance, reporting volume, staffing, and participant satisfaction are useful diagnostic measures, not final outcomes.

7. Price the distributed burden

Cost accounting should include time spent by participating organizations, data production, meeting load, approval delay, compliance work, and displaced local coordination—not merely the central office’s budget.

Cost incidence also matters. In humanitarian coordination, agencies and donors bear much of the direct cost of coordination, while affected populations bear much of the cost of non-coordination.¹¹ The party resisting expenditure may not bear the counterfactual harm.

8. Make measurement independent and plural

Do not allow rewarded actors to control the only performance channel. Use multiple measures, third-party evaluation where feasible, and opportunities for local actors and affected users to contest official representations.

9. Design dual accountability

Specify how funders, authorizers, member organizations, implementers, and affected users can influence the body. Representation alone is insufficient; the architecture needs appeal, review, and challenge procedures.

10. Specify amendment, exit, and sunset arrangements at founding

Define:

how scope changes;
how members appeal decisions;
how leadership can be replaced;
how data and capabilities remain portable;
how the function can return to members or move elsewhere; and
what evidence triggers renewal or dissolution.

This principle is normatively attractive but poorly observed in practice. The reviewed evidence contains almost no longitudinal study of an institution being deliberately sunset and its functions transferred.

Design hypotheses worth testing

A. Prefer composed architectures to unitary ones

Two literatures suggest dividing coordination work across multiple bodies.

Boundary chains connect specialized intermediaries. A “key chain” links one boundary organization to several partners in parallel; a “linked chain” moves information through successive organizations that deepen customization; a “networked chain” connects organizations across chains. The proposed advantage is lower transaction cost without requiring one body to serve every user. The stated precondition is complementary skill and non-overlapping mutual dependence among the links.²⁸

Layered governance separates constitutional or strategy-level orchestration from administrative monitoring and control. A study of the Universal Postal Union distinguishes a member-oriented constitutional layer responsible for socialization and strategy from an administrative organization responsible for accounting, monitoring, and facilitating collaboration.²⁹

These are plausible architectures, not proven superior designs. Neither has been compared with a unitary body on common outcomes.

B. Give the coordinating layer wider visibility than authority

A review of the “mirroring” literature found that organizational ties often follow technical dependencies in firms and consortia, while open collaborative projects frequently behaved differently. In the review’s design-oriented cases, “partial mirroring”—knowledge boundaries wider than operational boundaries—was associated with strong performance, particularly among systems integrators, although the authors warned that selection of successful cases created survivorship bias.³⁰

The resulting hypothesis is memorable: place decision rights at the interface, but give the coordinating layer visibility above it. This may avoid both the field-level staffing mismatch seen in humanitarian clusters and the excessive centralized control seen in large public IT programmes.

C. Match centralization to the type of interdependence

A plausible but untested moderator may explain conflicting findings:

Where interdependence is pooled around a common unit—a client, physical system, or shared standard—centralized integration may help.
Where interdependence is reciprocal among autonomous producers, directive centralization may impede adaptation and capability-building.

No study in the reviewed evidence tests this hypothesis directly. It should guide experimentation, not be presented as an established rule.

7. Important uncertainties and unresolved questions

Centralization remains unresolved

A foundational four-network mental-health study is widely cited as finding that centralized integration, external control, stability, and resources predicted better client outcomes. Only its bibliographic record and secondary descriptions were available in the reviewed material, so the pro-centralization finding could not be substantively verified here.³¹ Later configurational evidence points in different directions depending on setting.

The unresolved issue is not merely whether centralization helps. It is which task characteristics, information structures, and accountability arrangements determine when it helps.

Local autonomy is not measured

Loss of local autonomy is prominent in criticism of dedicated coordination bodies but is not directly measured in the reviewed studies. The canonical theory explicitly concentrates on network-level tensions rather than tensions experienced inside participating organizations.

This is a material absence. Local autonomy can support contextual adaptation and legitimacy, but it can also enable free-riding, inconsistent implementation, or avoidance of collective obligations. The evidence does not provide a settled way to value the tradeoff.

Capture and distribution are poorly observed

Studies usually report average network performance. They rarely show:

who gains agenda-setting power;
whose information becomes authoritative;
which organizations absorb reporting costs;
whose preferred standards are adopted;
which communities lose discretion; or
who faces higher exit costs.

The IETF study shows that coordination delay can fall disproportionately on academic and nonprofit participants. Beyond such cases, capture frequency and distributional incidence remain largely unmeasured.

Dependency after withdrawal is largely unknown

The Communities That Care follow-ups observed outcomes one year and 1.5 years after study-provided resources ended, making them the only meaningful post-withdrawal evidence in the reviewed material. No study follows a coordination body through deliberate dismantling or transfer of its functions back to members.

The proposition that dedicated infrastructure atrophies members’ coordination capacity is therefore mostly untested. The patent-pool case shows that capability substitution can be beneficial; the defence-acquisition cases show that it can undermine oversight.

The positive evidence is usually bundled

The strongest positive programme combines coordination with data, money, technical assistance, training, programme content, and fidelity controls. Mission-office evidence similarly bundles organizational form with exceptional recruitment, flexible contracting, and concentrated resources. The literature rarely separates the effects of:

the coordinator;
shared data;
funding;
technical assistance;
enforcement;
standards;
programme content; and
leadership quality.

Cost measurement is thin

Beyond a few programme-specific figures, systematic accounting for coordination costs is rare. Partner time, delayed decisions, data cleaning, duplicate reporting, reduced discretion, and displaced local work are usually omitted. This makes both claims of excessive overhead and claims of underinvestment difficult to assess.

The functional taxonomy itself remains unvalidated

No study compares backbones, secretariats, integrators, programme offices, and standards bodies on a common functional measure. Their apparent similarity is a cross-literature synthesis, not an empirical demonstration that all constitute a stable institutional class.

Conclusion

Dedicated coordination infrastructure is neither a universal solution nor mere overhead. It is a consequential design choice for governing interfaces.

Its potential value lies in changing the architecture of collective action: clarifying dependencies, making the system visible, assigning decision rights, aligning resources, resolving disputes, and supplying capabilities that fragmented participants cannot provide individually. Its characteristic danger is that the coordinating layer becomes detached from the work—accumulating meetings and reporting—or becomes so central that it distorts information, monopolizes expertise, suppresses local knowledge, or turns into a bottleneck.

The most defensible design stance is therefore minimum sufficient, problem-specific, and reversible infrastructure:

Diagnose the interface failure.
Specify the functions required to repair it.
Compare dedicated and non-dedicated carriers.
Couple responsibility to usable authority, independent information, and resources.
Preserve sponsor capacity and local decision rights where they matter.
Evaluate handoffs, outcomes, distribution, and total burden—not administrative activity.
Build in appeal, amendment, portability, and exit.
Consider composed architectures that separate strategy, administration, technical integration, and local implementation.

The evidence supports this stance as the best available hypothesis, not as a proven formula. The next decisive research would compare alternative coordination architectures on the same problem, isolate components of bundled models, measure local autonomy and distributional effects, and observe what happens when dedicated infrastructure is removed. Until then, the central discipline is to resist confusing the visible apparatus of coordination with coordinated action itself.

Steven van den Oord, Patrick Kenis, Jörg Raab, and Bart Cambré, “Modes of Network Governance Revisited: Assessing Their Prevalence, Promises, and Limitations in the Literature,” Public Administration Review 83, no. 6 (2023): 1564–1598, doi:10.1111/puar.13736. ↩
Margaret R. Kuklinski, Abigail A. Fagan, J. David Hawkins, et al., “Benefit-Cost Analysis of a Randomized Evaluation of Communities That Care,” Journal of Experimental Criminology 11, no. 2 (2015): 165–192, doi:10.1007/s11292-014-9226-3, PubMed Central. ↩ ↩²
Keith G. Provan and Patrick Kenis, “Modes of Network Governance: Structure, Management, and Effectiveness,” Journal of Public Administration Research and Theory 18, no. 2 (2008): 229–252, doi:10.1093/jopart/mum015. ↩ ↩²
Jody Hoffer Gittell and Anthony L. Suchman, Summary of Relational Coordination Research (2013), research summary by the theory’s originators, PDF. ↩
Miruna Petrescu-Prahova and Carter T. Butts, “Emergent Coordinators in the World Trade Center Disaster,” International Journal of Mass Emergencies and Disasters 26, no. 3 (2008): 133–168, doi:10.1177/028072700802600301. The reviewed representation was limited to abstract and secondary description. ↩
UK National Audit Office, Health and Social Care Integration, HC 1011, Session 2016–17 (2017), report. ↩
Laurie Day, Caroline Bryson, Clarissa White, et al., National Evaluation of the Troubled Families Programme: Final Synthesis Report (UK Department for Communities and Local Government, 2016), report; UK Ministry of Housing, Communities and Local Government, National Evaluation of the Troubled Families Programme 2015–2020: Family Outcomes—National and Local Datasets, Part 4 (2019), report. ↩
Stacy B. Dale, Arkadipta Ghosh, Deborah N. Peikes, et al., “Two-Year Costs and Quality in the Comprehensive Primary Care Initiative,” New England Journal of Medicine 374 (2016): 2345–2356, doi:10.1056/NEJMsa1414953. The evidence reviewed included the authoritative abstract rather than the full article. ↩
J. David Hawkins, Sabrina Oesterle, Eric C. Brown, et al., “Sustained Decreases in Risk Exposure and Youth Problem Behaviors After Installation of the Communities That Care Prevention System in a Randomized Trial,” Archives of Pediatrics & Adolescent Medicine 166, no. 2 (2012): 141–148, doi:10.1001/archpediatrics.2011.183; J. David Hawkins et al., “Youth Problem Behaviors 8 Years After Implementing the Communities That Care Prevention System,” JAMA Pediatrics (2014), doi:10.1001/jamapediatrics.2013.4009. Trial records were accessed through authoritative bibliographic abstracts; the benefit-cost paper was inspected in full. ↩
J. Michael McWilliams, Laura A. Hatfield, Michael E. Chernew, Bruce E. Landon, and Aaron L. Schwartz, “Early Performance of Accountable Care Organizations in Medicare,” New England Journal of Medicine 374 (2016): 2357–2366, doi:10.1056/NEJMsa1600142, PubMed Central. ↩
Julia Steets, François Grünewald, Andrea Binder, et al., Cluster Approach Evaluation 2: Synthesis Report (Global Public Policy Institute and Groupe URD for the Inter-Agency Standing Committee, 2010), PDF. ↩ ↩² ↩³
Jörg Raab, Remco S. Mannak, and Bart Cambré, “Combining Structure, Governance and Context: A Configurational Approach to Network Effectiveness,” Journal of Public Administration Research and Theory 25, no. 2 (2015): 479–511, doi:10.1093/jopart/mut039. The reported substitutability result was verified from an institutional-repository record rather than the full article. ↩
ORS Impact and Spark Policy Institute, When Collective Impact Has an Impact: A Cross-Site Study of 25 Collective Impact Initiatives (2018), report. ↩
Pierre Azoulay, Erica Fuchs, Anna P. Goldstein, and Michael Kearney, “Funding Breakthrough Research: Promises and Challenges of the ‘ARPA Model,’” Innovation Policy and the Economy 19 (2019): 69–96, doi:10.1086/699933. ↩
National Academies of Sciences, Engineering, and Medicine, An Assessment of ARPA-E (2017), doi:10.17226/24778. The reviewed artifact was the report highlights rather than the full consensus report. ↩
Michelle Shumate, Shaun Dougherty, Joshua-Paul Miles, et al., “Network Effectiveness in Context,” Northwestern University Institute for Policy Research Working Paper WP-21-35 (2021), working paper; subsequently published in Journal of Public Administration Research and Theory 33, no. 4 (2023). ↩
US Government Accountability Office, Coast Guard: As Deepwater Systems Integrator, Coast Guard Is Reassessing Costs and Capabilities but Lags in Applying Its Disciplined Acquisition Approach, GAO-09-682 (2009), report page. ↩
US Government Accountability Office, Defense Acquisitions: Role of Lead Systems Integrator on Future Combat Systems Program Poses Oversight Challenges, GAO-07-380 (2007), report page. ↩
Oliver Campion-Awwad, Alexander Hayton, Leila Smith, and Mark Vuaran, The National Programme for IT in the NHS: A Case History (University of Cambridge, 2014), PDF. ↩
Timothy Simcoe, “Standard Setting Committees: Consensus Governance for Shared Technology Platforms,” American Economic Review 102, no. 1 (2012): 305–336, doi:10.1257/aer.102.1.305. ↩
David A. Balto, “Standard Setting in a Network Economy,” Cutting Edge Antitrust Law Seminars International speech, February 17, 2000, FTC transcript. The speaker explicitly described the remarks as preliminary personal views that did not necessarily reflect those of the Commission or its commissioners. ↩
Yixin Liu and Chao Tan, “The Effectiveness of Network Administrative Organizations in Governing Inter-jurisdictional Natural Resources,” Public Administration 101, no. 3 (2023): 932–952, doi:10.1111/padm.12834. ↩
Jing Li, Xing Shi, Huaqing Wu, and Liwen Liu, “Trade-off Between Economic Development and Environmental Governance in China: An Analysis Based on the Effect of River Chief System,” China Economic Review 60 (2020): 101403, doi:10.1016/j.chieco.2019.101403. The evidence reviewed was the publisher-supplied abstract. ↩
Keyvan Vakili, “Collaborative Promotion of Technology Standards and the Impact on Innovation, Industry Structure, and Organizational Capabilities: Evidence From Modern Patent Pools,” Organization Science 27, no. 6 (2016): 1504–1524, doi:10.1287/orsc.2016.1098. ↩
Daniel Nguyen, Camille Heylen, Cora Cunningham, et al., “Effectiveness of Coordination in Humanitarian Water, Sanitation, and Hygiene Response,” Conflict and Health 20 (2026), article 19, doi:10.1186/s13031-026-00756-7. ↩
Julia Adler-Milstein, David W. Bates, and Ashish K. Jha, “A Survey of Health Information Exchange Organizations in the United States,” Annals of Internal Medicine 154, no. 10 (2011): 666–671, doi:10.7326/0003-4819-154-10-201105170-00006; Julia Adler-Milstein et al., “A Survey of Health Information Exchange Organizations in Advance of a Nationwide Connectivity Framework,” Health Affairs 40, no. 5 (2021), doi:10.1377/hlthaff.2020.01497. Both were accessed through authoritative abstracts. ↩
David H. Guston, “Boundary Organizations in Environmental Policy and Science,” Science, Technology, & Human Values 26, no. 4 (2001): 399–408, doi:10.1177/016224390102600401; David W. Cash, William C. Clark, Frank Alcock, et al., “Knowledge Systems for Sustainable Development,” Proceedings of the National Academy of Sciences 100, no. 14 (2003): 8086–8091, doi:10.1073/pnas.1231332100. The latter was accessed through its authoritative abstract. ↩
Maria Carmen Lemos, Christine J. Kirchhoff, et al., “Moving Climate Information off the Shelf: Boundary Chains and the Role of RISAs as Adaptive Organizations,” Weather, Climate, and Society 6, no. 2 (2014), doi:10.1175/WCAS-D-13-00044.1; Christine J. Kirchhoff, Rebecca Esselman, and Daniel G. Brown, “Boundary Organizations to Boundary Chains: Prospects for Advancing Climate Science Application,” Climate Risk Management (2015), doi:10.1016/j.crm.2015.04.001. The second source was available through an abstract rather than retained full text. ↩
Peter D. J. Kilfoyle and Alan J. Richardson, “Governance and Control in Networks: A Case Study of the Universal Postal Union,” Accounting, Auditing & Accountability Journal 28, no. 4 (2015): 551–580, doi:10.1108/AAAJ-04-2014-1687. The evidence reviewed was the publisher-supplied structured abstract. ↩
Lyra J. Colfer and Carliss Y. Baldwin, “The Mirroring Hypothesis: Theory, Evidence and Exceptions,” Industrial and Corporate Change 25, no. 5 (2016): 709–738, doi:10.1093/icc/dtw027. ↩
Keith G. Provan and H. Brinton Milward, “A Preliminary Theory of Interorganizational Network Effectiveness: A Comparative Study of Four Community Mental Health Systems,” Administrative Science Quarterly 40, no. 1 (1995): 1–33, doi:10.2307/2393698. Only an authoritative bibliographic record and secondary descriptions were available in the supplied evidence; the article’s substantive text was not directly inspected. ↩