Yongjian Ke

Governing project networks: how far the rules from above reach

· 8 min

A large infrastructure project is rarely delivered by an organisation. It is delivered by a set of firms that assemble for the job, separate when it ends, and reassemble in a different combination for the next one. Most governance research looks at one of those combinations at a time and asks how a single project network runs itself. Three outputs from a study led by Ralf Müller at BI Norwegian Business School ask a different question: who sets the conditions under which those networks form at all, and how far down does that reach (Müller et al., 2022; Unterhitzenberger et al., 2023; Müller et al., 2024). The work was funded by the Project Management Institute and ran through seven country teams. I was the co-investigator for Australia.

Three layers, not one

The framework separates three things the field had largely treated as one. Network governance is the governance of a single network delivering a single project. Governance of networks is the governance of the several networks an owner or investor runs at the same time: a knowledge-sharing network that trains firms in a new standard between projects, an information-sharing network that operates during tendering, an execution network that builds the asset. Metagovernance sits above both. It is the set of semi-permanent policies and guidelines issued by a government or a large investor that decide what kinds of networks are possible in the first place.

Metagovernance is borrowed from political science, from Bob Jessop's account of how governments govern systems that are supposed to govern themselves. It has four modes and a fifth that balances them. Meta-exchange decides which markets and project types exist, which in practice means what is on the government's list of projects. Meta-organisation decides which kinds of organisation may take part, including whether special purpose vehicles are permitted. Meta-heterarchy decides how much freedom a network has to organise itself, which is where public tendering rules sit. Meta-solidarity decides what collaboration is encouraged, such as funding a training programme so that a whole supply base learns the same standard. None of this is written for a particular project. All of it constrains every project that follows.

At the bottom layer, multilevel governance theory splits the network in two. Type I is the hierarchical part: owner, client organisation and tier one suppliers, with clear lines of authority and accountabilities that are meant not to overlap. Type II is the networked part: suppliers coordinating with each other at task level, often informally and case by case, with responsibilities that do overlap. Both are present at once, which is why governance theories built for either hierarchies or markets describe a project network poorly on their own.

The two are joined by interface bodies of three levels of formality. Clubs are voluntary and informal. In one case, representatives of independent firms presented themselves to the client as a single company; in another, network members joined in immediately, and without being asked, to fix a technical fault that would otherwise have delayed the project. Agencies are formal, headed by a Type I representative and staffed from the subcontractors. In a Scandinavian railway project employing about 120 subcontractor resources, ten agencies were formed, each led by an appointed representative of a beneficiary group. Boards are the most formal. In a school construction project, the city government set up three of them, covering legal, technical and financial matters, reporting to the city.

What the interviews showed

The qualitative study covers 28 networks, using 124 interviews across ten countries, collected between May and December 2020. A case was defined as a network of at least three companies that had worked together several times on different projects in the previous five years. Three of the cases are Australian.

Networks form in two ways, and the way decides much of the rest. Orchestrated networks are assembled by a prime contractor through a defined selection process and tend to be hierarchical, with Type I governance dominant. Emergent networks come out of prior working relationships, chance meetings and reputation, and tend to be more democratic, with stronger Type II governance. Most cases were hybrids of the two, but the tendency is consistent, and the choice is made at formation.

Selection at tender stage weighs location, price, quality, trust, commercial position, digital capability, prior experience of the contractor, knowledge of the market, previous relationship with the client and firm culture. Price matters and is routinely offset by trust and quality, with a higher price accepted where the expected quality justifies it. The evaluation runs both ways: contractors assess the prime contractor before deciding whether to bid.

Once work starts, trust does most of the governing. Parties frequently begin before all contracts are signed. Underperformance is usually met by other network members helping the failing firm rather than by penalties, on the reasoning that the project still has to be finished; the cost to that firm arrives later, in reputation and in not being asked again. Contractual penalties are a last resort. Control is applied only to the extent needed, largely through peer observation and reviews, and interviewees described a general drift from control towards trust as relationships mature.

National differences appear where the framework predicts them. Chinese networks were strictly orchestrated through formal supplier selection systems. Icelandic networks were mostly emergent, Canadian ones mostly orchestrated. Scandinavia has been moving towards Type II arrangements over time and Lithuania in the opposite direction. These are not preferences held by individual firms. They follow from national policy, procurement rules and established practice, which is the metagovernance layer doing its work.

Whether a project network ends up hierarchical or democratic is mostly settled by how it was assembled, and it was assembled before the project began.

What the survey tested

The second output tests the framework quantitatively, using 225 usable responses gathered by snowball sampling through professional bodies and the research team's own contacts. Type I and Type II were confirmed as distinct constructs rather than two ends of one scale. Metagovernance was significantly associated with Type I governance. Its association with Type II, the delivery end, was not significant.

What happens in between is the more useful result. When the clarity of accountabilities at the governance of networks level enters the model, the effect of metagovernance on Type I governance nearly disappears, with the standardised coefficient falling from 0.246 to 0.035 and 85 per cent of the effect running through accountabilities. Clarity of responsibilities, meaning work done to accepted professional standards and monitored, absorbs 67 per cent of it. For Type II the pattern differs again: clearly defined accountabilities and informal partner selection weaken the influence of metagovernance, while clearly defined responsibilities strengthen it.

Put plainly, a policy set by a government or an investor does not reach the people delivering the work on its own. It reaches them through whether somebody in between has written down who is answerable for what. Where that has been done well, the writing down carries the effect and the policy above adds little on top of it.

Some secondary results are worth keeping. Liberal governance, democratic leadership and outcome-based control were significantly more common in networked topologies than in hierarchies, in Europe than in China, in international projects than national ones, and in larger projects than smaller ones. Working to professional standards, and enforcing them, was more strongly expressed in engineering projects than in IT or organisational change projects, and in networks of more than 100 organisations than in networks of fewer than 10.

The white paper adds a performance analysis that neither journal article reports. There, the three layers together account for 51 per cent of project and governance performance. Clear accountabilities have a direct positive effect that was not conditional on any other variable measured. The effect of clear responsibilities depends on the delivery layer: where Type II governance is weak, unclear responsibilities produce the lowest performance in the sample and clarifying them lifts performance steeply; where Type II is strong, unclear responsibilities do little damage. The influence of metagovernance on performance is stronger under more authoritarian structures, while democratic structures compensate for weak metagovernance and flatten the gain from strong metagovernance.

What follows for practice

Write the accountabilities between organisations, not only inside the contract. This is the one variable in the set with a direct effect on performance that was not conditional on anything else measured, and it is also the variable that absorbs the influence of every policy above it. It belongs at the level of the owner's portfolio of networks: who is answerable for what across the networks, who to contact when something technical goes wrong, and where an issue escalates to.

Do not expect a policy to reach the delivery end by itself. The survey found no significant direct association between metagovernance and the networked part of governance, which is where the work is actually coordinated. A policy meant to change how suppliers work with each other needs something in between to convert it, and on this evidence that something is the definition of roles and answerability.

Decide the formation method with the structure you want in mind. Formal selection produces hierarchy; informal formation produces the democratic pattern. Public procurement rules make the first largely unavoidable for public projects, and that is a trade rather than an accident. The transparency requirement and the flexibility the delivery end needs pull against each other, and the tender stage is where the trade is made.

Match the interface body to the issue. Clubs work for one-off problems where the parties trust each other's capability. Agencies work for recurring themes needing formal representation from both sides. Boards work for compliance and external scrutiny. The warning in the white paper is against being locked into a single type, and the worst combination in the data is a weak delivery layer with responsibilities nobody has defined.

What the papers do not settle

The framework was developed on large construction projects and megaprojects, with networks typically above 30 organisations, while the survey drew on a wider mix: 53 per cent of respondents described networks of fewer than 10 organisations, and about a third of the projects cost less than one million euros. The paper notes that this blurs the boundary conditions. The results are best read as a pattern across project types rather than a settled account of megaprojects. The sample is also concentrated, with 64 per cent of the projects in China, and recruitment was by snowball sampling, so no response rate can be reported.

The design is cross-sectional, and each network is described by one person, most often a project or other manager. The causal ordering, running downwards from metagovernance to the individual network, follows from the logical and temporal sequence of the layers rather than from anything observed, and the authors say as much. Several constructs sit near the lower bound of accepted reliability, and one of them, collaboration across networks, did not meet the thresholds and was left out of the analysis, so that part of the framework remains untested. The qualitative analysis reports its findings at an abstract level without quoted interview material, which follows from the method it uses.

The performance results, including the figure of 51 per cent, appear in the white paper rather than in either journal article. Both articles treat the link between governance layers and project performance as work still to be done, so that figure is best taken as indicative.

Papers discussed on this page