n.V1-9.08 | Coordination Nobody Can Be Compelled To Do

01| A common charge against development work is that it has no coordination architecture. Humanitarian response, the argument goes, has clusters, country teams, response plans and pooled funds — a system that assigns responsibility and produces a single plan. Development work in the same city has a dozen agencies operating in parallel with no shared plan, no shared data and no shared accountability.

02| The observation about the dozen agencies is accurate. The explanation is wrong. Development has built coordination machinery repeatedly, for twenty years, at the highest political level available to it. Understanding why that machinery did not produce a shared plan is more useful than believing it was never built.

Where the clusters came from

03| Start by noticing that the crisis sector's coordination system is a retrofit, adopted in living memory, in response to precisely the complaint now made against development.

04| After the 2004 Indian Ocean tsunami, an independent Humanitarian Response Review examined how the international response had performed. It found fragmented responses, duplication of effort, and insufficient engagement with governments and national actors. The cluster approach — sectoral coordination groups, each with a designated lead agency — was designed to fix that, and rolled out from 2005.

05| So the comparison is not between a naturally coordinated field and a naturally chaotic one. It is between a field that was told the same thing about itself twenty years ago and built a specific mechanism in response, and a field that was told the same thing and built something else.

06| It is also worth knowing what the clusters achieved, because they are frequently held up as a model by people who have not read the evaluations. The system's own multi-phase evaluation found systemic improvement in coordinated response — genuine progress — alongside a troubled early rollout, uneven performance between country cases and between clusters, weaknesses in the Provider of Last Resort provision meant to guarantee someone fills a gap, and, in the evaluation's own terms, an undisputed increase in workload. The system also spent much of its first decade being criticised for excluding the national and local organisations doing much of the actual work.

07| That is a mechanism that works, imperfectly, at a cost. It is not a miracle, and holding it up as the standard development failed to meet overstates what it delivers.

What development built instead

08| Development's answer to the same problem was not a delivery-coordination system. It was a series of political agreements about how donors would behave.

09| The Paris Declaration on Aid Effectiveness, in 2005, committed donors to align behind country systems, harmonise their procedures, reduce duplicative missions and analyses, and be accountable for results — with indicators and target dates. Accra followed in 2008, Busan in 2011. Busan created the Global Partnership for Effective Development Co-operation, which ran monitoring rounds in 2014, 2016 and 2018. Underneath these sit sector-wide approaches, joint programming, UN country frameworks, and in most capitals a standing set of donor coordination groups organised by sector.

10| By volume of architecture this is more coordination machinery than the humanitarian system has, not less. What it produced is the interesting part.

11| The OECD's own progress report found that globally, donors and developing countries had fallen short of the goals they set themselves for 2010. Its more pointed finding is the one worth carrying into any argument about this: partner country authorities had gone further in implementing their commitments than donors had. The governments receiving aid did more of what was agreed than the agencies giving it. And donor requirements and processes were found to be generating unproductive transaction costs, drawing down limited national capacity, and not always fitting national priorities and systems — which is what the agreements had been written to stop.

12| The underlying trend did not reverse. Later World Bank work tracking proliferation and fragmentation — the number of donors per country, the number of activities, their average size — shows the structural pattern continuing well past the agreements meant to curb it. And the political energy dissipated: after three monitoring rounds the exercise faded, prompting people inside the field to ask openly whether the Paris agenda had simply disappeared.

The scale of the fragmentation

13| The numbers are worth stating because they are rarely stated. By 2007 more than 90,000 aid projects were running simultaneously across the developing world, with the most heavily targeted countries handling more than 2,000 projects in a single year. By 2019 the count of aid transactions had risen to more than 222,000, at an average size of $1.4 million.

14| Average project size has fallen persistently for decades: from roughly $15 million in the early 1970s to around $1 million by the late 2000s, and average ODA grant size fell from $1.7 million in 2000 to $0.76 million by 2021. Some recipient governments — Ethiopia is the standard example — deal with more than 250 separate donor agencies.

15| The recognised consequence is that a larger number of smaller projects multiplies transaction and coordination costs for recipients, and that this burden falls hardest on exactly the countries with least administrative capacity to absorb it.

16| One caution, and it matters for how this argument gets used. What fragmentation describes is a coordination failure among funders and implementers, whose primary victim is the recipient government's administrative capacity. It is a real mechanism with real costs. It is not the mechanism by which a child in a settlement does not learn to read. Keeping those separate is the difference between an argument that holds and a document in which the absence of a particular kind of programme appears as a cause of poverty — a claim nobody would defend if it were stated plainly.

The difference that actually matters

17| Both sectors identified fragmentation and both responded. One built a mechanism that partly works. The other built commitments that its own monitoring shows were not met, principally by the parties who made them. The honest explanation is structural rather than moral.

18| A cluster coordinates delivery inside a bounded event. There is a disaster with a start date, a defined geography, a lead agency named in advance for each sector, and a stated obligation — the provider of last resort — that someone must fill a gap if no one else does. Failure to coordinate shows up immediately and visibly, as a camp with three water projects and none for sanitation.

19| Development in a settlement has none of these features. There is no start date and no end date. There is no lead agency; the World Bank, a UN agency, a bilateral donor and a national ministry are not in a hierarchy with each other. Nobody is the provider of last resort, so a gap can persist indefinitely without any organisation being in breach of anything. And because there is no single moment of delivery, non-cooperation never becomes visible as failure — it looks like two organisations pursuing separate mandates, which is what it is.

20| Add the incentive from the earlier chapter. Coordination costs an agency autonomy over its own portfolio, slows its disbursement, and makes its results harder to attribute to itself in a report to its own funder. Every one of those is a cost to the organisation and a benefit to the system. Voluntary commitments do not survive that arithmetic, and Paris was voluntary, as was Busan, as is the Global Partnership.

21| So the charge that development has no coordination architecture should be dropped, because it is easy to refute and it wastes the stronger argument. The stronger argument is this. Development attempted coordination at the highest level three times in fifteen years, monitored its own compliance, published findings showing the donors fell short of their own targets while recipient governments did better — and then allowed the monitoring itself to lapse. The problem is not an absent mechanism. It is that every mechanism built has been one no participant can be compelled to use, in a field with no lead agency, no provider of last resort, and no moment at which failing to coordinate becomes visible to anyone outside the room.

22| A settlement with a dozen agencies and no shared plan is not evidence that nobody tried. It is what voluntary coordination looks like after twenty years.