What Procurement Is For?
Everyone knows it's broken. The architecture is doing exactly what it was designed to do.
Public procurement is one of the largest machines governments possess for shaping markets and capability. An estimated $13 trillion runs through procurement systems globally every year. The scale of the instrument is settled. Its purpose isn’t.
The system is failing at the work governments increasingly ask it to do. Everyone in the system knows this. The European Court of Auditors documented a lost decade. The OECD has been writing about it for fifteen years. There is a thirty-year academic library on the subject. The institutional reform community has converged on the diagnosis. Reform packages are issued every five years like a liturgy. Compliance frameworks. Strategic action plans. Capability-building programmes for procurement officers. Little changes.
The visible failure is bad procurement. The deeper failure is what procurement is now actually for.
I. The three assumptions nobody questions
Public procurement rests on three foundational assumptions: outputs are specifiable before procurement begins. Markets exist to supply them. Competition on stated criteria reveals value.
For governments buying trucks and uniforms, all three hold. For governments commissioning solutions that don’t yet exist, building capability in institutions that don’t yet have it, or shaping markets that haven’t yet formed, all three collapse simultaneously.
Richard Nelson and Sidney Winter established this in 1982. Innovation operates under genuine uncertainty: search, selection and routines. Optimal suppliers cannot know they are optimal before they try. Solutions are discovered through the process, not in advance of it. Edler and Georghiou made the institutional case in 2007. Mazzucato, Spanó and Doyle updated it for Brazil in 2024. The argument has not moved in two decades because the framework hasn’t either. A governing logic that cannot be amended by its own evidence is a defence mechanism with footnotes.
II. The decade the evidence produced
Procurement is for converting public money into public capability. The lost decade was a decade in which it stopped doing that.
The European Court of Auditors’ Special Report 28/2023 covers EU procurement between 2011 and 2021. The decade after the financial crisis. The decade of directive reforms. The decade in which everything was supposed to get better. It got worse.
Single-bid procedures nearly doubled, from 23.5 percent to 41.8 percent. Average bidders per contract fell from 5.7 to 3.2. Award procedures grew longer, not shorter - from 62.5 days to 96.4. In eight member states, more than 80 percent of contracts went to lowest price alone. Direct cross-border procurement rose from about 2 percent before the single market to about 5 percent, where it has stayed across the 2010s.
The 2014 directive reforms had no demonstrable effect on the level of single bidding or direct awards. They had effects. None of them were the ones intended. The Competitiveness Council responded in May 2024 by calling for an EU-wide strategic action plan. The same response made in 2014. The same response that produced the lost decade.
The absurdity inside evaluation panels is built in. A handful of officials sit in a room reading bid documentation for work that has not been done before, in markets that do not yet exist, by suppliers whose potential capability is under scrutiny. Most of them have never built anything under genuine uncertainty in their professional lives. They are scoring people who have, against criteria written by people who haven’t, on behalf of institutions that reward the criteria over the capability.
The matrix is not a tool for finding the best bid. It is a tool for producing a defensible record of having chosen one. Expertise the panel does not possess is laundered through criteria the panel can apply. The decision looks technical. It is procedural. Innovation is, by definition, new. The system rewards the old. Nobody got fired for hiring McKinsey.
III. The procurement that excluded the ecosystem it claimed to serve
Earlier this year my firm bid on an EU-funded tender run by an African institution, for an assessment of digital entrepreneurship and innovation regulatory environments across multiple African countries. The eligibility criteria (not stated accurately in the terms of reference), applied to the prime contractor, would have excluded virtually every firm headquartered in any of the target countries.
What was left was the Big Four and a handful of legacy international development consultancies. The contract went to one of them - a firm publicly identified by its own staff as a strategic partner of the contracting institution. The work that comes back from these awards is generic. The local ecosystem watches it happen. Same load. Same cycle.
If you have ever bid into a public-sector tender that excluded you on a ten-year track record requirement in a field that did not exist ten years ago, you already know everything in this piece. The LinkedIn page “Crappy Funding Practices” is full of similar stories. None of them surprises anyone who has been bidding for long. Dejection. Then frustration. Then wondering whether to bid again. Until you don’t.
IV. Why nothing changes
The standard reform account says officials need better training. The implication is that the system is malfunctioning. It is not. It is stable at the equilibrium its incentive architecture demands.
The first layer is leverage. Donella Meadows mapped twelve places to intervene in complex systems, ranked by power to produce change. Tweaking parameters - thresholds, weights, criteria - sits near the bottom. Changing the goals or paradigm of the system sits at the top. Every procurement reform since 1994 has been a parameter intervention. Thirty years of reform at the weakest leverage point in the system.
An objection deserves to be taken seriously. The procurement architecture was not built to fail. It was built to prevent ministers awarding contracts to their friends, to prevent state capture, to prevent the conversion of public money into private fortunes through opaque award. That problem is real. In many jurisdictions the architecture is still the only thing standing between public money and capture. Procurement officers who insist on the matrix are not stupid. They are responding to a problem audit institutions exist to address. The question is not whether that problem matters. It does.
The question is whether the cost - capability never built, markets that never form, technical expertise the state has spent thirty years not developing - is one we have ever priced. The reform conversation has presumed for thirty years that the downside is acceptable because the alternative is corruption. That presumption now needs auditing on its own terms.
The second layer is the loops. Four reinforcing mechanisms sustain the architecture, each compounding the others. The first is incumbency-reproduction: rules drafted with reference to firms that already exist, panels staffed by people who came through those firms, awards going to firms whose corporate histories match the rules. The second runs through audit. Audit defines what is auditable. Procurement produces what audit can confirm. The unit of accountability becomes the file rather than the outcome. The third absorbs reform - each new package adds a compliance layer that strengthens the architecture it claims to challenge. The fourth absorbs the workarounds: labs, challenge funds, sandboxes, each treated as a discrete pilot rather than precedent. The fifteenth lab opens. The fifteenth one is also absorbed.
These loops stabilise because the procurement function consolidates funding decisions, rule-setting, evaluation, certification of legitimate suppliers, and audit of outcomes inside a narrow institutional band. The procurement officer who writes the terms of reference scores the bids, signs the contract, monitors delivery, and signs the closure document. The firms that win contracts advise on the design of the next competition. The audit body that signs off the file uses the framework the procurement officer was meant to comply with. This is role collapse, and it is what holds the loops in place.
The third layer is concentration. Procurement systems do more than fail to prevent it. They actively reproduce it. Mariana Mazzucato and Rosie Collington documented the dynamic at full length in The Big Con (2023): three decades of consulting contracts have hollowed out public-sector capability and concentrated state expertise in a handful of firms. The procurement framework is the operational expression of that hollowing-out. The Big Four win because the rules were drafted by people whose institutional reference points are firms that look like the Big Four. Those rules are scored by panels whose members have never built anything outside an institution that looks like the Big Four. On behalf of buyers whose own career incentives reward the appointment of firms that look like the Big Four.
Each of these advantages on its own can be argued. Bundled together - rules, audit, relationships, language register, bid-preparation capacity, and the next competition’s pre-positioning all aligned around the same set of firms - they produce a structural advantage no incoming supplier can contest. Bundled advantage is how incumbency reproduces. The UN system runs the same play. I have written about it elsewhere. Adjusting the scoring weights inside an architecture designed to reproduce its own incumbents is exactly as effective as it sounds.
The institutional reform community has reached parallel conclusions. Leonardo Quattrucci’s 2025 TIAL white paper, citing Jennifer Pahlka, frames public procurement as “compliance with absurdity” and identifies seven recurrent institutional fallacies that produce static markets, lock-in to incumbents, and labyrinthine requirements that exclude small firms. The diagnosis is now mainstream. What follows from it politically is not.
V. The instruments that hold it in place
The architecture has operational instruments. The reform literature treats them as technical. They are not. They are political.
The logframe locks design into a linear chain - input, output, outcome, impact - that assumes the answer is known at the start. Des Gasper’s 2000 critique named it “lock-frame”: once written, the matrix becomes resistant to revision even when conditions change. The Doing Development Differently manifesto, signed in 2014 by more than 400 development practitioners and scholars, named the same problem: logframes “incentivise rigid, linear program logic, which does not reflect reality” and “make it difficult for programme managers to adapt to changing circumstances.” Eleven years on, the logframe is still required by the World Bank, the EU, FCDO and every major UN agency.
Lowest-price-wins is the explicit award criterion in 55 percent of EU public procurement procedures, according to the European Commission’s own data. In eight member states, more than 80 percent of contracts are awarded on price alone. The 2014 directive introduced the Most Economically Advantageous Tender criterion specifically to allow public buyers to factor in quality, sustainability and innovation. The proportion of contracts using lowest price as the sole criterion has barely moved.
Programme cycles run one to two years for most bilateral and multilateral procurement. Anything that would actually compound - institutional capability, market formation, ecosystem development, behavioural change - sits structurally outside the window. By the time the work would begin to deliver, the programme has closed, the team has dispersed, and the next cycle is buying something different from somebody else. The Thinking and Working Politically community of practice has been documenting this since 2013. The donor architecture has not adapted.
Each instrument was designed for a reason. Each reason was correct in 1995. Together they produce a procurement architecture that cannot commission anything that takes longer than two years, costs more than the squeeze allows, deviates from the logframe, wins on anything other than price, or comes from outside the small set of suppliers the architecture has already pre-cleared.
The workaround economy is what you get after thirty years of trying. Innovation labs. Challenge funds. Accelerators. Sandboxes. Each one a pressure valve, an attempt to route innovation through a bypass while leaving the main road unreformed. The lab opens. The pilot succeeds and doesn’t scale. The champion moves on. The system absorbs it.
The brochure is always handsome. The reform never arrives. Each lab is a programme. None of them is a capability. Programme-rich, capability-thin - the structural failure of procurement reform across thirty years, and the condition Section VIII’s corrections are designed to reverse.
VI. AI is making it worse
Procurement is for converting public money into public capability. AI is the test the system is failing in real time. In 2025, the UK government awarded 521 public contracts for AI worth £1.17 billion - more than double the value awarded the previous year. The accountability regime processing this is the same one that produced the lost decade.
The pathology runs on both sides of the table. On the supplier side, AI bid-writing platforms - AutogenAI, documented in the OECD’s 2025 Governing with Artificial Intelligence report; Procurement Sciences, $30 million Series B in November 2025, 300+ contractor clients - sit on top of foundation models trained on the public corpus of government writing. That corpus is decades of incumbent output. Smaller firms using these tools draft in a stylistic register the underlying model learned from incumbent material. On the buyer side, AI-assisted procurement assessment runs on the same kind of model, calibrated to the same register. The two industrialisations meet in the middle. The system is being industrialised on both sides at speeds the procurement officer cannot match.
Sir Geoff Mulgan, who teaches a course on AI in government at UCL, posted on LinkedIn in September 2025 in response to the FT’s reporting on the spend. Much of it was justified, he wrote. What was not justified was the apparent absence of four specific things: any serious evidence assessment of what does and doesn’t work, any open data to enable such assessment, any skin in the game for contractors so that payment depends on delivery, and any serious comparison of the deals other governments are getting alongside attempts at pooled purchasing. He invited correction. If the evaluation, the open data and the pooled purchasing arrangements existed, he wanted to be pointed to them. Absent them, in his phrase, what the UK government might be doing was faith-based procurement, approved by ministers with little tech background, egged on by vendors.
Pushback came from inside government. The Head of Data Science and the Chief Data and AI Officer at the Department for Business and Trade pointed to a 49-page departmental evaluation of its Microsoft 365 Copilot pilot, published 28 August 2025. The evaluation is real. It is also one product, one department, one three-month window, described in its own methodology as “evidence-without-a-counterfactual,” reporting high user satisfaction but no productivity improvement, with the value-for-money assessment explicitly deferred and the environmental impact methodology still under development. None of which addresses Mulgan’s points on open data, contractor accountability, or cross-government comparison and pooled purchasing. The most carefully prepared rebuttal to the critique substantiates it.
The end-state is already visible in adjacent territory. EU Marie Skłodowska-Curie postdoctoral fellowship applications grew from 7,044 in 2022 to 17,066 in 2025 - a 142 percent increase. Proposals drafted by paid consultants using AI routinely achieve near-perfect scores. Research Commissioner Ekaterina Zaharieva is now exploring tie-breaker lotteries among proposals above the quality threshold for the next Horizon Europe programme. Research funding is not procurement. But the architectural failure is identical: a selection system that admits it cannot distinguish quality among AI-assisted submissions, and defaults to the next-cheapest defensible option - assign awards by chance.
The Commission is considering randomisation because the alternative - building evaluators with the standing to read past AI-polished surfaces, holding officials accountable for the assessments they make, governing judgment rather than depersonalising it - is the one move the system structurally cannot take. The pretence of judgment is being formally retired in the one corner of the EU’s expert-evaluation architecture where the volume problem became impossible to ignore. The procurement architecture is next.
VII. What the frontier already proves is possible
Working models exist. They are rare, fragile, treated as exceptions rather than precedent. They should be treated as precedent. The frontier cases work because they change the unit of control: from compliance with a fixed brief to governed search under uncertainty. They cluster into three patterns. The patterns matter, because the conditions under which procurement can be reformed are different from the conditions under which it must be bypassed.
Exemption. The UK’s Advanced Research and Invention Agency was given statutory exemption from the Public Contracts Regulations 2015 and an initial £800 million budget, raised to at least £1 billion through 2029 in the June 2025 Spending Review. Programme Directors run roughly £50 million budgets with two-signature approval, commissioning research, taking equity, funding prizes. Gavi’s pneumococcal Advance Market Commitment committed $1.5 billion to guarantee a market for newly licensed pneumococcal vaccines in low-income countries - operating outside standard tender mechanisms via a pull instrument. The ten-to-fifteen year lag between high-income and low-income vaccine introduction was closed inside a year. The lesson is uncomfortable: these worked because they were granted formal carve-outs from the procurement architecture, not because the architecture was reformed to accommodate them.
Architecture. Rwanda’s Umucyo connects 1,194 government agencies, 13,800 suppliers, and 24 government information systems - digital procurement infrastructure built coherently rather than the existing system digitised piecemeal. Joyeuse Uwingeneye led Rwanda’s Public Procurement Authority through September 2025 and chaired the African Public Procurement Network - open to all 54 African states, with 43 currently active - through 2024. Ukraine’s ProZorro consolidated PPE demand at the start of COVID because the open-data architecture existed before the emergency. The EU’s Pre-Commercial Procurement instrument funds R&D across multiple suppliers in three competitive stages, risk shared at market conditions, no purchase commitment until results are demonstrated. Over 600 PCPs have been launched across Europe. PCP is used a fraction as often as it should be, because it requires officials to exercise judgment at every gate - but where it is used, it works.
Capability. Finland built capability among public buyers through KEINO, its Competence Centre for Sustainable and Innovative Public Procurement (operations ended in 2024). It sat, with Norway and Estonia, among Europe’s top performers on innovation procurement on the European Commission’s 2024 horizontal policies benchmark. Estonia digitalised 100 percent of its government services by 2024, built on principles embedded in law and code: once-only data, sovereignty, transparency. ARIA’s Programme Directors sit in this bucket as well as the exemption bucket: individuals selected for the standing to exercise judgment, granted the legal scaffolding to do so. Most jurisdictions did the reverse - set targets first, retrofit capability into a calcified architecture, produce compliance theatre.
Three patterns. None mutually exclusive. None exotic. None requiring new technology or new economics. Each was a political choice, made once, not made elsewhere. Kattel and Mazzucato name what these cases share: dynamic capability, the institutional ability to sense what the market is revealing and reconfigure when the evidence shifts. The deeper lesson is the one the reform literature elides. Asking how to make our procurement system more like ARIA’s is the wrong question. ARIA’s defining feature is that it isn’t subject to a procurement system at all.
This is what a stable system looks like when it is asked to do work it was not designed to do. The exemptions are the architecture’s tacit admission that its own logic does not survive contact with the task. Where the work is genuinely uncertain, the system spawns carve-outs and calls them exceptions. The exception framing is what keeps the architecture intact. It comes apart the moment somebody treats one of those exceptions as precedent and acts on it.
VIII. The five corrections
The corrections are not conceptual. Each is operational somewhere right now. Together they would change what procurement is for.
Buy the problem, not the solution. Stop specifying the answer. Define the question, the performance criteria, the success conditions, and pay whoever can satisfy them. Gavi did this with vaccines. ARIA does this with research. The instrument is older than DARPA. There is no excuse for not using it.
Run the contest in stages. Stop awarding under uncertainty in a single round. Fund five suppliers to design. Three to prototype. One to deliver. Reduce at each gate based on actual evidence rather than predicted capability. PCP runs this way. Pay small in stages for the contribution. The single-round award under uncertainty is the cheapest mechanism in compliance terms and the most expensive in outcome terms.
Make judgment legible. Drop the matrix as the decision. Use it to structure argument, not to replace it. Named officials. Reasoned, documented decisions. Accountable for outcomes, not for process compliance. Sabel and Zeitlin showed what this looks like at scale: framework goals, implementer autonomy, peer review of outcomes, periodic revision as evidence accumulates. The matrix produces the appearance of reasoning. This produces the substance.
Audit the portfolio. Stop scoring contracts one at a time. Score the portfolio. The current architecture records four failures and one breakthrough as four failures and one anomaly. The breakthrough disappears. Until the unit of audit shifts, instruments designed for uncertain work will keep failing the tests designed for routine work. This is the most consequential change available, and the one most resisted by audit institutions whose authority depends on contract-level scrutiny.
Build the institution before the instrument. Capability does not retrofit. Finland’s KEINO existed before the 10 percent target. ARIA’s exemptions were granted before the first Programme Director was hired. Umucyo was operational before it became continental infrastructure. Every challenge fund that failed to scale did the reverse: deployed the instrument first, tried to retrofit capability into a calcified architecture, failed.
Five corrections. None radical. None expensive. Transferable. The logic works in any jurisdiction. The reason they are not in place is political.
IX. A position
What has not happened - in any country that signed up to the plurilateral WTO Agreement on Government Procurement (22 parties covering 49 WTO members), in any multilateral framework, or in any bilateral donor architecture - is the political act of asking the question this piece poses. The UK’s Procurement Act 2023, which took effect in February 2025, left the architectural premises intact: same logframe-derived assumptions, same lowest-price defaults, same compliance-track audit. The EU’s response to the lost decade documented by its own Court of Auditors has been another strategic action plan. Germany’s Public Procurement Acceleration Act (August 2025), Canada’s Procurement Improvement Action Plan (March 2024), and Australia’s revised Commonwealth Procurement Rules (July 2024) are the most recent reform vehicles in three OECD jurisdictions. None has acted on what it would actually take.
Procurement is currently being used to convert public money into the institutional reproduction of a settled distribution of who gets to do public sector work. The architecture protects that distribution. The audit institutions police it. The reform packages reinforce it. The workaround economy disguises it. AI is industrialising it.
What follows is what could change.
Stop.
Stop pretending logframes are neutral planning instruments.
Stop using lowest-price-wins as the default decision rule for anything except commodity goods.
Stop awarding contracts on ten-year track records in fields that did not exist ten years ago.
Stop running competitions with one bidder and calling them competitive.
Stop pretending evaluation panels have technical expertise they don’t have.
Stop treating exemption as reform.
Stop responding to the lost decade by calling for the next strategic action plan.
Start.
Start auditing the portfolio rather than the contract.
Start pricing the cost of the architecture: capability never built, markets never formed, technical expertise the state has spent thirty years not developing.
Start treating procurement as the political instrument it is.
Start naming bundled advantage as the design feature it is.
Start unbundling the roles the procurement function has collapsed - rule-setting, evaluation, certification, audit.
Do more.
More statutory exemption for work the architecture cannot commission.
More pull commitments for outcomes that do not yet exist.
More staged procurement under uncertainty.
More investment in the capability of the buyers, not the suppliers.
Do less.
Less reliance on consulting contracts to substitute for state capability.
Less faith in labs, sandboxes and challenge funds the architecture is built to absorb.
Less reform-package theatre.
The next lost decade is already underway. It will produce more reform packages, more compliance frameworks, more capability programmes, more strategic action plans. None will work. None is meant to.
Procurement is a choice. The choice that has been made is to keep the architecture intact. That choice can be unmade. Exemption is not unmaking. ARIA, Gavi, PCP, KEINO - the architecture granted those carve-outs and survived them. Whoever unmakes it instead decides what the next architecture is for. The diagnosis is settled. What is missing is the political act.