Twenty Years of Buying Public Expertise

Consider a public agency that commissions a transportation study. The consultant delivers the analysis, staff review it, and elected officials adopt a plan. The contract worked. Then conditions change — a subdivision is approved, a bridge closes, growth arrives faster than the forecast — and a different question comes up: can the agency update the analysis itself? That depends on what it kept. Usable data, documented methods, a working model, and people who understand all three. A finished project and a stronger institution are related achievements, but one does not guarantee the other.

That distinction runs through two decades of federal argument about contracting, and the federal record deserves a local planner’s attention because it is the only place the argument has been audited. Between 2003 and 2026 the Government Accountability Office (GAO), the Congressional Research Service (CRS), and the Office of Management and Budget (OMB) asked three successively harder questions about buying services: does it save money, who is in control of the work, and who keeps the knowledge? The first proved hard to answer. The second produced a real policy. The third has never been answered — and artificial intelligence is now making it the only one that matters.

A word about the evidence. Most of what follows comes from GAO reviews of selected agencies and selected contracts. They are case studies, not samples, and they do not show that every contract weakens government; contractors train staff, document methods, and build capabilities that outlast them all the time. What the audits document is a mechanism — a way an agency can buy a service repeatedly without becoming any better at understanding it. Read what follows as a description of that mechanism, not a verdict on consulting.

Does it save money?

In the early 2000s, federal sourcing policy was about competition. The George W. Bush administration made competitive sourcing a management priority, and the 2003 revision of OMB Circular A-76 set procedures for comparing public and private performance of commercial activities, competitions that federal employees could win. A 2012 CRS review described the period as a landscape of overlapping approaches to sourcing rather than a single, consistent choice between public and private provision.[1]

Even the price question proved slippery. In 2008 GAO found that the Department of Labor’s competitive sourcing assessments left out costs — the employee time and expenses of preparing for, conducting, and implementing the competitions themselves — so the savings were overstated.[3] The lesson was not that contractors cost more. It was that competitions have costs regardless of who wins, and a credible comparison has to include the full cost of changing and managing a service arrangement. Hold on to that; it comes back when we reach AI.

Who is in control?

By mid-decade a second concern was getting harder to separate from price: whether government still had enough expertise to direct the work it was buying. In 2007 GAO examined professional and management support contracting at the Department of Homeland Security. More than half of the 117 statements of work it reviewed included services closely supporting inherently governmental functions, and in nine case studies officials had not assessed the risk that government decisions could become dependent on contractor judgments.[2]

Here is how that dependence develops. A public official keeps the authority to approve a recommendation while a contractor defines the problem, selects the assumptions, and identifies the alternatives. Reviewing a finished report gives the agency far less influence than understanding how its conclusions were produced, and contract compliance cannot answer the question this raises: how independent is the public decision maker’s judgment?

The policy response turned toward the workforce. Congress required civilian agencies to develop guidance for considering which contracted work could be brought back inside, and in May 2010 GAO testified that none of the nine agencies it reviewed had met the deadline, citing incomplete information, coordination problems, and trouble recruiting and retaining staff. Twenty-four agencies had launched pilots on their contractor use; the machinery for deciding what belonged inside government was still being built.[4]

The clarification came in September 2011. OFPP Policy Letter 11-01 distinguished inherently governmental functions, which federal employees must perform, from work closely associated with those functions and from functions critical to an agency’s mission. Closely associated work could be contracted with safeguards. Critical functions could involve contractors too — provided the agency kept enough internal capability to maintain control of its mission and operations.[5]

Who keeps the knowledge?

That proviso is the hinge of the whole twenty years. Policy Letter 11-01 recognized that government needs employees who can understand requirements, weigh alternatives, and manage contractor work, and it directed agencies to preserve opportunities for employees to develop those competencies and maintain continuity of knowledge. Permission to contract a function did not remove the need for public expertise; retaining it became, on paper, part of managing the work.[5]

Implementation was another matter. CRS’s 2012 review described workforce decisions tangled up with appropriations, staffing constraints, and inconsistent sourcing requirements.[1] Rebuilding a capability means recruiting people, training them, and sustaining a career path for them, and an agency with contracts that must deliver this year cannot wait for a team that will be ready in three. A sourcing decision made under those conditions tends to reproduce the existing arrangement even when the officials making it would prefer more capacity of their own.

From here the cycle is easy to see. An agency buys expertise because it lacks staff. If the engagement gives its employees little chance to do the underlying work, they gain little experience, and the next procurement starts where the last one did. Whether the cycle runs depends on how the engagement is designed — which is the point.

Two federal efforts showed what designing for retention could look like. GSA launched 18F in March 2014 as a technology team that helped agencies build and buy digital services — specialized expertise shared across government rather than maintained separately in every department, and by its own count 450 projects in ten years.[6] For small agencies that cannot staff every specialty, the organizational model matters as much as the projects. Then in 2016 the Federal Source Code Policy addressed the same problem at the level of the asset: agencies paying for custom software were not consistently making the code available for reuse, so the same thing was being bought more than once. OMB required agencies to obtain sufficient rights for reuse and secure delivery of the code and supporting materials, and set up a three-year pilot to release a portion of new custom code as open source.[7]

The software policy is the precedent for planning. It made continued access to the products of publicly funded work an explicit objective of the purchase. A planning contract can do the same if the deliverable is defined as the documented analytical workflow and the data, not just the report.

Did any of it stick?

Not reliably. In 2020 GAO went back to DHS and examined service contracts requiring heightened management attention. The department was not consistently planning for the government workforce needed to oversee that work, and in six of eight selected contracts officials had not identified specific activities for implementing oversight safeguards. DHS later addressed the recommendations on identifying sensitive functions and oversight activities; the workforce-planning recommendation was still open in GAO’s July 2026 update, DHS disagreed with it, and nothing further had been done.[8]

In 2023 Mariana Mazzucato and Rosie Collington put institutional learning at the center of their critique of consulting in The Big Con, arguing that heavy reliance on consultants erodes an organization’s ability to learn and to build its own capabilities.[9] Their thesis reaches further than any audit can, but it names the shift the audits had been circling for fifteen years: judge a consulting engagement by what happens to the client’s knowledge over time, not only by the usefulness of the advice.

In January 2024 GAO reviewed the Department of Veterans Affairs and found that VA had not fully implemented the 2011 guidance. Contract data were unreliable, guidance and training were incomplete, and workforce planning did not adequately address oversight capacity. VA agreed with seven recommendations.[10] A current management problem traced in a straight line to a policy adopted more than twelve years earlier.

The record shows recurring weaknesses in three places: identifying sensitive work, planning the public workforce, and sustaining informed oversight. It does not say consulting is wasteful or that insourcing everything would save money; surges, rare specialisms, and independent review all justify outside help. The hard case is recurring work central to an agency’s mission, where the question is whether each contract leaves the agency more capable of understanding and directing it — or exactly as dependent as before.

AI arrives, 18F departs

The last two years changed the procurement landscape and sharpened the question at the same time.

A March 2025 executive order consolidated purchasing of common goods and services through GSA and gave it a larger coordinating role in certain government-wide IT acquisitions.[11] An April 2026 order made fixed-price, performance-based contracts the default, with written justification for other contract types, senior approval above set thresholds, and exceptions for emergencies and certain development work.[12] Both address how government organizes its buying. Neither touches the capacity problem, and the second depends on it: a fixed price only disciplines a contractor if the buyer can define an adequate result and judge whether it was delivered. A predictable price for a report says nothing about whether its model is sound or whether anyone on staff can run it next year.

In the same period, GSA eliminated 18F. GAO’s June 2026 cloud-procurement report records the elimination in March 2025 — and records several agencies telling GAO that 18F had helped them with their workforce challenges. The same report found ten of 24 agencies citing difficulty hiring and retaining skilled staff and seventeen reporting trouble tracking their cloud costs.[13] Created in 2014 to share scarce technical capacity across government, 18F was removed eleven years later by the agency that housed it, and the fact is recorded in a report documenting the shortage it was built to relieve. Access to sophisticated commercial technology and the expertise to buy and manage it are different things. Government now has more of the first and less of the second.

AI brings the whole history to a point. OMB’s April 2025 guidance on acquiring AI addressed vendor dependence directly, calling for attention to knowledge transfer, portability, and rights in data and models: access to a powerful system leaves an agency exposed if it cannot move its information, understand the arrangement, or change suppliers.[14] GAO’s April 2026 review of AI acquisitions then found the familiar problems in the new setting. Officials could not get access to technical specialists to evaluate proposals and struggled to understand AI-related costs (the Department of Labor’s accounting lesson, eighteen years on); at the four agencies examined in depth, lessons from AI acquisitions were not being systematically collected; and GAO distinguished agency-directed acquisitions from cases where vendors introduced capabilities before the agency had defined any requirement.[15] That last finding is the 2007 DHS finding with the nouns changed: the supplier helps decide what government believes it needs, and the institution lacks the capacity to evaluate the proposal on its own.

The question for a small agency

None of this binds a metropolitan planning organization. OFPP policy letters govern federal agencies, and nothing above shows that any particular MPO has mismanaged a contract. But an MPO exists because federal law requires one, its planning money is largely federal, and the work program it spends that money through is a federally required document.[16] The questions travel better than the rules do:

• Who can reproduce the analysis once the contract closes?
• Who controls the usable data, in a usable form?
• What can staff do after the project that they could not do before?
• How will the work be updated, and by whom?

They apply whether an agency hires a consultant, employs its own analysts, or uses AI.

As of October 2026 the federal picture is mixed. VA has closed several recommendations after improving its contract inventories and analysis procedures, but GAO’s February 2026 update reported that it had collected information about staff competencies without yet analyzing whether the workforce could actually oversee the work.[10] DHS’s workforce-planning recommendation remained open in July.[8] Better information and procedures are progress. They are not the same as having the people.

So go back to the agency in the first paragraph, the one that wants to update its own transportation study. After twenty years of federal policy that recognized the problem, named it, and wrote it into guidance, the honest answer to “can you do it yourself?” is still: it depends what the contract said. AI may change the cost of doing and maintaining analytical work. It does not change the fact that the capability has to be designed into the engagement, or it will not be there afterward. That is the starting point for the broader argument in The Civil-Industrial Complex: the comparison that matters is not the price of this year’s plan but the capacity the institution will have several years after it is paid for.

Sources

Sources and recommendation status checked on 20 October 2026. Numbered citations and source titles link to the original publications.

[1] Congressional Research Service. Sourcing Policy: Selected Developments and Issues. R42341, L. Elaine Halchin, 7 February 2012.

[2] U.S. Government Accountability Office. Department of Homeland Security: Improved Assessment and Oversight Needed to Manage Risk of Contracting for Selected Services. GAO-07-990, 17 September 2007; publicly released 17 October 2007.

[3] U.S. Government Accountability Office. Department of Labor: Better Cost Assessments and Departmentwide Performance Tracking Are Needed to Effectively Manage Competitive Sourcing Program. GAO-09-14, 21 November 2008.

[4] U.S. Government Accountability Office. Sourcing Policy: Initial Agency Efforts to Balance the Government to Contractor Mix in the Multisector Workforce. GAO-10-744T, 20 May 2010.

[5] Office of Federal Procurement Policy. Policy Letter 11-01: Performance of Inherently Governmental and Critical Functions. Federal Register, 12 September 2011, 76 FR 56227.

[6] U.S. General Services Administration. At 10 years, GSA’s tech consulting team 18F celebrates 450 projects. 19 March 2024.

[7] Office of Management and Budget. Federal Source Code Policy: Achieving Efficiency, Transparency, and Innovation through Reusable and Open Source Software. Memorandum M-16-21, 8 August 2016.

[8] U.S. Government Accountability Office. DHS Service Contracts: Increased Oversight Needed to Reduce the Risk Associated with Contractors Performing Certain Functions. GAO-20-417, 7 May 2020; recommendation updates include July 2026.

[9] Mariana Mazzucato and Rosie Collington. The Big Con: How the Consulting Industry Weakens Our Businesses, Infantilizes Our Governments, and Warps Our Economies. Penguin, 2023. See also the authors’ UCL Institute for Innovation and Public Purpose book talk, June 2023.

[10] U.S. Government Accountability Office. VA Acquisition Management: Oversight of Service Contracts Needing Heightened Management Attention Could Be Improved. GAO-24-106312, 25 January 2024; recommendation updates include February 2026.

[11] The White House. Eliminating Waste and Saving Taxpayer Dollars by Consolidating Procurement. Executive Order 14240, 20 March 2025.

[12] The White House. Promoting Efficiency, Accountability, and Performance in Federal Contracting. Executive Order 14402, 30 April 2026.

[13] U.S. Government Accountability Office. Cloud Computing: Federal Government Needs to Address Procurement Challenges. GAO-26-107530, 23 June 2026; see footnote 193 on 18F.

[14] Office of Management and Budget. Driving Efficient Acquisition of Artificial Intelligence in Government. Memorandum M-25-22, 3 April 2025.

[15] U.S. Government Accountability Office. Artificial Intelligence Acquisitions: Agencies Should Collect and Apply Lessons Learned to Improve Future Procurements. GAO-26-107859, 13 April 2026.

[16] 23 CFR Part 450, Subpart C — Metropolitan Transportation Planning and Programming, including § 450.308 (funding and unified planning work programs) and § 450.310 (MPO designation). eCFR.