The phenomenon of “hallucination” occurs when a large AI language model generates text that appears authoritative despite being entirely fabricated. Because these outputs appear flawless, both identifying and correcting the errors present a significant professional challenge as hallucinations are hard to catch in the first place, let alone correct.

The seminal framework for subsequent enforcement remains Mata v Avianca (SDNY, June 2023), the case that first exposed the professional hazards of automated fabrication. In that instance, legal counsel submitted filings containing six fictitious citations and fabricated judicial excerpts provided by ChatGPT. While the $5,000 penalty was comparatively modest, the ruling fixed the baseline for professional accountability, confirming that the obligation to verify remains a professional duty that cannot be transferred to technology.

By 2026, the initial judiciary tolerance has been replaced by a significant escalation in punitive measures designed to address this increasing problem:

• Rising Punitive Costs: Judiciary responses have sharpened into significant financial deterrents. In March 2026, the Sixth Circuit imposed personal fines of $15,000 on two legal professionals, alongside double costs and additional fees, for filings riddled with fictitious citations; the court observed that previous, lighter penalties had not effectively discouraged such professional failings.

• Licensure and Practice Suspensions: The courts now categorise automated fabrication as a professional breach of duty. By June 2026, the Ninth Circuit issued $2,500 sanctions and six-month practice bans against two attorneys who sought to characterise their reliance on fabricated citations as simple clerical or typographical errors.

• Total Procedural Collapse: Chronic reliance on unverified material has led to unprecedented structural consequences. In April 2026, the Alabama Supreme Court penalised a practitioner who utilised AI to draft a formal apology for a previous fabrication, only for the apology itself to contain further non-existent legal references. This culminated in June 2026 when a Mississippi judge was forced to remove counsel from both sides and abandon a trial entirely after determining that fabricated content had compromised the integrity of all the party submissions.

The Number That Should Stop You

The AI Hallucination Cases Database, maintained by legal academic Damien Charlotin and now the most cited tracker in this field, had logged, by early July 2026, 1,694 court and tribunal decisions worldwide, each one a case where a court or tribunal engaged with hallucinated material, not merely where AI use was alleged.

The composition of that number has shifted too. In 2023, roughly seven in ten flagged cases involved self-represented litigants, people without legal training, using a free tool and trusting it. Across the full database by July 2026, lawyers were the party at fault in close to four in ten decisions and rising. Increasingly, the people with a professional duty are not verifying information.

Hallucinations Are Spreading to the UK

The same database lists, in early July 2026, 59 documented UK hallucination cases. The most notable is R (Ayinde) v London Borough of Haringey, where a barrister and solicitor submitted five fabricated cases to the High Court in a judicial review. The court imposed a wasted costs order and referred both to their professional regulators.

The incidents of AI hallucination cases have reached the largest UK legal firms. In a judgment handed down on 22 May 2026 (Re an Office-Holder; Cork v Smith), a Judge delivered a public admonishment of a junior solicitor, for using AI to research a straightforward procedural question, which cited a fabricated insolvency rule. Although the judge accepted there had been no deliberate attempt to mislead and stopped short of contempt proceedings, the firm referred itself and three solicitors to the Solicitors Regulation Authority and the published judgment stands as the sanction.

Recent, high-profile publications from Big 4 accounting firms have also attracted uncomfortable public scrutiny, prompting withdrawals or significant fee refunds, after independent fact checkers identified that citations and claims were fabricated. These reports attributed specific AI deployments or academic research to organisations that subsequently disputed the claims. The incidents share the same pattern as the court sanctions. Professional content, produced with AI assistance, where verification did not happen. This is not a junior lawyer under deadline pressure; these are the most process driven firms in professional services, where clients took the content at face value until someone outside the firm checked.

The Same Failure Is Reaching Procurement

For procurement this is no longer a hazard in waiting. In the United States, where procurement disputes go to specialist tribunals, the fabrications have already arrived, but so far are limited to bidder challenges. The US Government Accountability Office threw out four challenges from one bidder in a single ruling, calling its reliance on invented case law an abuse of process, and the Armed Services Board struck out a brief after finding more than seven in ten of its citations were false.

In the UK, on the buyer’s side, one of the higher risk stages of the above-threshold public procurement lifecycle, is at assessment. This is where a defined process is followed to select the successful bidder and the agreed scores, narratives and outcome are provided to bidders in an Assessment Summary.

Where a bidder has the statutory right to the Assessment Summary and a defined challenge process, the consequences of an AI hallucination can be both reputational and financially costly for a public organisation.

And the potential use of language models in procurement is wide, from developing procurement strategy, meeting transcripts, to the monthly calculation of contract performance deductions. But hallucination exposure risk is not evenly distributed across the procurement lifecycle. The higher risks are concentrated where three conditions meet: legal consequence, third-party reliance and an authoritative source, such as a public body, regulated firm, court of law, or a statutory regulated profession, from which information is by default trusted.

When a language model is then used to interpret, rather than reproduce, its inherent fluency obscures the boundary between accuracy and invention. Without rigorous verification, these assertions, regardless of their veracity, are accepted as authoritative fact.

The Rules Are Forming Faster Than The Legislation

Courts have not waited for legislators. In the US, over 30 federal district courts had adopted some form of AI disclosure requirement by April 2026, covering three things: disclose whether AI was used, name the tool, and certify that a human has personally verified every citation and factual claim.

In England and Wales, the Civil Justice Council opened an eight-week consultation in February 2026 on whether formal rules are needed to govern AI use in preparing court documents. As of early July 2026 the Council had published an update on the consultation responses but not its final report, which is expected later in the year. The direction of travel so far points toward relying on existing professional standards rather than new rules for professionally drafted documents.

What Good Governance Looks Like

Within the public sector, the need for robust governance is now; waiting for a legal challenge to expose systemic failings is a significant risk. The approach must be risk-centric rather than technology-driven. In scenarios where legal consequence, third-party reliance and authoritative sourcing converge, pre-publication verification should be non-negotiable and the duty of accuracy rest with the author. For lower risk outputs, a less intensive review may be sufficient. However, the fundamental principle remains: all model-generated content is unverified by default. Sophisticated linguistic fluency frequently masks factual fabrication. The governance framework should clearly distinguish between those who conduct the screening and those who provide final approval.

Organisations are committing multi-million pound investments in AI, chasing efficiency and competitive advantage. That deployment needs a parallel level of investment in governance, defined ownership and measurable outcomes, or the efficiency it buys is only exposure that has not yet surfaced.

The questions worth asking, whatever the source of the output, are:

• Verify the source: has every citation and factual claim been checked against a primary source, not simply accepted?

• Own the process: is the use of AI transparent, and could you defend the method that produced the content?

• Validate the record: for meeting notes and transcripts, has someone, who was present, confirmed it reflects what was said?

• Assess the risk: if the content turns out to be fabricated, what happens to the decision, the contract or the reputation resting on it?

Write the policy before an incident writes it for you. If the honest answer to any of those is “we assumed it was fine”, the exposure already exists.

____________________________________________________________________________

Sources

• AI Hallucination Cases Database, Damien Charlotin (figures as at update of 2 July 2026) Inclusion test: decisions that address the use of AI in more than a passing reference and where the court or tribunal has found, or implied, that a party relied on hallucinated material; mere allegations are excluded, though the author includes some alleged but unconfirmed cases as a judgement call.

• Mata v. Avianca, Inc., case background

• Lawyer Sanctioned for Failure to Catch AI “Hallucination”, American Bar Association (Rule 11 certification and duty of candour)

• Federal judge hands down $110K penalty against 2 lawyers for AI errors in court documents, ABA Journal (Oregon)

• Brigandi Case: How a $110,000 AI Hallucination Sanction Rewrites Risk, CorePro (Oregon, single-attorney framing)

• Sixth Circuit Sanctions Attorneys for Fake Citations, What Does This Mean for Use of AI?, National Law Review (Sixth Circuit, $30,000)

• US Appeals Court Fines Lawyers $30K Over AI Filing Errors, JD Journal (Sixth Circuit)

• Two O.C. Attorneys Suspended Relating to AI Fabricated Cites, Metropolitan News-Enterprise (Ninth Circuit)

• US judge stops case after lawyers on both sides cite AI-hallucinated cases, Legal Cheek (Mississippi)

• Alabama Justices Toss Case Over ‘Egregious’ Attorney AI Misuse, Bloomberg Law (Ibach v Stewart)

• Would you hire the lawyer who just got sanctioned for using AI?, Fortune

• UK AI Hallucination Cases, Natural and Artificial Law

• Verify Before You File: UK High Court on AI Hallucinations, Unverified Submissions and Solicitors’ Duty to the Court, SCC Online

• Law firm Pinsent Masons and three solicitors referred to SRA after ‘astonishing’ AI failures, Law Society Gazette

• Judge admonishes Pinsent Masons over misleading letters drafted by junior lawyer using AI, The Global Legal Post

• KPMG pulls report on AI usage due to apparent hallucinations, TechCrunch

• KPMG’s AI report becomes an accidental demo of AI hallucinations, The Register

• Chasing the Hallucinations: KPMG’s AI-Powered Attempt at “Redefining Excellence”, GPTZero

• Deloitte was caught using AI in $290,000 report to help the Australian government, Fortune

• Gen-AI Misuse in Procurement Litigation, Burr & Forman (David P. J. Timm), 6 January 2026 (US procurement-tribunal AI cases: GAO, COFC, ASBCA)

• When AI Backfires: GAO Dismisses Protests Over Fake Citations, Centre Law & Consulting (GAO Oready dismissal)

• Guidance: Assessment Summaries, GOV.UK (Procurement Act 2023, above-threshold assessment summaries)

• Guidance: Contract Award Notices and Standstill, GOV.UK (standstill period and below-threshold position)

• Civil Justice Council publishes update on consultation responses regarding use of AI for preparing court documents, Herbert Smith Freehills Kramer

• Use of AI in preparing court documents, Courts and Tribunals Judiciary (CJC)

• AI Disclosure Requirements for Lawyers: What Courts Require in 2026

Background only, not directly cited: Ironclad Report Finds 99% of Lawyers Trust Using AI As Their Roles Expand, PR Newswire

The pattern across every case was similar. A company described its product as AI-powered. Consumers or investors paid a premium on the basis of that claim. The AI either did not exist, did not have the capability promised, or was — in the most striking case — manual human labour dressed up as automation.

Albert Saniger, the founder and former CEO of Nate Inc, claimed his shopping app automated above 90 per cent of transactions. The actual automation rate was essentially zero. Overseas contractors were completing purchases manually. Investors had committed $42 million on the strength of his AI capability claims. In April 2025 the Department of Justice filed criminal charges against Saniger personally — the first criminal prosecution for AI washing — for wire fraud and securities fraud.

AI technology commands a price premium. In every US case the AI label was being used to extract money — from consumers or from investors — based on false product claims.

This AI washing is not uniquely an American problem.

The two problems that AI washing creates

By “AI washing” I mean selling a price premium on the strength of AI claims that are not substantively true, or are not substantively relevant.

Type 1: Misrepresentation. A product labelled as AI-powered when the underlying capability is something materially different — standard automation, rules-based processing, a workflow tool, or in Nate’s case, human beings. The buyer pays a premium for AI capability and receives something else: the basis of the enforcement action in the United States.

Type 2: Bolt-on AI premium. This is more subtle and arguably becoming more prevalent in the UK. A supplier takes an existing platform — a contract management system, a procurement tool, a supplier performance dashboard — and adds an AI integration. This might be an API link to a large language model such as GPT-4, applied to one part of the workflow. It might be a chatbot layer over existing data. The product is now described as AI-powered or AI-native. The underlying system however has not changed and the price has been revised upward to reflect the AI label rather than any material change in capability.

Neither version is a properly developed AI system. And the difference between those and a genuinely capable AI tool matters considerably — commercially and contractually.

What a properly developed AI system looks like

A procurable AI capability is defined less by the label and more by whether it has measurable, testable performance claims, documented limitations, human oversight controls, and an audit trail. It is defined by what it can do compared to a conventional workflow tool — delivering a material capability uplift, not simply executing fixed rules. It is also defined by whether it has demonstrable capability in compliance and commercial terms, not just technical ones. If it has none of these, a supplier is not in a position to justify a premium price.

The DoNotPay case illustrates the distance between claim and reality at the other end of the scale. The company marketed itself as the world’s first robot lawyer, promising to generate perfectly valid legal documents and replace the $200 billion legal industry with artificial intelligence. The FTC found that the company had not conducted any testing to determine whether its AI output was equivalent to a human lawyer’s work, and had not retained any lawyers to assess it. The product could not deliver what was claimed, at any level.

Most procurement teams will not encounter fraudulent products such as Nate Inc. But they may encounter an AI-labelled product that does not meet the standard of a fully developed system, and where the price does not reflect that gap.

Why public sector procurement is especially exposed

UK public bodies are potentially more exposed than other sectors as they are under considerable financial pressure. The Government has tied £2 billion of AI investment to a target of £14 billion in efficiency savings by 2028/29. AI is not being considered as a solution — it is being actively pursued as one, with savings targets attached. Alongside the efficiency imperative sits a second pressure: the Government’s ambition to position the UK as a global leader in AI adoption, with public sector procurement expected to demonstrate that commitment in practice. Buyers are therefore being pushed from two directions — to cut costs through AI and to be seen embracing it — and that combination creates exactly the conditions in which AI washing thrives: buyers motivated to find solutions, moving faster than their ability to verify the product.

In the United States, enforcement agencies are now actively pursuing suppliers on behalf of buyers harmed by false AI claims. In the UK, existing regulators can act on misleading claims, but there is no single, FTC-style programme focused specifically on AI marketing claims and buyers cannot assume it will protect them in time.

Three factors make the sector particularly vulnerable.

The first is the weak buyer knowledge. A lack of AI technical knowledge within buying teams makes it difficult to verify product claims — and suppliers know that procurement teams are under pressure to modernise and demonstrate savings from technology investment.

The second is the lack of disclosure. A May 2026 techUK report on AI procurement found that supplier disclosures about AI capability are largely inadequate, and that buyers currently lack either the leverage or the technical knowledge to demand better. More significantly, the report found that AI is regularly being activated within existing contracted products — sometimes without buyer knowledge and frequently without any contractual provision to govern it. Organisations are discovering that systems they purchased for one purpose are now making or influencing decisions through an AI layer they did not commission, did not specify, and cannot audit.

The third is in contract management. AI systems present a challenge not found in conventional software. Model drift — the tendency of AI behaviour and performance to change over time as the system processes new data — means that what was delivered at contract start may not be what is operating a year later. The techUK report found that questions about model drift, transparency and liability are going unanswered in procurement conversations. Governance frameworks designed for conventional software are poorly suited to managing AI procured from third parties.

The practical test

The key questions are both commercial and technical. A supplier who genuinely understands their product should be able to explain what the AI does, how it does it, what the outputs are, and any risks and behaviour changes over time. Suppliers relying on a general-purpose language model API will struggle with those questions — not necessarily out of dishonesty, but because they are dependent on third-party products they do not fully control.

But the questions also go beyond the technical. They touch on contractual accountability, organisational risk, and the buyer’s own governance responsibilities. A capable supplier and a well-managed implementation are not the same thing. Buyers should not assume that purchasing a sophisticated AI product resolves the governance question — it only begins the conversation.

Before procurement and at contract review:

If a supplier cannot answer these questions clearly, the product may still be useful. It may even represent good value. But it should not be priced or procured as a sophisticated AI system and the governance applied to its outputs should reflect what it actually is.

The broader point

The US enforcement agencies acted because consumers and investors were losing money on the basis of claims that could not be substantiated. The legal framework used — consumer protection law, securities law — was not written for AI. It was written to stop deception. The AI label simply became the latest vehicle for a familiar kind of commercial dishonesty.

In public procurement, the equivalent harm is a procurement decision made on a false premise, public money committed to a product that does not deliver the claimed capability, and governance gaps that emerge when AI is operating inside contracts without the transparency and accountability that responsible use requires.

The label is not the capability. A properly developed AI system can be verified. A rebranded workflow, or a general-purpose language model bolted to an existing platform and priced as transformation, cannot — and will not answer the questions above with any confidence.

A structured buyer-side checklist covering supplier due diligence and organisational readiness for AI deployment is available separately. Get in touch if that would be useful.

That is the test worth applying before the contract is signed, not after the invoice arrives.

I am not a lawyer, so this is not a legal analysis of the judgement. It is a practical reflection on what the judgement may mean for suppliers and contracting authorities, from a procurement practitioner’s point of view.

For those less familiar with UK public procurement, the “standstill period” is a pause, between notifying bidders of the preferred bidder and contract signature. It provides unsuccessful bidders a short window to challenge a flawed procurement process. It also provides contracting authorities with a safeguard before contract signature, when potential issues can be surfaced and addressed before the contract is entered into.

Where a supplier has issued formal legal proceedings during the standstill period, the procurement is automatically suspended under section 101 of the PA2023. This prevents the authority from proceeding to contract signature and provides time for the court to review the merits of the challenge pending either an application from the authority to lift the suspension or the matter proceeding to trial.

Under the previous regulations, applications to lift suspensions were governed by the American Cyanamid principles for interim injunctions. In broad terms, the court considered whether there was a serious issue to be tried, whether damages would be an adequate remedy for the claimant and the balance of convenience for the parties. In practice, many suspensions were lifted because suppliers often found it difficult to show that damages would not be an adequate remedy.

The Procurement Act 2023 changes the framing. Section 102(2) requires the court to consider:

a. The public interest in, among other things:

i) upholding the principle that public contracts should be awarded, and contracts should be modified, in accordance with the law;

ii) avoiding delay in the supply of goods, services or works provided for in the contract or modification;

b. the interests of suppliers, including whether damages are an adequate remedy for the claimant;

c. any other matters the court considers appropriate.

That does not mean that every automatic suspension will be upheld. The particular facts of the case still matter. But it suggests that damages will no longer do the heavy lifting in the way they often did under the previous regime. The public interest in legally compliant procurement now sits much higher in the consideration. That is the significance of this case.

The court was dealing with a particular factual position: an incumbent supplier, an existing continuing service, allegations of non-compliance and no persuasive public interest reason to allow the authority to proceed immediately to contract award. A key point in the judgment is that damages would likely have been an adequate remedy for the claimant, but that did not lead to the suspension being lifted.

Adequacy of damages remains relevant under section 102(2), but the test now sits within a wider legal assessment that also weighs the public interest in lawful procurement and the interests of suppliers.

For suppliers, the judgment has strengthened the practical value of an automatic suspension. What the supplier really wants is not damages but the opportunity to compete for the contract. But this is not a cost-free option. If the suspension remains in place, the court may expect the supplier to provide a cross-undertaking in damages. That could expose the supplier to compensating the authority if the claim ultimately fails.

For contracting authorities, the message is equally practical. They should be cautious about assuming that an application to lift an automatic suspension will succeed simply because damages would be an adequate remedy. Authorities will need to weigh up whether an application to lift a suspension is likely to succeed. The assessment will depend not only on the adequacy of damages, but on the strength of the procurement record, the nature of the alleged breach, the public interest in proceeding, the impact of delay and the evidence available to support those arguments.

The wider lesson is upstream. Authorities should have in place strong quality control checks throughout the procurement lifecycle, particularly for high value, high risk or operationally sensitive procurements. This includes evaluation records, moderation notes, assessment summaries, notices, decision approvals and audit trails. If the procurement is challenged, those records may become central to whether the authority can proceed. Parkingeye is a reminder that the evidence trail authorities build may, months later, be what determines whether they can lawfully sign the contract at all.

This is the data intelligence problem. And it is not, at its heart, a technology problem.

The comfort of vanity metrics

Jeff Bezos noticed something unusual at Amazon. The dashboards looked good — on-time delivery, fulfilment efficiency, operational performance were all tracking in the right direction. But customer complaints were rising. The metrics and reality had diverged.

What Bezos had identified is the vanity metric: a measure that reassures leadership without truly testing the outcome. It shows activity. It shows process completion. But it does not show whether the system is working as intended.

It is an easy trap to fall into, particularly for procurement and contract management. The percentage of spend under contract tells you about contracting activity — it does not tell you whether those contracts are delivering value, managing risk or producing the outcomes the organisation needs. The number of suppliers onboarded tells you about supplier registration — but not necessarily whether the supplier data is complete, current, or useful when decisions need to be made.

None of these are wrong to track. The problem lies in how those metrics are interpreted, and in the limits of what they can actually show. When vanity metrics become the primary reporting approach, they create a false sense of security — and that false sense of security makes it harder, not easier, to see what is actually happening.

The illusion of control

In public sector organisations, the same issue can be subtler because the pressures that produce it are entirely understandable.

Teams are under pressure to demonstrate performance. Leaders want reassurance. Dashboards are built to provide it. So the data that rises to the top is the data that shows the system functioning — compliance rates, savings figures, contract coverage — while the data that would create discomfort stays lower down, in the detail, in the gaps between systems, and in the areas nobody is quite sure how to report consistently.

The result is that leaders make consequential decisions on the basis of only part of the picture.

The risk is not abstract. A 95% compliance rate can coexist with coding gaps that distort the picture entirely. Savings targets can be met on paper while contract leakage of thousands of pounds goes unnoticed. Contract registers can show clean, current contracts while extensions and variations accumulate in email chains that nobody has formally captured. The dashboard stays green. The exposure grows.

What the data missed

I have seen this in practice. On one large refurbishment project, the main performance and completion datasets confirmed that the contractor was meeting the agreed targets. Properties were being completed and handed over. On paper, the picture looked positive and the contractors were proactive partners throughout the project. But complaints were starting to come through, and resources were being pulled back into properties that had supposedly been completed.

The issue was not visible in the headline dataset because the problem sat after handover: repeat visits, resident frustration and post-handover rectifications. When the pattern was investigated, the cause was traced to a failing component in a toilet cistern. It was a small operational issue, but it had a wider impact on resources, resident experience and confidence in delivery. A more granular view of post-handover rectifications during practical completion would have identified the pattern earlier and allowed a different component to be fitted.

The point is not about blame. The headline data was accurate within its own limits, but it was not granular enough to show an emerging pattern at the point where it could still be addressed quickly.

The shift from data to intelligence starts when organisations become clearer about the decisions their data is meant to support.

That sounds obvious, but it is often where the gap appears. One dataset may be designed for financial control, another for procurement compliance, another for contract management, another for operational performance. Each may be accurate within its own boundaries. The problem is that commercial decisions rarely sit neatly within one boundary.

A contract renewal decision, for example, is not just a procurement timetable issue. It may involve demand changes, supplier performance, market capacity, service risk, inflation, legal exposure, operational resilience and affordability. If those signals are held separately, the organisation can appear well-informed while still making decisions from a partial picture.

From reporting to intelligence

Better commercial intelligence connects those signals before the decision is made.

That does not necessarily require a major technology programme. It often starts with a more disciplined approach to question design:

What decision is this data intended to support?

What would we do differently if the answer changed?

Are we looking at total spend, addressable spend or genuinely contestable spend?

Are savings being reported as cashable savings, avoided cost, efficiency or baseline movement?

Where is spend increasing because of price, volume, demand, specification or contract leakage?

Do supplier KPIs tell us whether the service is working, or only whether the supplier has reported against the measures in the contract?

Is the data granular enough to identify repeated local issues before they become accepted as operational noise?

Are contract extensions, variations and operational workarounds visible in the reported record, or sitting outside the data picture?

Where do finance, procurement, contract management and service teams see the same issue differently?

These are not abstract data questions. They are commercial judgement questions. They help distinguish between information that reassures and information that supports action.

The same point applies to published data. In public procurement, the Procurement Act 2023 and the Central Digital Platform will create a much richer published record across the procurement and contract lifecycle. That is a major step forward for transparency. But publication is not the same as intelligence. A notice may show that a contract has been awarded, a supplier has been named or performance information has been reported. It may not explain whether the procurement route generated meaningful competition, whether the KPIs are measuring the right things or whether the data is complete, comparable and consistently interpreted.

That does not undermine the value of transparency. It shows its limits. Published data can tell us more than before, but it still needs context, structure and judgement before it can support better commercial decisions.

The point is not to make reporting more complicated. In many organisations the answer may be to simplify it: fewer metrics, clearer definitions, better baselines and stronger links between spend, contracts, suppliers, risk and outcomes.

Data does not become commercial intelligence because it is collected, cleaned or displayed on a screen. It becomes intelligence when it changes what an organisation notices, challenges and decides.

The gap between having data and using it well is not primarily a technology gap. It is a question design gap.

That matters for what comes next. AI-enabled tools are already being presented as a way to analyse procurement and commercial data faster. Some will be useful. But speed is not the same as intelligence.

AI will not fix weak questions. It may simply answer them faster, at greater scale and with more confidence than they deserve.

What does your organisation’s data help you see — and what might it still be missing?

I am developing a short question bank for procurement and commercial teams who want to test whether their data is supporting better decisions or simply producing better reports. If that would be useful, let me know.

These weaknesses are not just procedural lapses. Poor scoping and weak contract management are creating significant value leakage pre- and post-contract award.

Commercial awareness is an often-used phrase that can mean different things. In some settings it is reduced to market knowledge. In others it is treated as an ability to navigate procurement rules, manage a sourcing exercise, or secure a saving at the point of award. All these matter. But none of them is enough on its own.

Commercial awareness is the ability to make sound judgements across the full commercial lifecycle to support delivery of the intended outcome, including when pressures, risk, change and operational reality begin to bite.

That may sound obvious. In practice, it is not how many organisations view it.

Too much attention is still placed on the procurement event itself: getting the business case approved, running the competition, documenting the decision and signing the contract. But the strength of a commercial arrangement is not proved at contract award. It is proved later, when ambiguity in scope starts to create friction, when risk sits awkwardly between the parties, when performance slips, when assumptions no longer hold and when the right to enforce a clause is a poor substitute for the service delivering.

In my view commercial awareness must be understood as a form of stewardship, not just process competence.

One of the most persistent weaknesses in commercial thinking is the illusion of transfer: where it is believed risk has been successfully addressed because it has been pushed into a contract clause, wrapped in indemnities, or backed by insurance requirements. On paper, the position may look robust. Poorly judged risk transfer however often reappears elsewhere: in inflated pricing, weak competition, fragile delivery models or disputes that consume time and energy, long after the original decision has been made. A right to sue is rarely an adequate substitute for service continuity.

A prime example of financial risk transfer is The Ministry of Justice (MOJ) Prison Maintenance Contract (2015), where contracts awarded to Carillion and Amey provided an annual estimated “saving” of £15m compared to previous costs. The saving was soon eroded prompting a MOJ injection of £15m to address the increasing poor prison conditions arising from a maintenance backlog.

Real commercial awareness is a broader and more disciplined view of the commercial lifecycle.

The starting point is not the tender. It is earlier than that. If commercial, operational and contract management thinking are absent at business case development and when requirements are being shaped, problems can be built in from the outset. Unrealistic expectations, weak understanding of delivery conditions, over-confidence in what the market can absorb and poor alignment between outcomes, scope and accountability all create difficulties that are expensive to resolve later. By the time a contract manager inherits the arrangement, many of the problems are baked in.

This is where commercial awareness becomes much more than a procurement skill. It becomes a judgement skill.

It means understanding how scope, pricing, liability, incentives, governance and delivery interact. It means recognising that a saving secured in procurement can simply become a cost pressure in contract management. It means knowing when to insist, when to compromise and when a supposedly firm contractual position is commercially unwise. It means understanding suppliers as delivery partners in a practical sense, not in the sentimental language that sometimes surrounds collaboration, but in the harder sense of understanding what will support resilient performance over the life of a contract.

There is a welcome shift to the broader interpretation in the public sector. The launch of the Government Commercial Agency this month, consolidates commercial expertise from across departments into a single integrated service for the sector. The Procurement Act 2023, together with the revised contract management Playbook and the separate Capability Programme, also points towards a stronger emphasis on commercial capability across the procurement lifecycle.

The real test is whether this broader way of thinking is consistently embedded in practice within the public sector generally.

That challenge is especially important in the public sector, where the consequences of weak commercial capability are rarely confined to budgets. They affect service resilience, public confidence, operational continuity and, ultimately, the credibility of the original business case. In that context, commercial awareness cannot be reduced to knowing the rules or following the process. It is the discipline of protecting intended outcomes from the first commercial assumptions through to delivery, change and exit.

For me, that is the real shift.

Commercial awareness should not be treated as a loose behavioural phrase or shorthand for business knowledge. It is a practical, judgement-based discipline: the ability to see how decisions made at one stage create pressures at another, to challenge weak assumptions early and to stay engaged beyond contract award, where intended value is either protected or quietly lost.

If we want better outcomes from public money, that is where the focus must shift: away from procurement as an event and towards commercial stewardship across the full lifecycle.

This is not about adding another slogan to the system. It is about strengthening a discipline often interpreted too narrowly for the responsibility it carries.

How well does your organisation connect the business case, procurement and contract management in practice — and how much value is being protected, or quietly lost, after award?

Do not confuse symptoms with causes

Too much of what is labelled transformation is still inward looking. It focuses on structures, processes and accountability lines, while giving little attention to the wider forces that can weaken or destabilise the organisation: financial shocks, digital disruption, climate pressures, capability loss, pandemics and geopolitical instability. These are increasingly familiar territory for organisations.

The systems thinking approach remains valuable, but it needs to be applied more completely and grounded in the reality of the business environment. The real test is not to understand the organisation as it is, but to understand how it behaves under pressure: what is changing around it and how to develop the resilience to adapt before strain becomes failure.

The risk is not only that organisations fail to respond. It is that they respond too narrowly or too reactively, reshaping structures and cutting capacity without asking what must be preserved to remain resilient. In trying to solve the immediate problem, they can quietly reduce their ability to manoeuvre.

For public organisations, building adaptability is particularly difficult. Unlike commercial organisations they are not, in the main, income generating bodies. They cannot withdraw from core services when conditions become more difficult. They operate within an environment of statutory duties, political oversight, financial constraint and public accountability. That makes shallow or inward looking transformation particularly risky. Yet public organisations also have a potential advantage: they can often achieve more through collaboration, standardised processes and platforms and more strategic use of shared suppliers, because they are not constrained by the same competitive dynamics.

Dysfunctional systems

In some organisations the existing system contains self-reinforcing loops where informal workarounds become normalised; normalisation weakens demand for discipline and the absence of discipline then justifies more workarounds. In that environment, new tools, training or ways of working are easily neutralised unless there is strong leadership intervention.

How drift goes unnoticed

Systems do not usually fail dramatically. They drift through reinforcing feedback, normalised workarounds, weak ownership, little challenge and misleading performance signals. This creates the impression that the system is still functioning and masks the evidence of deeper strain until the tipping point is missed.

The visible failure comes later, after the tipping point. This is when commercial pressure, normalised compromise and weak or blocked challenge begin to outweigh the organisation’s capacity to correct course.

Conclusion

Real transformation is not a change programme applied to the surface of an organisation. It is a disciplined effort to understand how the system behaves, including under pressure, what future conditions it must withstand and which feedback loops, controls, buffers and leadership actions are needed to prevent drift and build resilience.

Not because it was bad. Not because it was dated. But because so much of it still felt uncomfortably familiar.

It explored carbon reduction in the private rented sector and the barriers to improving older housing stock. I expected it to read like a period piece from an earlier policy era. In some respects, it does. The policy framework has changed, the legal duties are tougher, and the language around climate, safety and housing quality has become sharper. But what struck me most was not how much had changed, it was how much of the underlying system had not.

Even then, the real issue was not simply energy efficiency. It was the collision between old housing stock, weak incentives, fragmented accountability, affordability pressures and a policy habit of treating connected problems as though they could be solved one at a time. Looking back, it feels less like a historical snapshot and more like an early view of a system that was already under strain.

At the time, I was looking particularly at the private rented sector. It was difficult to reach, inconsistent in quality and shaped by a set of incentives that did not support the outcomes public policy was seeking. Landlords were expected to improve properties where the direct financial benefit often sat more obviously with the tenant. The stock itself was older and harder to improve. Regulation was uneven. Financial support was limited. There was also a real risk that if pressure increased without workable support, some landlords would leave the sector altogether. That would not simply be a private market issue. It would become a supply issue.

That logic still feels very familiar.

What has changed since then is not that the problem has been solved. It is that the pressures have intensified and multiplied. We are now asking the housing system to do several difficult things at once. We want more homes. We want safer homes. We want lower carbon emissions. We want better energy performance. We want improved health outcomes for residents. We want homes that can cope with hotter summers, wetter winters and more extreme weather. We want all of that while keeping housing affordable and viable to provide.

None of those aims is wrong. They are all necessary. But they do not sit in neat compartments.

That is where so much of the public debate still goes wrong. Housing is repeatedly discussed as though one bottleneck has arrived on its own. One week it is planning. The next it is regulation. Then it is viability, retrofit, landlord behaviour or building safety. Each of these matters. But in practice they interact. Tighten one part of the system without thinking through the others and pressure simply reappears somewhere else. It shows up in cost. It shows up in delay. It shows up in stalled schemes. It shows up in damp and mould risks when older homes are altered without enough thought to ventilation and moisture movement. It shows up in provider caution. It shows up in landlords quietly deciding that the numbers no longer work.

This is why I have become increasingly wary of single issue explanations. They are usually too neat for the reality people are dealing with. England does not have one housing problem. It has several overlapping delivery problems which happen to meet in the same place. There is a supply problem. There is an existing stock problem. There is a retrofit problem. There is a climate adaptation problem. There is a viability problem. There is also a governance problem, because accountability for the whole outcome is rarely located in one place.

That matters because people do not experience these issues separately. Residents experience them together. If a home is expensive to heat, vulnerable to damp, difficult to retrofit and poorly suited to more extreme weather, that is not four policy issues. It is one lived reality. In the same way, if a provider is trying to improve safety, meet standards, maintain ageing stock, reduce emissions and still make future development stack up, those are not separate operational tasks. They are one combined delivery challenge.

London shows how these pressures travel through the system. The private rented sector is shrinking at the same time as demand remains high and affordability is worsening. That matters because, for many younger and lower paid households, private renting is no longer simply expensive. It is edging towards unsustainable. When supply tightens in an already pressured market, the result is not just higher rents. It is reduced access, reduced mobility and a system that becomes harder to enter unless you already have income, family support or luck on your side. When lower and middle earners can no longer live within reasonable reach of work, housing stops being a private problem and becomes an urban systems failure. And the costs do not stay in housing budgets. They travel – into NHS waiting rooms, into emergency services, into social care, into the lost productivity of workers priced too far from employment.

Looking back at my thesis, I am less interested in whether every recommendation still stands than in what the comparison reveals. The policy language has evolved. The standards have tightened. The wider context has become more demanding. But the deeper pattern remains. We still struggle to align incentives with outcomes. We still underestimate the friction created when old housing stock is asked to meet new expectations. We still talk as though objectives can be added one by one without altering the behaviour of the whole system.

That is why I do not think housing should be treated simply as a housing debate. It is really a test of whether public policy can cope with interconnected reality. It sits at the meeting point of regulation, finance, health, climate, infrastructure, politics and delivery. If those elements are considered in isolation, the result is often frustration on all sides. Policy sounds right, but delivery feels harder. Standards rise, but progress slows. Pressure increases, but capability does not always follow.

What my old thesis reminded me is that many of today’s failures are not new failures at all. They are older tensions, left unresolved for too long, now intensified by tighter legal duties, climate change, rising costs and a more fragile economic environment. That does not make the objectives wrong. It does suggest that the way we frame the problem is still too narrow.

Housing was never just a housing problem. It was always a question of whether policy, incentives, regulation, finance and delivery could be made to work together in the real world.