Insolvency Service will use AI to identify rogue directors

The government has decided that the old tools are no longer enough for a problem that has grown both larger and harder to see. From next year, the Insolvency Service expects to have a new specialist taskforce fully up and running, armed with AI-driven analytics and backed by £25 million over five years, to hunt down directors who repeatedly walk away from debts by liquidating or dissolving companies and then starting again under a fresh name.

The practice has a respectable origin story. Businesses fail; markets turn; demand evaporates; cash runs out. Insolvency law exists in part to provide an orderly end to that story and, crucially, to allow those who have tried and failed in good faith to return and try again. Britain has never pretended that failure is a moral stain in itself. Yet the same tolerance for legitimate second chances has created a seam of exploitable weakness, and it is this seam that the Insolvency Service now says it must mine with better technology and sharper enforcement.

“Phoenixism” is the general term for a business that rises again from the ashes of an insolvent predecessor. In many cases it is mundane and lawful: assets are sold out of insolvency, staff are rehired, work continues, creditors take what recovery they can, and the productive capacity of the business is preserved. The abusive version is something else. It is the deliberate use of dissolution or insolvency as a method of shedding liabilities, sometimes repeatedly, while the same individuals carry on trading behind a new corporate shell. The losers are predictable: suppliers left unpaid, employees short-changed, customers with warranties that evaporate, and the taxpayer, through HMRC, watching arrears disappear into the administrative graveyard of struck-off companies.

The numbers that have pushed ministers and officials to act are stark. HMRC estimates that phoenixism accounted for 22 per cent of total tax losses in 2022-23, within an overall figure of £3.8 billion. That implies a tax-loss problem in the region of £800 million a year attributable to this kind of behaviour. In an era when governments of every stripe insist that fiscal room is limited, that is not a rounding error. It is also, politically, the most combustible kind of loss: not a complex avoidance scheme available only to the well advised, but a form of evasion that looks and feels like an open invitation to those willing to treat the corporate register as a revolving door.

The taskforce will be a 50-person unit, announced after the chancellor committed the funding in November’s budget. It began work in April and is expected to reach full operating capacity next year. Its immediate remit is to investigate suspicious insolvencies and dissolutions, particularly patterns that suggest a director is using the system to evade tax and write off debts. That language matters. It signals that the unit is not being built merely to improve administrative efficiency, but to pursue wrongdoing as such, and to do so in a way that produces sanctions, including disqualification.

Behind the initiative lies an uncomfortable assessment of past performance. A National Audit Office report in 2024 found that some small businesses were easily exploiting weaknesses in government systems to evade tax, and that there had been a lack of focus on phoenixism. The same report noted that the Insolvency Service had disqualified only seven directors for phoenixism between 2018-19 and 2023-24, out of 6,274 disqualified directors in total. Even allowing for the fact that phoenixism can be difficult to prove and can shade into legitimate business rescue, that figure suggests a level of enforcement wildly out of line with the scale of the harm estimated by HMRC.

It is that gap between the size of the problem and the smallness of the official response that the government now intends to close. The question is whether AI, plus a new unit with an explicit mandate, can do what broader institutional and legal changes have not yet managed: alter incentives, raise the odds of detection, and convince would-be repeat offenders that the risks have changed.

Dave Magrath, director of investigation and enforcement services at the Insolvency Service, has offered a tellingly cautious note. Enforcement outcomes, he said, are “really important”. But he added that enforcement alone will not solve the problem, because the underlying tension is structural: how to preserve the principle that honest directors can start again while stopping those who treat corporate death as a business model. There is also a wider economic framing that cannot be ignored. In a period when the country is “striving for economic growth”, ministers are reluctant to create a regime so harsh or so blunt that it chills entrepreneurship or discourages risk-taking in small business.

This is the argument that always appears when policymakers attempt to tighten the rules around company formation and dissolution. If it is too easy to incorporate, too easy to dissolve, and too hard to trace responsibility through successive entities, then bad actors will thrive. If it is too hard to incorporate, too hard to start again after failure, and too easy to be labelled a rogue, then the dynamism of the small business sector may be stifled. The difficulty, in practice, is that the costs of inaction fall on those least able to lobby: the small supplier unpaid by a dissolved customer, the local subcontractor left with a hole in cashflow, the public purse absorbing the losses and then raising the burden elsewhere.

The case for using AI begins with volume. Hundreds of thousands of companies dissolve each year. Within that mass are many entirely benign closures: dormant companies never used, trading businesses that wind down properly, entities dissolved after mergers, ventures that end without drama. Yet within the same stream may also be repeated patterns of harm: directors associated with multiple dissolutions leaving debts behind, common addresses reused, clusters of companies formed and struck off, or networks of related individuals and entities who appear whenever money is owed and vanish when it is time to pay.

Magrath put it plainly: the service needs “tech power” to find the needle in the haystack. The promise of AI-powered analytics is that it can identify relationships and patterns that a human investigator would struggle to assemble from disparate datasets at scale. If the taskforce can pull together information involving HMRC, Companies House and Insolvency Service records, then it may be able to surface risk indicators quickly, allowing the limited number of investigators to focus on the most egregious cases rather than wading through low-signal paperwork.

That is also where the limits become obvious. AI cannot create powers the state does not have. It cannot force better data sharing if agencies remain guarded or if legal gateways are narrow. It cannot automatically turn suspicion into evidence that meets the standard required for disqualification or prosecution. And it cannot by itself resolve the fundamental definitional challenge: distinguishing the legitimate director who failed and starts again from the director who engineered failure as a strategy.

The government appears to recognise that technology must be paired with a tougher legal framework. As part of the taskforce’s drive to disqualify more rogue directors, the Company Directors Disqualification Act is set to be amended, following a recent consultation, to extend the circumstances in which directors who break the law can be disqualified. The direction of travel is clear: more routes to disqualification, and potentially a broader set of behaviours deemed sufficient to justify restrictions.

The emphasis on disqualification is significant because it is one of the few sanctions that directly constrains repeat behaviour. Fines can be avoided by those who have already shown willingness to leave creditors unpaid. Prosecutions are slow, costly, and rare in the small business sphere. Disqualification, if applied quickly and consistently, can change the equation by preventing an individual from simply setting up the next vehicle. Yet disqualification is only as effective as the system that enforces it. If identity checks are weak, if “shadow directors” can hide behind proxies, or if monitoring is lax, then a ban becomes another rule that the determined can sidestep.

Recent cases cited in reporting illustrate the breadth of conduct the authorities want to deter: a director of a Burton fire alarm installation company who paid himself almost £400,000 across two companies while paying HMRC just £5,368, and an Oxfordshire landscaping boss who ignored his director ban and left £300,000 in unpaid tax across two companies. These examples are useful not because they are necessarily typical, but because they show how abuse is often entangled with personal extraction of cash, repeated use of corporate entities, and a willingness to disregard existing restrictions.

Caroline Sumner, chief executive of R3, the association for restructuring, turnaround and insolvency professionals, welcomed the taskforce as a response to “a longstanding issue”. Her framing is instructive: phoenixism, she argued, undermines confidence in the business environment and leaves creditors, including small businesses and HMRC, out of pocket, so coordinated action is essential. That points to the wider economic cost beyond the headline tax figure. When suppliers believe they are trading in a market where some counterparties can routinely stiff them without consequence, they raise prices, tighten credit terms, or refuse to supply. The result is a less efficient, less trusting business ecosystem, in which honest firms pay a hidden premium for the existence of dishonest ones.

Sumner also highlighted measures that sit alongside enforcement: strengthened identity checks for directors and improved data sharing. These are the plumbing of an effective regime. Identity checks matter because the easiest way to evade sanctions is to obscure who is really in charge. Data sharing matters because phoenixism is often visible only when you connect dots across multiple registries: tax arrears, director appointments, dissolution patterns, insolvency proceedings, and address histories. If each dataset remains in its own silo, the state ends up with the same disadvantage as an unpaid supplier, reacting after the company has vanished rather than preventing harm in the first place.

There is, however, a further complication: the government’s own desire to make it easy to start and run a business. Over the past decade the UK has pushed for simpler incorporation and quicker processes, partly to encourage entrepreneurship and partly to keep the country competitive. Companies House reforms have been introduced to strengthen the integrity of the register, but the basic convenience of the system remains a selling point. The more friction the state introduces, the more it risks criticism from those who see regulation as a brake on growth. Yet the alternative is to tolerate a system in which the honest subsidise the dishonest.

That trade-off is often presented as a binary choice, but it need not be. A system can remain simple for the majority while becoming more discriminating for those who trip reasonable risk markers. This is where the Insolvency Service’s language about patterns of repeated failure becomes important. Magrath suggested that the “heart of the solution” may come from a civil enforcement consultation, potentially leading to restrictions on directors who show repeated failures causing harm in lower-level cases, while still allowing them to contribute economically. In other words, a graduated response rather than an all-or-nothing regime, with early interventions for those whose history suggests rising risk.

The success of that approach will depend on whether restrictions are designed to be both fair and meaningful. If the threshold is too low, honest directors will feel hounded for misfortune. If too high, the most persistent offenders will still slip through until the damage is severe. If the restrictions are easily avoided, they will be treated as paperwork. If they are applied inconsistently, they will be perceived as arbitrary. Technology may help to standardise detection, but it cannot remove the need for judgment, nor the political responsibility for setting the balance.

There is also a question of whether £25 million over five years is enough. In one sense, it is a serious commitment for a targeted unit. In another, it looks modest against the stated scale of harm. If phoenixism truly contributes roughly £800 million a year in tax losses, then even a modest improvement in compliance would pay for the unit many times over. But that assumes the money can be recovered and that behavioural change follows. The taskforce is being asked to move from a world where disqualifications for phoenixism were counted in single digits to one where meaningful deterrence is plausible. That is not simply an operational challenge; it is a cultural change in how vigorously the state pursues white-collar wrongdoing at the small business end of the economy.

AI is being sold, in part, as a way to make that cultural shift feasible: to give investigators a sharper targeting tool so that effort is not wasted, and to make it harder for repeat offenders to hide in plain sight. Yet the state will also have to be clear about what it wants the technology to do, and how it will be governed. AI models can amplify the biases present in the data they are trained on, flagging certain sectors or geographies disproportionately if historic enforcement or reporting patterns were uneven. If the system produces false positives, it can waste investigative time and create unjustified stress for legitimate directors. If it produces false negatives, it will offer a comforting illusion of control while the worst actors adapt.

That suggests the need for transparency of a particular kind: not the release of sensitive investigative criteria, which would help offenders, but clarity about oversight, appeal mechanisms, and the principles guiding automated triage. There is a political risk here too. Ministers are keen to show that government is modernising and embracing technology, but they will find little applause if AI becomes a byword for bureaucratic overreach or for errors that harm innocent businesses. The taskforce will have to demonstrate not only toughness, but competence.

The broader point is that abusive phoenixism thrives on a mismatch between speed and scrutiny. It is relatively quick to set up a company, quick to trade, and, if one is willing to ignore creditors, often quick to dissolve. The investigation and sanction process is slow. By the time a case reaches a disqualification decision, the director may have already moved on to the next entity and the money may have gone. The objective of the new unit, supported by analytics, is to compress that timeline, spotting risky patterns earlier and intervening before harm is repeated. If that happens, the deterrent effect is likely to be more significant than any individual enforcement headline.

None of this removes the need for the mundane disciplines of enforcement: enough staff, enough legal support, coherent cooperation with HMRC and Companies House, and a willingness to pursue cases that are complex but important. The previous record, in which only seven directors were disqualified for phoenixism over six years, did not arise purely because investigators lacked clever algorithms. It arose because the system did not prioritise the problem, did not connect the data effectively, and did not apply sanctions at a rate commensurate with the estimated damage. AI may change the mechanics; it will not change the incentives unless the government sustains attention, funding and political will beyond the launch period.

For businesses that play by the rules, the test will be practical rather than rhetorical. They will look for faster action against serial offenders, fewer stories of directors cycling through multiple companies while leaving tax and supplier debts behind, and a Companies House register that feels less like a directory of disposable entities and more like a record that carries accountability. For the Insolvency Service, the test will be whether its new tools can turn what has long been an open secret into a demonstrably riskier game.

In the end, the argument for this taskforce is not only about revenue, though revenue is a compelling motivator. It is about the legitimacy of the corporate framework itself. Limited liability is a privilege granted in exchange for certain responsibilities. When that privilege is repeatedly used to impose losses on others without consequence, the bargain frays. The government is betting that data, analytics and sharper law can tighten the bargain again, without crushing the entrepreneurial impulse that the system was designed to protect.