Process vs practice: tax and welfare prosecutions in New Zealand

Prosecution is a mechanism to incentivise compliance. This may be, for example, to encourage compliance with the tax or welfare systems. New Zealand does not have a single agency responsible for making decisions on prosecutions. Instead, more than 40 agencies may make decisions on, and conduct, prosecutions (Crown Law, 2024). Each of these agencies is responsible for developing their own procedures relating to enforcement and prosecution, taking into account statutory obligations and practical considerations such as available resources (Crown Law, 2024).

This study has two parts. In December 2025 the New Zealand tax authority, Inland Revenue (IR) published prosecution guidelines (IR, 2025a). In January 2026, the social welfare agency, the Ministry of Social Development (MSD) did the same (MSD, 2026). These publications outline the agencies’ approach to prosecution activity, although they are both subject to the Solicitor-General Prosecution Guidelines (Crown Law Te Tari Ture o te Karauna, 2025). The first part of this research compares the two sets of prosecution guidelines. This comparison is necessary because research has shown that tax evaders are prosecuted at a significantly higher dollar value of offending than welfare fraudsters in New Zealand (Marriott, 2024). Therefore, the first step in this study is to address the research question:

Tax and benefit non-compliance are suitable for comparison as they both comprise financial fraud that can be quantified (Gottschalk, 2010). Moreover, they have the same victims (state and society) and typically do not involve other confounding factors such as violence or drugs. Therefore, similar prosecution guidelines may be expected from the two agencies. Notwithstanding this expectation, comparison of the prosecution guidelines highlights several differences, e.g. explicit recognition of Te Tiriti o Waitangi/the Treaty of Waitangi by MSD and not by IR, a wider range of enforcement tools for the MSD when compared to IR, and different emphasis on using resources between the two agencies ^[1]. However, these differences are not sufficiently substantial to explain the differences in financial thresholds for prosecutions in the two agencies. Thus, the second part of this study is a theoretically informed explanation for the differences found. Lipsky’s “street level bureaucracy” is used to provide insights into why the tax and welfare agencies produce different prosecution outcomes despite operating under broadly similar legal frameworks.

Street-level bureaucracy illustrates how broadly framed prosecution guidelines are operationalised within public agencies. The theory shows that the issue is not that IR and MSD apply significantly different legal tests but that agencies translate those tests into practice through organisational choices that facilitate discretion. Street-level bureaucracy structures the analysis around several related issues: rationing; routinisation; screening and non-escalation; policymaking through non-action; organisational constraints; operational cues; administrative workability; social control; and monitoring. These concepts highlight where decision authority sits, what oversight devices are built into the process, how each agency designs its enforcement “pyramid”, the practical availability of alternatives to prosecution, how “moderate seriousness” and related thresholds are defined, and how offender circumstances and reparation are treated. Used in this way, street-level bureaucracy provides an explanatory bridge between the formal similarity of the overarching prosecution framework and the different outcomes observed across tax and welfare administrations.

This article contributes to scholarship on prosecution policy and financial offending by showing that divergent tax and welfare prosecution outcomes in New Zealand cannot be understood solely through formal policy comparison. By applying street-level bureaucracy, it demonstrates that the crucial explanation lies in the organisational structuring of discretion, particularly in the largely opaque decisions that occur before cases enter the formal prosecution pathway. The article concludes that greater transparency is more important than formal consistency in written rules.

The article commences in Section 2 with a brief outline of the literature on prosecutions for financial fraud and white-collar crimes, as well as an introduction to street-level bureaucracy, the analytical framework adopted in the study. Section 3 provides data showing the different prosecution outcomes in the two agencies, along with methodological considerations. Section 4 highlights similarities and differences in the IR and MSD prosecution policies. Section 5 provides a theoretically informed analysis of these differences, with conclusions drawn in Section 6.

This section is in two parts. The first provides a brief outline of some of the literature pertaining to prosecution guidelines. The second engages with the theoretical framework adopted in this study: street-level bureaucracy.

In every country, most tax evasion is dealt with through civil and administrative regimes or negotiated settlements, rather than the criminal justice system (Levi, 2010). The issue of whether crime should be pursued through criminal or non-criminal proceedings is one that has been well canvassed in the literature. Levi (2010) for example, examines the international evidence on the impact of civil and criminal penalties on tax non-compliance. Levi reports that because prosecution occurs publicly, whereas civil or administrative penalties are confidential, this may encourage tax offenders to pay financial penalties to avoid the publicity of a prosecution and a criminal record. However, Levi also suggests that increased prosecution “might be justified for purposes of moral retribution as well as perceived social fairness” (2010, p. 493). As will be shown later in the article, retribution does not feature in the IR, MSD or Solicitor-General prosecution guidelines. Fairness, in the form of equitable criminal justice outcomes, is incorporated in the IR guidelines. MSD guidelines on equity are less clear, although reference is made to fair processes and the potential for bias to undermine decision-making.

Levi notes international cases of “strategic prosecutions” where high profile individuals are prosecuted for tax non-compliance “to leverage the effects of their interventions and to produce general deterrence” (2010, p. 494). Other examples include amnesties or warnings of potential enforcement areas by tax authorities. Levi provides examples of high-net-worth individuals who were assisted in tax secrecy by international banks which, when discovered, resulted in nearly 15,000 individuals “voluntarily” disclosing Swiss bank accounts to avoid possible criminal prosecution in the USA. However, what is not known is whether avoiding the visibility of prosecution or avoiding significant financial penalties was the key driver in the change of behaviour.

The absence of transparency around decisions limits an outsider’s ability to assess decision-making processes and to potentially challenge the application of rules or discretion by those making decisions on whether prosecutions proceed. This is observed in research, for example Sklansky’s observation that “we know much less about prosecutors than we do about the other main officials in the criminal justice system: judges and the policy. Unlike judges, prosectors generally do not announce the grounds for their decisions…[and]…Prosecutors carry out most of their work behind closed doors” (Sklansky, 2016, p. 474). In New Zealand, judges’ sentencing notes may be requested from courts, and data from the New Zealand Police that is not subject to suppression orders can be requested under the Official Information Act 1982. However, detail on cases that do not proceed into the prosecutorial decision-making process cannot be obtained.

Research examining the prosecution or non-prosecution of major financial crimes in the USA finds prosecution decisions are influenced by a prosecutor’s level of risk tolerance, the potential economic impact of a successful conviction, as well “the desired outcome” (Gilsinan et al., 2015, p. 5). The desired outcome includes fixing blame and satisfying a desire for a scapegoat, along with financial penalties that generate income for a government. As noted above, retribution does not feature in the prosecution guidelines examined for this study. However, as will be discussed below, the financial outcome can impact on the prosecution decision in New Zealand.

Research from the UK illustrates the impact of different prosecution decisions for tax and welfare fraud. Tax fraud costs the Crown nine times the value of benefit fraud, but the Department of Work and Pensions employ three and a half times more compliance staff than His Majesty’s Revenue and Customs (adjusted for the size of tax and benefits) (Tax Watch, 2021). In the 11 years to 2021, there were 23 times more prosecutions for benefit crimes (85,745), compared to tax crimes (3,665) (Tax Watch, 2021). Resourcing was also identified as a relevant factor with approximately 3.5 times as much resource dedicated to investigating welfare fraud and error as tax evasion. Research has highlighted issues around resourcing of fraud cases in the criminal justice system (Fitzpatrick, 2017).

Research has also noted the constraint of resourcing on tax prosecution decisions (Braithwaite, 2010; Levi, 2010). This impacts on the tax authority, whose mandate is to collect the highest revenue at the lowest cost, considering available resources. However, resourcing is not typically a factor identified as limiting benefit fraud prosecutions.

A related regulatory framework is Ayres and Braithwaite’s theory of responsive regulation (1992). Ayres and Braithwaite (1992) argue that effective regulation is most effective when agencies begin with cooperative strategies, while retaining the capacity to escalate to more punitive sanctions where non-compliance persists. This is commonly represented through the “enforcement pyramid”, in which the broad base consists of education, persuasion and negotiated compliance, while progressively more severe sanctions, including prosecution, are options that may be pursued when other lower-level enforcement has failed. The framework is relevant to tax and welfare prosecution because both IR and MSD present prosecution as a last resort within a wider hierarchy of compliance and enforcement tools. However, responsive regulation also highlights the conditions under which escalation occurs, including who makes this decision, what evidence is needed to support the decision and how consistency is achieved across different regulatory institutions.

The second part of this section sets out the theoretical framework used for analytical purposes in this article – Lipsky’s street-level bureaucracy. The value of this framework is that it shifts attention from the formal prosecution tests to the organisational conditions under which prosecution decisions are made. Even when agencies are subject to the same or similar overarching legal standards, street-level bureaucracy predicts that day-to-day decision practices will be shaped by workload pressures, resource constraints, and the need to translate ambiguous policy goals into routinised, defensible decisions.

Lipsky’s theory of street-level bureaucracy starts from the position that public policy is continually made and remade in everyday administrative practice (Lipsky, 2014). Street-level bureaucrats are public officials who interact directly with the public and exercise discretion over the allocation of benefits and burdens, including access to support, compliance interventions and sanctions (Lipsky, 2010). For this article, the value of the framework is not as a general theory of public administration, but as a way of examining how broad prosecution standards are translated into practical decisions within IR and MSD.

Following street level bureaucracy, officials operate under resource constraints, high workloads, limited information, and ambiguous or competing organisational goals (Lipsky, 2010). Discretion is a structural feature because formal rules cannot anticipate every case or resolve the tensions between efficiency, fairness, deterrence, rehabilitation and responsiveness. The practical consequence is that routine decisions about screening, escalation, and non-action become part of policy as experienced by citizens.

This is particularly important for prosecution policy because both IR and MSD operate under broadly similar overarching legal tests, including evidential sufficiency and the public interest. The key question is therefore how those common standards are converted into administratively usable thresholds. Lipsky’s framework highlights the organisational routines that determine which cases are treated as minor, which are resolved through compliance or recovery processes, and which are escalated to prosecution.

A central concept is the development of coping mechanisms. These include rationing scarce enforcement attention, screening cases according to organisational priorities, and simplifying complex circumstances into workable indicators of seriousness, risk or recoverability (Lipsky, 2010). Such practices help agencies manage caseloads and justify decisions, but they also shape outcomes by making some enforcement paths more available than others.

Street-level bureaucracy also explains why agencies develop procedural safeguards and review structures to control discretionary risk. Separating investigation from prosecution decision-making, requiring legal review, using panels or anonymising case material may promote consistency and reduce bias. However, these practices also show that discretion is organised rather than eliminated. They are part of the institutional framework through which agencies make prosecution decisions defensible while maintaining throughput in a resource-constrained environment.

Lipsky’s framework is also useful because it recognises the social control function of public administration. This is especially relevant to welfare enforcement, where policy and practice may communicate expectations about deservingness, work, dependency and appropriate conduct (Lipsky, 2023). In the prosecution context, this shows how benefit fraud and tax fraud may be framed differently from a moral perspective, and how those frames may influence the discretion exercised before cases enter the formal prosecution pathway.

Although street-level bureaucracy is not a criminological theory, it has been used to examine discretion in criminal justice and tax administration, including sentencing work, tax audits, taxpayer negotiations, and the effects of performance incentives on officials’ behaviour (Halliday et al., 2009; Cohen and Gershgoren, 2016; Raaphorst, 2018; Keulemans and van de Walle, 2020; Wang, 2025). This existing use supports its application in this study to prosecution policy, where the central issue is how discretionary judgement is structured before formal prosecution decisions become visible.

This section provides data on recent prosecution outcomes by IR and MSD. However, several methodological considerations follow from the use of Official Information Act 1982 (OIA) prosecution data. Firstly, the data provide a snapshot of completed or reported prosecution activity rather than the full enforcement pathway, i.e. they do not show the number of matters identified, or resolved through administrative penalties, settled through recovery processes, or screened out before prosecution was considered. Secondly, the IR and MSD figures are not symmetrical because each agency records and reports enforcement activity according to its own institutional and statutory functions. The comparison therefore focuses on broad patterns in prosecution volume and quantum rather than treating the figures as a complete measure of offending or enforcement effort. These caveats are important because the data can identify a disparity in prosecution outcomes, but they cannot by themselves explain how discretionary decisions are made before cases enter the formal prosecution process.

A further limitation is that the study is desk-based. It relies on published policies, annual reports and OIA-derived data rather than interviews or internal agency records. As a result, it can identify differences in prosecution outcomes and formal decision frameworks, but it cannot directly observe how officials interpret those frameworks or exercise discretion before they reach the formal prosecution stage.

In June 2025, OIA requests were submitted to IR and MSD for prosecution data for the three most recent years available. Data from IR is outlined in Table 1.

Data provided by MSD is outlined in Table 2.

Across the three years, 99 successful benefit fraud prosecutions were completed, exceeding the 76 tax prosecutions over the comparable period. However, the average quantum of benefit fraud was substantially lower at $62,219.49, compared with $412,256.78 for tax fraud, i.e. the average value of prosecuted tax fraud is nearly seven times that of welfare fraud. The observation that tax offending is prosecuted at a higher threshold of offending than welfare fraud is not new; it was first identified nearly 15 years ago (Marriott, 2012).

Both IR and MSD pursue only a small number of prosecutions annually. For example, IR initiated 50 prosecutions in the year 2025 (a 32% increase from the prior year) (IR, 2025b), while MSD completed 45 prosecutions in the 2024 / 25 year. These figures are small relative to the volume of compliance interventions undertaken by MSD. MSD’s integrity program in that year managed over 276,000 early intervention cases and 6,298 formal investigations, showing that only a tiny fraction of investigated cases escalated to prosecution (MSD, 2025a, p.63). In contrast, completed investigations (tax audits) numbered 4,300 in 2024 / 25 (Inland Revenue, 2024), which must be viewed against 4.7 million individual taxpayers. Therefore, while both agencies prosecute sparingly, targeting only the most serious cases, the same cannot be said for the less visible investigations, where MSD’s surveillance and monitoring far exceeds that of IR.

Institutional resourcing further entrenches these disparities. IR has an appropriation “to protect the integrity of the tax system and functions that the Commissioner administers” (IR, 2025b, p. 52). The budgeted amount for this service is NZ$126m for the 4.7 million individual taxpayers in New Zealand in 2024 (IR, 2025c) ^[2]. The MSD has an output expense for “investigation of overpayments and fraudulent payments and collection of overpayments” (MSD, 2025a, p. 62). Funding allocated to this activity is approximately half allocated to IR’s integrity function at $63.8m (MSD, 2025a, p.63). However, there are around one-tenth the number of people receiving core benefits in New Zealand at approximately 430,000 individuals (excluding the universal pension) (MSD, 2025b). Thus, the funding allocated to the MSD investigative function per beneficiary is approximately 5.5 times the funding allocated to protecting the integrity of the tax system per taxpayer. This resource configuration reflects political priorities rather than relative harm and reinforces class-based distinctions in enforcement intensity.

As noted, both IR and MSD policies are explicitly subject to the Solicitor-General’s Prosecution Guidelines. The Solicitor-General’s guidelines “do not limit prosecutors’ discretion and prosecutors are expected to exercise their judgement” to allow for different circumstances in cases (Crown Law Te Tari Ture o te Karauna, 2025, p. 6). This section reviews the two agencies prosecution guidelines to identify similarities and differences.

The guideline comparison was conducted through a textual analysis of the publicly available prosecution policies published by IR and MSD, read alongside the Solicitor-General’s Prosecution Guidelines that govern public prosecutions in New Zealand. The analysis first identified the common legal framework applying to both agencies, particularly the evidential sufficiency and public interest tests, before comparing how each agency translated those requirements into operational policy. Inductive analysis of the prosecution guidelines highlighted five key themes: purpose and institutional mandate; scope and legal basis; decision-making authority and oversight; the range of enforcement options available before prosecution; and the treatment of offending seriousness, offender characteristics, remediation and alternatives to prosecution. These themes are used to structure this section and highlight the differences between formal policy differences and their likely practical significance. The comparison is then interpreted through the lens of street-level bureaucracy to assess whether the written guidelines explain observed prosecution outcomes, or whether the more important explanation is found in the discretionary space in which cases are resolved administratively or kept outside the formal prosecution pathway.

Both agencies’ prosecution policies focus on protecting the integrity of their system and encouraging voluntary compliance. As noted in the Solicitor-General prosecution guidelines, prosecution will ordinarily be the appropriate response to “criminal offending of at least moderate seriousness” where it is a proportionate response to the circumstances of the case (Crown Law Te Tari Ture o te Karauna, 2025, p. 7). Both agencies adopt a graduated enforcement model before considering prosecution. The hierarchies of approaches are outlined in Table 3 and discussed in more detail later in the article.

IR uses prosecution as an enforcement activity “usually of last resort” and against “those who manipulate and abuse the system to reduce the tax they are required to pay or to obtain refunds or entitlements they are not eligible for” (IR, 2025a, p. 3). MSD’s policy, in contrast, protects the welfare system by early intervention and debt recovery, outlining a three-tier integrity approach (early correction, intensive facilitation, then formal investigation and prosecution). While both policies focus on deterring deliberate wrongdoing, MSD’s policies emphasise early intervention.

The Commissioner of Inland Revenue’s duty is to collect “over time the highest net revenue that is practicable within the law having regard to available resources, promoting compliance, and compliance costs” ^[3]. While prosecution activity has similar objectives, there is a tension between collecting revenue, promoting compliance, holding the non-compliant to account, and balancing demands on resources.

The prosecutor will decide which charge or charges are most appropriate. These will reflect the seriousness and extent of the offending and provide the court an appropriate basis for sentencing (IR, 2025a, p. 15). Where it is possible to charge the offending under either the Tax Administration Act 1994 (TAA) or the Crimes Act 1961, the prosecutor selects the charge that is most likely to lead to the “just, efficient and effective resolution of the criminal proceedings; and accurately and adequately reflect the seriousness and extent of the alleged tax offending” (IR, 2025a, p. 16).

MSD’s Prosecution Policy references the legislation that may be used by MSD to prosecute criminal non-compliance. However, the policy also notes that “charges under other Acts, such as the Crimes Act 1961, may be considered where this is more appropriate” (MSD, 2026, p.7). Prior research indicates that, at least historically, the Crimes Act 1961 has been frequently used by MSD, and infrequently used by IR, as shown in Table 4. Harsher maximum penalties exist under the Crimes Act 1961.

Unlike IR, which incorporates deterrence as part of the decision-making process, MSD’s response includes denunciation alongside deterrence: “any response will seek to denounce and deter the conduct or offending, including the recovery of any debt” (MSD, 2026, p. 2). In addition, only MSD policy includes reference to Te Tiriti o Waitangi. This is surprising as the Supreme Court has confirmed that tikanga is an important part of interpreting the law in New Zealand (Crown Law Te Tari Ture o te Karauna, 2025) ^[4]. The MSD guidelines also explicitly refer to Māori, with reference to MSD’s strategic goals and objectives that include:

• the intention to act “reasonably, honourably and in good faith towards Māori”; and

• to recognise and provide for Māori perspectives and values and take positive steps to ensure protection of Māori interests (MSD, 2026, p. 2).

Within IR, the decision to prosecute is made by senior management. Specifically, a business unit Group Lead decides whether to prosecute at the conclusion of an investigation. If a lesser enforcement option, e.g. a penalty or warning, is selected instead, the decision will usually be made at the Team Lead level. Before any charges are filed, there will be a formal legal review that considers whether the evidential and public interests tests are met. This is documented in a legal opinion and peer reviewed by an IR prosecutor. Disagreements with the legal opinion recommendation result in an internal review and escalation to the Legal Services Leader to make the final decision on prosecution.

Within MSD, after an Investigator completes a fraud investigation and believes prosecution might be warranted, they first consult with their manager and an MSD lawyer. At this stage, they consider whether non-prosecution options (such as those mentioned in Table 3) could suffice. The lawyer will assess if the evidential and public interest tests are met, much like IR’s process does. If both tests are met, the case is referred to the Prosecution Review Panel for a final decision. Cases are anonymised when they go to the Prosecution Review Panel to minimise the potential for bias.

As with IR, prosecutors and investigators have separate roles. MSD’s policy emphasises that while they should cooperate, prosecutors make decisions independently of investigators. Investigators gather facts; prosecutors apply legal tests. This separation mirrors the independence principle in the Solicitor-General’s guidelines, ensuring investigations are not influenced by biases in the field: the final call is made by those considering the legal merits.

Each approach (IR’s hierarchical review versus MSD’s panel) has advantages. IR’s model leverages strong subject-matter expertise and clear accountability up the chain for each decision. MSD’s panel model provides collective judgement and an extra layer of impartiality through anonymity. Both systems involve multiple people, guarding against any single individual pushing a prosecution without oversight. However, both agencies lack transparency on decisions made at lower levels not to prosecute, i.e. before engagement with the formal prosecution policy is enacted. This generates the opportunity for discretion to result in different outcomes in the two agencies.

As noted above, IR have five potential enforcement options. The first of these is to take no action. This action may be appropriate where “the non-compliance is minor, inadvertent and quickly corrected” (IR, 2025a, p. 7). Taking no action is similar to “deferred prosecution agreements” visible in other jurisdictions that result in no action being taken in return for compliance over a specified period (Parker and Dodge, 2023). The advantage with a deferred prosecution agreement is that it provides some consequence in the event of non-compliance, which taking no action in isolation does not. Taking no action is not an option in the MSD hierarchy of approaches.

The next level of enforcement is to take an educative approach. Both agencies adopt this tool. It is used when a taxpayer was genuinely mistaken or misunderstood their tax obligations and they are likely to comply in the future because of the education provided (IR, 2025a). MSD are less likely to consider that the offending was an error and instead provide “advice to the offender of their inappropriate behaviour and how to comply in the future” (MSD, 2026, p. 8).

Further escalation of penalties at IR is the use of an informal warning. This may be used in conjunction with a shortfall penalty and/or education. MSD may also issue an offender a warning that they have committed a specific offence, but MSD has decided not to prosecute in this instance (MSD, 2026).

The fourth option for IR is imposing a shortfall penalty. Where a shortfall penalty is imposed in relation to a tax shortfall, it is likely to act as a bar to a prosecution ^[5]. This option may be used as an alternative to prosecution where there is a high likelihood that the taxpayer will voluntarily pay the shortfall penalty; or the Commissioner can enforce payment; or bankrupting or liquidating the taxpayer would be an appropriate compliance outcome, despite payment being unlikely to be recovered. MSD’s prosecution guidelines also explicitly refer to financial penalties as part of their enforcement tools. If someone obtains a benefit payment by fraud, a penalty of up to three times the amount of the excess may be recovered by MSD^[6].

MSD have two other further options before prosecution. The first is civil recovery to recover any overpayment. The second option is assigning a client to Low Trust Client Management, where clients have been identified as having taken advantage of the benefit system. When a client is assigned to Low Trust Client Management, they have restricted access to online services and are required to have in-person contact and provide additional verification when seeking financial assistance. This assignment may be for a specified period or permanent, and it can be instead of, or as well as, prosecution (MSD, 2026). Unlike IR, when a debt is established with MSD through deliberate fraud “then an overpayment will be established, and recovery of this money will be sought” (MSD, 2026, p. 6). In addition, prosecution action will be considered.

Prosecution is the final and most resource-intensive option for both agencies. IR will consider prosecution usually after other enforcement options have been considered and rejected. MSD’s prosecution policy references “offending of at least moderate seriousness” for prosecution consideration (MSD, 2026, p. 6). Following MSD’s policy “prosecution will ordinarily be the appropriate response to serious criminal offending” but “should otherwise be used only where it is proportionate to the circumstances of the case” (MSD, 2026, p. 7).

Thus, the common philosophy is that prosecution is reserved for wilful, significant fraud – not mere mistakes or small lapses. Both agencies focus their criminal enforcement on the more serious cases in terms of non-compliance, using lesser measures to correct or discipline the rest. The exact tools differ (e.g. tax shortfall penalties or welfare penalties) but the strategy of graduated enforcement is similar.

Certain features of the offending will impact on the prosecution decision. Those outlined in the IR Prosecution Guidelines include:

• the seriousness of the offending;

• the individual’s involvement and level of culpability;

• whether there is a need to deter the type of offending; and

• whether the likely sentence that would be imposed on conviction would be only minor (IR, 2025a).

Personal characteristics and circumstances are a further factor considered by IR, including age, disabilities or mental health issues, prior criminal history, and other aggravating or mitigating features, including expressions of remorse or efforts to restore losses.

IR do not usually prosecute a taxpayer who has made a pre-notification full voluntary disclosure about their tax position IR, 2025a). This is to encourage voluntary disclosure. The prosecution guidelines are explicit that a “decision to prosecute in these circumstances will be an exceptionally rare occurrence” (IR, 2025a, p. 17). This is because voluntary compliance promotes the tax system’s integrity via cooperation, and it allows IR to balance enforcement while considering resource priorities.

MSD’s policy also includes a clause that if there is another way of dealing with a particular case that is effective and appropriate in the circumstances, then it should be used (MSD, 2026). However, MSD are less explicit about the extent to which voluntary disclosure impacts on prosecution, with the prosecution policy writing “if someone has made a full voluntary disclosure that was not prompted by the commencement of an investigation, this will be a factor to be weighed in the consideration of whether to prosecute or not” (MSD, 2026, p. 6).

IR’s prosecution guidelines observe that offending against the tax system is “unusual in that there is a possibility that an offender may be able to substantially right the wrong they have committed against Inland Revenue and the wider community by (re)paying the tax shortfall along with penalties and interest” (IR, 2025a, p. 12). While the prosecution guidelines explicitly state that the Commissioner will not enter negotiations or plea discussions regarding how the impact of payments made towards outstanding tax liabilities will impact on the prosecution decision-making process, they also state:

Where the taxpayer can substantially rectify the situation before Inland Revenue files charges, this is often a powerful mitigating factor in favour of the Commissioner deciding that prosecution is not in the public interest. However, each situation must be assessed on its merits, and in some circumstances the damage done to the integrity of the tax system means the public interest will favour prosecution, even where the taxpayer has repaid the tax shortfall (noting they will nevertheless get credit for the rectification at sentencing should they be convicted) (IR, 2025a, p. 12).

MSD’s prosecution policy is significantly more detailed than IR’s particularly in its outlining of factors that contribute both to the seriousness of the offending and factors that would count against prosecution. However, both agencies’ factors are ultimately similar to those provided in the Solicitor-General’s Prosecution Guidelines, just with different levels of detail.

From a street-level bureaucracy perspective, the punishment objectives referenced in prosecution guidelines are important because they provide the normative cues through which officials translate broad public interest tests into practical enforcement decisions. Although criminal punishment is conventionally understood to serve multiple purposes, including deterrence, reparation, retribution, incapacitation and rehabilitation (Zedner, 1994) - the prosecution guidelines for both IR and MSD refer primarily to deterrence, with IR also incorporating reparation and MSD also incorporating denunciation. This narrows the decision frame, directing officials towards outcomes that protect system integrity and recover losses rather than towards the full range of penal goals. With reference to street-level bureaucracy, MSD’s additional reference to denunciation signals a broader and different institutional framing of what prosecution is meant to achieve. Denunciation strengthens the expressive and moral dimension of enforcement, potentially sharing how officials classify cases, what they regard as proportionate, and when they treat prosecution as necessary rather than excessive. In this way, punishment objectives are not merely abstract statements of principle; they are part of the organisational architecture that structures discretionary judgement in everyday prosecutorial practice.

In Lipsky’s theory, public officials operate under resource constraints, competing policy goals, high workloads and discretion, which in practice becomes public policy. The prosecution thresholds used by IR and MSD illustrate the Lipskian concept of rationing. While operating as legal safeguards, they also ration scarce prosecutorial attention by identifying which cases are sufficiently serious to justify escalation. When the guidelines state that prosecution is exceptional, they legitimise the routine resolution of many cases through non-prosecutorial pathways. Discretion is therefore structurally necessary. Graduated enforcement also reflects what Lipsky describes as a coping mechanism: it enables agencies to reconcile competing goals such as deterrence, welfare protection, revenue collection, compliance, fairness and efficient administration.

Te Tiriti o Waitangi has particular implications for enforcement discretion because it requires public agencies to consider whether apparently neutral enforcement practices may produce uneven burdens for Māori, and whether Māori perspectives and interests have been recognised in the design and application of discretionary processes. Thus, MSD’s explicit reference to Te Tiriti o Waitangi signals that prosecutorial discretion should be exercised with attention to Māori interests and the broader relationship between Māori and the Crown. IR’s absence of an equivalent statement does not remove the relevance of Treaty obligations, but it may make the Treaty dimensions of enforcement less visible within operational decision-making. A Treaty-consistent approach would therefore require transparency about how discretionary judgements are made and how agencies ensure that enforcement practices give meaningful effect to Treaty principles.

The enforcement pyramids adopted by IR and MSD are best understood through Lipsky’s concept of routinisation. Public officials simplify complex policy environments by creating ordered decision pathways that make discretionary judgement administratively workable. The hierarchy of penalties becomes a mechanism for channelling discretion under pressure rather than a neutral compliance tool. It tells officials how to move from education to warnings, penalties, recovery, monitoring or prosecution, converting broad standards into practical enforcement routines.

The two agencies’ formal hierarchies also reveal the Lipskian importance of screening and non-escalation. From a street-level bureaucracy perspective, the more significant discretion may occur before cases are categorised as potentially prosecutable. Officials decide whether matters are minor, suitable for education or recovery, appropriate for monitoring, or serious enough for legal review. These decisions are the practical points at which policy is made. However, they remain largely opaque. There is little visibility over the processes or judgements that determine whether a case ever enters the formal prosecution pathway. Greater transparency at this point would reveal how discretion is operationalised in practice and help explain why prosecution outcomes differ across agencies despite a shared overarching legal framework.

The low prosecution volumes outlined in section three are similarly explained by Lipsky’s concept of policymaking through non-action. Decisions not to prosecute are important administrative decisions that determine how the law is experienced in practice. In the case of IR, practical capacity is a significant constraint, meaning that enforcement severity is symbolically maintained while rarely used. Deterrence therefore operates more as a background threat than as a likely sanction. This is a form of selective enforcement: the possibility of prosecution remains visible, but the actual use of prosecution is rationed through organisational judgements about seriousness, resources and strategic value.

IR’s repeated references to resources provide a direct example of Lipsky’s claim that discretionary practice is shaped by organisational constraint. The IR prosecution guidelines refer to resource considerations in seven separate places, including where prosecution is justifiable but would consume resources “that could be better used elsewhere” (IR, 2025a, p. 14). This frames prosecution as a rationed institutional choice shaped by efficiency and the strategic allocation of enforcement capacity. In contrast, the absence of similar references in the MSD guidelines suggests a different organisational framing of prosecution decisions. Following Lipsky, this difference highlights how resource constraints generate the coping mechanisms through which agencies screen, prioritise and resolve cases without prosecution.

The treatment of remediation illustrates Lipsky’s concept of operational cues. Both agencies recognise that subsequent compliance may affect whether prosecution proceeds, but they structure that consideration differently. IR’s prosecution guidelines treat reparation relatively broadly: “efforts to repair damage or restore losses” are identified as mitigating, and where a taxpayer has “rectified the non-compliance” this is a factor likely to weigh against prosecution (IR, 2025a, pp. 11 / 14). This gives decision-makers a clear cue to treat restored compliance as evidence that a punitive response may be unnecessary. MSD adopts a narrower approach. Its policy refers to encouraging voluntary compliance and disclosure, but only a “full voluntary disclosure” made before the commencement of an investigation is identified as a factor that may or may not impact on whether prosecution occurs (MSD, 2026, p. 6). These differences show how broad public interest standards are converted into administratively usable indicators. IR’s broader recognition of remediation creates more room for discretion to resolve cases without prosecution, whereas MSD’s narrower formulation limits the circumstances in which subsequent compliance can displace a prosecutorial response.

Financial penalties as an alternative to prosecution show how discretion is shaped by administrative workability. Both prosecution guidelines incorporate financial penalties, but IR and MSD structure that option differently. For MSD, a financial penalty may be considered if it is appropriate to the offending. For IR, in contrast, a shortfall penalty may be used as an alternative to prosecution where there is a high likelihood that it will be paid. This makes ability to pay an explicit operational consideration in deciding whether a case can be resolved without prosecution. Lipsky’s framework suggests that officials simplify complex decisions by relying on practical proxies such as recoverability and the likelihood of future compliance. Because welfare recipients are likely to have fewer financial resources, their capacity to absorb a financial penalty is more constrained than that of tax offenders, who may have retained the funds connected to the non-compliance. The result is that a formally similar alternative to prosecution may be more realistically available in the tax context than in the welfare context.

The distinction between benefit fraud as a crime of need and tax fraud as a crime of greed is connected to Lipsky’s concept of social control through the meanings attached to welfare and tax offending. Benefit fraud is often best understood as a crime of need, arising in contexts of poverty and insecurity, even though this does not remove individual responsibility or the need for system integrity. Tax fraud, in contrast, involves the deliberate retention or concealment of resources by those with greater economic capacity. The resulting moral asymmetry is significant. Those most in material need may be subject to more intensive surveillance and greater likelihood of prosecution, while offending associated with greater financial advantage may be more readily diverted into repayment or negotiated compliance.

Reference to the available enforcement options provides further insight into Lipsky’s concepts of sorting and institutional monitoring. From a street-level bureaucracy perspective, enforcement options are organisational tools for classifying cases and managing workloads. IR’s inclusion of a “take no action” option indicates a degree of flexibility to screen out minor or quickly corrected non-compliance without activating a more intensive enforcement pathway. MSD, in contrast, does not include a comparable option, but instead includes civil recovery and Low Trust Client Management among the enforcement options. Thus, the enforcement pyramid creates the practical routines through which officials decide which cases are ignored, monitored or escalated. The contrast illustrates how discretion is structured differently across the two agencies, with IR retaining greater scope for administrative non-response, while MSD adopts mechanisms that extend institutional oversight even where prosecution does not occur.

Increasing use of artificial intelligence and automated decision-making is also relevant to this analysis because it can shift monitoring and discretion from visible human judgement to less visible technical systems. Automation may appear to standardise decision-making, but it can also embed policy assumptions and resource-driven shortcuts into systems that are difficult for affected individuals to understand or challenge. The Australian Robodebt scheme provides a cautionary comparator. In the Robodebt scheme, automated income averaging was used to raise welfare debts, but the scheme produced inaccurate results and was unlawful (Commonwealth of Australia, 2023). The significance of Robodebt for the present study is that automated tools can intensify the problems identified by street-level bureaucracy: screening, non-escalation and enforcement decisions may become less transparent when mediated through data-matching or algorithmic processes. The use of AI may therefore make discretionary practices appear more objective while reducing the visibility of the policy choices and institutional priorities embedded within them.

Are there any solutions to these issues? The key problem identified in this study is the opacity of discretionary decisions and, in particular, decisions not to prosecute. If discretion cannot be eliminated, it should at least be more visible and publicly accountable. Increased transparency need not be overly burdensome. For example, publication of more detailed prosecution decision data would lead to greater visibility of criteria used in decisions not to prosecute. Reporting on the use of alternatives to prosecution would also provide for greater institutional accountability on decision-making. Moreover, it would facilitate the comparison between agencies that are frequently critiqued for appearing to use different standards for similar offences. Without greater visibility over how cases are screened out or resolved through less visible punishment alternatives, the core issue identified in this article – discretion exercised beyond meaningful public scrutiny - remains unresolved.

This article has compared the prosecution guidelines of IR and MSD to examine whether differences in formal policy explain the divergent prosecution outcomes observed across tax and welfare offending in New Zealand. Although the two agencies operate under the same Solicitor-General framework and both present prosecution as a last resort, they structure discretion differently through their decision hierarchies, enforcement options, and treatment of remediation, resources and seriousness. However, these differences do not fully explain the disparity in prosecution practice. Rather, they show how formally similar legal tests are translated into distinct organisational settings that influence how cases are screened, escalated or resolved without prosecution.

Through a street-level bureaucracy lens, the central finding is that prosecution policy is not explained by what the written rules say. The key work is done through the organisational routines that make broad legal standards administratively workable under conditions of limited resources and competing mandates. Low prosecution volumes are evidence of rationing. Decisions not to prosecute, the availability of alternatives, and the practical criteria used to identify a case as worth pursuing are all part of how policy is made in practice. The comparison therefore supports Lipsky’s insight that discretion is not an unfortunate residue left by imperfect rules, but a normal and constitutive feature of public administration.

The significance of this finding is practical as well as theoretical. If discretion is inevitable, the task is to make its operation more visible and more accountable. Greater transparency over how cases are filtered before they become prosecutable, how alternatives to prosecution are used, and how non-prosecution decisions are reached would provide a firmer basis for assessing whether similar forms of financial offending are being treated similarly across the tax and welfare systems. Ultimately, the study suggests that the problem is not just process versus practice, but of the limited public visibility of the discretionary space between them.

Several policy recommendations follow from this analysis. In the interests of transparency, IR and MSD should publish more comparable data on the full enforcement pathway, including the number of matters identified, investigated, resolved through education, warning, penalty, recovery or monitoring, and referred for prosecution. Both agencies should provide clearer public explanations of the criteria used to screen cases out before the formal prosecution stage, particularly where decisions turn on seriousness, available resources, remediation, ability to repay or offender circumstances. In addition, alternatives to prosecution should be reviewed to ensure they are practically available across different socio-economic contexts and do not operate more favourably for those with greater financial capacity. IR and MSD should report how Te Tiriti o Waitangi obligations and equity considerations are incorporated into enforcement and non-prosecution decisions. Finally, where data-matching, risk-scoring or automated tools are used to support compliance or investigation work, agencies should disclose the role of those tools and maintain meaningful human review. These measures would not remove discretion, but they would make its use more transparent, comparable and accountable.

The study also has limitations. It is desk-based and relies on published prosecution policies, publicly available agency material and OIA data. While this approach allows for systematic comparison of formal frameworks and aggregate outcomes, it cannot reveal the internal reasoning, organisational routines or individual-level judgements that shape decisions before files enter the formal prosecution pathway. Future research would therefore benefit from qualitative interviews with agency staff, prosecutors, investigators, and others involved in compliance and enforcement decision-making. Such interviews would provide richer insight into how discretion is exercised in practice, how alternatives to prosecution are assessed, and why broadly comparable forms of financial offending may produce different prosecution outcomes.

To conclude, the contribution of the research is threefold. Firstly, an empirical contribution through the systematic comparison of the newly published prosecution guidelines of IR and MSD, highlighting the commonalities and differences. Secondly, a theoretical contribution using Lipsky’s street-level bureaucracy to explain why similar formal legal frameworks can produce different enforcement outcomes. Thirdly, a policy contribution. The article argues that the real issue is the limited visibility and accountability of discretionary decisions, especially decisions not to prosecute. This leads to the implication of the research which is that greater transparency is likely more important than formal consistency in written rules.

The researcher is grateful for the constructive feedback provided by two anonymous reviewers and the Journal Editor.