Challenges for individual assessment in digitalised welfare administration: the case of social assistance

Digitalisation has transformed welfare administration into what scholars term “screen-level bureaucracy”: an ICT-supported, rationalised, rule-based system (Bovens and Zouridis, 2002; Busch and Henriksen, 2018). In this study, screen-level bureaucracy refers to digitalised administration, where implementation relies on standardised procedures and discretion is tightly constrained (Bovens and Zouridis, 2002; Newman et al., 2022). By reducing professional judgement and shifting citizen–worker interaction to digital interfaces, this development is reshaping the way social assistance is administered and implemented across welfare states (Huby et al., 2024; Larasati et al., 2023; Meers et al., 2025).

We focus on decision-making regarding social assistance, a financial safety net of last-resort for people whose primary benefits are insufficient. Determining eligibility and needs requires an understanding of each applicant’s circumstances (Jewell, 2007; Saikkonen and Ylikännö, 2020), which involves a comprehensive individual assessment (Hussénius et al., 2025; Jewell, 2007). Previously, such assessment was carried out by street-level bureaucracy, grounded in professional expertise and discretion (Evans, 2013; Molander, 2016). With the rise of screen-level bureaucracy, however, individual assessment is increasingly conducted through digitalised and standardised procedures (Marienfeldt, 2024), which reconfigures discretion and the conditions under which decisions are made. This warrants attention, as welfare systems are increasingly digitalised internationally to improve efficiency and address perceived limitations of human-centred decision-making (Germundsson, 2022; Jørgensen, 2023; Larasati et al., 2023; Ranerup and Svensson, 2023).

Although previous research has examined frontline workers’ discretion (Ball et al., 2023; Busch et al., 2018; Ilyas et al., 2025; Soares et al., 2024), and professionalism in digitalised welfare (Bernhard and Wihlborg, 2022; Pedersen and Wilkinson, 2018; Taylor, 2017), less is known about how institutional provisions shape individual assessment in decision-making within digitalised welfare administration. This study addresses that gap by analysing the shaping of assessment through guidelines and ICT systems within screen-level bureaucracy, a system designed to promote consistency and equal treatment, but which may constrain the consideration of individual needs (Bovens and Zouridis, 2002; Buffat, 2015).

Finnish social assistance provides a compelling case for investigating screen-level bureaucracy. Since 2017, it has been administered through a nationally centralised and highly digitalised system, unlike the decentralised, local models common across Europe (cf. Hölsch and Kraus, 2004; Lunder, 2016; Minas et al., 2012, 2018; Petroons et al., 2025). By focussing on the process of reducing social assistance, which requires a comprehensive assessment of the recipient’s situation, this case illustrates how assessment is affected by digitalisation.

Given this context, this study addresses the following research question: how do institutional provisions shape individual assessment in social assistance within screen-level bureaucracy? To answer this question, we applied qualitative content analysis (Dey, 1993) to policy documents and guideline revisions related to social assistance decision-making. The revisions make it possible to scrutinise the ambition to file instructions for consistent individual assessment and to identify which aspects are difficult to structure as formal instructions.

This paper contributes to screen-level bureaucracy theory by conceptualising the often overlooked organisational conditions that shape individual assessment. Understanding these conditions is critical for evaluating accountability and transparency in digitalised welfare administrations, which are designed to promote consistency while accommodating individual needs. These conditions become more apparent as welfare systems move toward automation and artificial intelligence (AI)-driven delivery (Carlsson, 2025; Germundsson, 2022; Petersen et al., 2020; Petroons et al., 2025). This study underscores the need to carefully design assessment processes (Coles-Kemp et al., 2020), including mechanisms to safeguard consideration of vulnerable recipients and social inclusion (Halliday et al., 2024; Schou and Pors, 2019). The paper proceeds as follows: we first present the analytical framework for individual assessment and social assistance within screen-level bureaucracy; next, we introduce the Finnish case and research design; then, we present empirical findings; and finally, we discuss the implications for theory, policy, and future research.

The digitalisation of welfare administrations is increasingly justified on grounds of efficiency and organisational rationalisation (Busch and Henriksen, 2018; Carlsson, 2025; Høybye-Mortensen, 2019). This development reflects Bovens and Zouridis’s (2002) concept of screen-level bureaucracy, in which front-line decision-making is reconfigured as ICT systems and standardised procedures, and rules guide decision-making, reducing professional judgement and face-to-face interactions.

Screen-level bureaucracy signals a transformation in governance and practical implementation. Grounded in Weberian rule-based governance and Mintzberg’s (1979) machine bureaucracy, screen-level bureaucracy prioritises predictability, transparency and reduced discretion as core organisational aims (Muellerleile and Robertson, 2018; Newman et al., 2022). Related research on e-bureaucracy and e-government shows similar dynamics of ICT-enabled coordination and tightened control (Buffat, 2015; Cordella and Tempini, 2015; Twizeyamana and Andersson, 2019). Building on this literature, this study uses screen-level bureaucracy to describe a highly digitalised administration defined by mass processing, rigid rules, and reduced professional expertise and discretion (Bovens and Zouridis, 2002).

Screen-level bureaucracy promises advantages: standardisation is expected to enhance consistency, accountability and equal treatment (Bovens and Zouridis, 2002). By substituting schematic rules for professional judgement, decision-making becomes more uniform and less susceptible to individual bias or error (Bovens and Zouridis, 2002; Buffat, 2015). Digitalised processes aim to narrow the gap between “policy as written” and “policy as performed”, improving the accuracy of implementation (Busch and Henriksen, 2018; Ranerup and Henriksen, 2022). However, these benefits raise questions about their impact on individual assessment in social assistance, which is based on fundamentally different principles from standardised procedures (Lipsky, 2010).

Individual assessment is central to social assistance and goes beyond technical eligibility calculation to encompass an evaluation of recipients’ specific circumstances and needs (Hussénius et al., 2025; Jewell, 2007). In street-level bureaucracy, such assessment often relies on face-to-face encounters and professional judgement (Evans, 2010, 2013; Lipsky, 2010). Professional expertise shapes how hardship is understood and legitimises the exercise of discretion (Blomberg et al., 2013; Evetts, 2009; Ponnert and Svensson, 2016). This recognised competence entitles organisations to delegate discretion to frontline workers (Evans, 2010). While these circumstances allow for a comprehensive and context-sensitive assessment, it is resource-intensive and not immune to errors, leading to inconsistencies or unintended outcomes, especially under time pressure or institutional constraints (Jewell, 2007; Lipsky, 2010; Nordesjö et al., 2022).

In screen-level bureaucracy, administrative workers carry out assessments, mediated by guidelines and ICT systems designed to standardise decision-making. Recipients’ information is gathered digitally, which restructures their role in the process. Applicants submit their applications and articulate their needs through digital interfaces, creating challenges for those with limited skills or access (Hofmann et al., 2024; Meers et al., 2025). Combined with remote interactions and reduced professional expertise, these conditions may constrain flexibility when it comes to considering individual circumstances, leading to inadequate or arbitrary assessments (Bovens and Zouridis, 2002; Ranerup and Svensson, 2023; Zouridis et al., 2020).

At the same time, granted discretion remains an important formal component of social assistance (Evans, 2010; Molander, 2016). Here, granted discretion refers to the explicit delegation of judgement to workers (Hupe, 2013;, cf. Hupe’s “discretion-as-used”). Reforms can formalise expectations of discretion, and these are signalled through guidelines. In screen-level bureaucracy, however, this granted discretion is expected to be strictly limited by rules and ICT systems.

Whether in street-level or screen-level bureaucracy, decision-making is primarily steered by guidelines, variously described as organisational rules, manuals, or procedures. Guidelines translate legislation into practical instructions and aim to promote accountability, consistency, and transparency (Pottie and Sossin, 2005; Sossin and Smith, 2003; Weeks, 2016). In screen-level settings, their role becomes even more pivotal.

Given this importance, a structured approach is needed to analyse how guidelines shape individual assessment. This study elaborates on Daly’s (2023) analytical framework regarding binding and non-binding guidelines. Binding instructions use imperative language that prescribe a single course of action and promotes uniform compliance. Non-binding instructions employ permissive, open-ended, or vague formulations that leave room for interpretation and adjustment (Baker Collins, 2016; Daly, 2023; Sossin and Smith, 2003). Open-ended instructions create room for manoeuvre by setting broad but clear expectations, such as “evaluate the recipient’s functional capacity”, while vague instructions, such as “if necessary”, lack specificity and their intent is unclear. Both types of instructions can generate ambiguity, and multi-part or complex instructions often further complicate interpretation (Baker Collins, 2016). Granted discretion may take either binding or non-binding forms.

Finnish basic social assistance exemplifies screen-level bureaucracy through extensive digitalisation and centralisation. In 2017, responsibility for granting social assistance was transferred from municipal welfare offices to the National Insurance Institution of Finland (hereinafter Kela). This transfer restructured implementation into a nationwide administration (Varjonen, 2020) and standardised locally developed practices. Preventive and supplementary social assistance (together accounting for less than 10% of total social assistance expenditure), which involve greater discretion, remained under municipal responsibility (since 2023 they have been handled by wellbeing services counties). Centralisation aimed to standardise implementation nationwide; however, the role of social assistance as individually assessed financial aid of last-resort remained unchanged (Varjonen, 2020). In 2022, approximately 80% of applications were submitted digitally (Gov. bill, 2022). Even when applications are brought to an office, they are processed separately through the centralised benefit processing system, without face-to-face contact.

In Finland, social assistance is often received as a supplement to first-tier benefits when these are insufficient to cover monthly expenses and the applicant has no wealth or assets. An individual assessment is used to evaluate income and acceptable expenses, as well as the person’s individual living conditions. The assessment is particularly important in situations where social assistance is reduced due to the recipient’s non-compliance with their obligations (Blomberg and Kroll, 2020; Veilahti, 2020). The national guidelines formulated by Kela’s legal unit are based on its interpretation of the (Finnish) Social Assistance Act. These guidelines are frequently revised (288 updates since centralisation, as of July 2025). ICT systems support decision-making by calculating benefit amounts based on income and expenses reported in the digital application. However, the information provided by applicants must be manually verified as part of the process (see Figure 1).

In our empirical analyses, we focused on the reduction of social assistance. Reduction functions as a sanction and is applied when a recipient aged 17–64 refuses to accept work, participate in activation measures, or otherwise neglects their livelihood without a valid reason (Social Assistance Act, 1997). Individual assessment comes into play once non-compliance has been confirmed by the benefits officer. The first reduction can be up to 20%, and in cases of repeated misconduct, up to 40% (prior to February 2026). According to social assistance law, “A reduction can only be made if it does not endanger the minimum living standards required for a dignified life and if the reduction cannot be considered otherwise unreasonable. The reduction can last for a maximum of 2 months at a time, starting from the date of refusal or neglect.” (Social Assistance Act, 1997). The reduction of social assistance is a matter of concern because the wording of the Social Assistance Act necessitates individual assessment. Furthermore, the legislation has remained unchanged since its enactment, so changes in Kela’s guidelines cannot be attributed to legislative amendments. The intention of the law was that, based on individual assessment, the reduction could also be less than 20% or 40%. In practice, however, reductions have been applied as flat rates. There have also been unexpected developments in the use of reductions; for example, in 2023 their number increased significantly (Kela, 2025).

The centralisation of social assistance in a national agency, where implementation relies heavily on ICT systems, national guidelines, and the capacity to process applications on a mass scale, provides a strong context for studying individual assessment within screen-level bureaucracy (see Duméz, 2015). The following section presents the research design in more detail.

The research material includes Kela’s guidelines for basic social assistance, the Social Assistance Act, and government bills (see Table 1). The Social Assistance Act provides the legislative foundation for individual assessment. Government bills on centralisation and a subsequent reform shaped the decision-making process by defining the operational parameters of screen-level bureaucracy. The guidelines translate legislation into instructions that steer decision-making and individual assessment.

The primary data consist of guidelines related to the reduction process and the government bills. All guidelines issued since the centralisation of basic social assistance were reviewed. The first guideline analysed, dated 21 December 2017, was chosen because the major technical problems had been resolved by the end of that year (Ministry of Social Affairs and Health, 2020). Five other versions between 2017 and 2023 were selected based on substantial changes to the reduction process (see versions in Table 1). While the legislation remained largely unchanged, the revisions of the guidelines reflect ongoing efforts to experiment with different formulations to steer assessment within screen-level bureaucracy.

Data were analysed using qualitative content analysis (Dey, 1993), supported by NVivo. Particular attention was paid to how the guidelines enabled or constrained individual assessment. In addition to explicit language, the overall message was considered (Krippendorff, 2019), drawing on the idea that texts are interpreted within their institutional context (Daly, 2023). In screen-level bureaucracy, guidelines are designed to provide a standardised, step-by-step process for individual assessment that requires minimal interpretation, as professional expertise is not expected.

We began the analysis by inductively categorising the bills in NVivo into categories such as reform goals, cooperation between Kela and welfare offices, discretion in social assistance, consideration of individual circumstances, service provision, reduction process, and ICT systems in implementation. The text was analysed with attention to how they describe implementation within the centralised system and describe the nature of social assistance.

Next, we analysed the guidelines. First, we coded them inductively in NVivo into categories representing phases of the reduction process, such as duration of reduction, 20% reduction, cooperation with social services, fictitious case examples, and criteria for exemption from registering as a jobseeker.

In the second stage, we categorised the content of the guidelines in NVivo into binding and non-binding instructions. Binding instructions used rigid or imperative language, enforcing uniform application, for example, “the applicant must be informed of the threat of reduction”. Non-binding instructions used permissive, vague, or open-ended language, leaving room for adjustment, for example, “social services may be called if needed”, “exceptionally, reduction can continue over six months”, or “consider the recipient’s ability to function”.

Some instructions combined both forms, for example, starting with a permissive phrase and ending with a binding statement: “if the benefit reduction would make the recipient’s overall situation unreasonable, the basic amount must not be reduced”. We categorised these based on whether the binding or non-binding expression appeared first. Additionally, instructions explicitly granting discretion were categorised and could appear in either form. For example, “the time limit can be considered case-by-case” or “discretion must be used”, indicating that discretion was delegated to consider individual circumstances.

Finally, we compiled the main changes in each guideline version into a summary Word document, comparing the main changes in the guidelines and binding and non-binding instructions as well as the articulation of the discretion granted. This provided an overall picture of how the steering of individual assessment evolved. Findings are presented chronologically, starting with government bills, followed by analyses of the selected guideline versions.

The bill on centralisation restructured the implementation of social assistance by prioritising standardisation and efficiency without explicitly addressing the conditions for individual assessment. It stated that basic social assistance should be implemented consistently and require only limited discretion (Gov. bill, 2014, pp. 4, 23–24). Kela’s integrated data systems, digital application, and national guidelines were presented as tools to promote equal treatment and reduce stigma, while reinforcing standardised decision-making (Gov. bill, 2014, p. 21). Individual assessment was discussed mainly as a technical matter, although the bill noted that reductions require case-by-case assessment and knowledge of the recipient’s individual and family circumstances (Gov. bill, 2014, p. 8).

Contrary to the stated rationales for the reform, the bill also highlighted the importance of discretion: “Because of the last-resort nature of social assistance, and the regulations related to it, it is considered necessary to retain a considerable number of provisions that require discretion. With discretion, it is possible to evaluate each person’s individual situation and target support to ensure the level necessary for a dignified life.” (Gov. bill, 2014, p. 7). However, it did not address how such an assessment could occur digitally and without face-to-face contact or social work expertise. Consequently, benefits officers were tasked with identifying applicants’ needs for social services based on documents. Individual assessment was transferred into screen-level bureaucracy, with little consideration for practical implementation.

The first bill overlooked legislation concerning ICT systems, hindering cooperation and data exchange between Kela and municipal welfare offices, ultimately necessitating a new bill. The 2016 bill addressed authorisation and data sharing, such as transferring medicine payment commitments to pharmacies, but focused mainly on technical tasks. Kela was unable to access social services’ data, and strict regulations restricted data sharing, particularly regarding reductions (Gov. bill, 2016, pp. 6–7, 16–17).

Furthermore, the 2016 bill reinforced recipient self-determination in the application process, giving them greater responsibility for navigating a fragmented system. For instance, it required applicants to explicitly request transfers for expenses managed by welfare offices (Gov. bill, 2016, pp. 5–6). Both bills sidelined the challenges faced by vulnerable groups, posing a risk of further exclusion in a centralised and digitalised system.

The 2022 bill sought to address the shortcomings identified in earlier bills, particularly those linked to centralisation and digitalisation, which had weakened individual assessment and detached social assistance from social work (Gov. bill, 2022, pp. 24–25). It acknowledged the barriers faced by vulnerable groups, such as a lack of digital skills and devices, and recognised that the absence of Kela offices exacerbated exclusion. By reintroducing consideration of individual circumstances (Gov. bill, 2022, pp. 31, 37–38, 53 and 69) and reaffirming social assistance as part of social services (Gov. bill, 2022, pp. 31–32), the bill signalled a shift towards restoring discretion and improving accessibility, albeit within the constraints of screen-level bureaucracy.

The bill emphasised the importance of social services’ information for individual assessment, but enhancing this meant increasing the administrative burden on the authorities. It introduced a requirement for benefits officers to consult social services before applying a 40% reduction and to consider both the recipient’s standpoint and the social worker’s judgement in the reduction process (Gov. bill, 2022, pp. 51–52, 72). However, the final decision-making authority remained with Kela, and the role of social workers was not guaranteed in the process.

The bills suggest that recurring reforms involve the persistent challenge of incorporating individual assessment with centralised, standardised administrative structures. While the 2022 bill sought to mitigate rigidity by integrating social workers’ expertise and intensifying inter-organisational cooperation, it left largely unclear how standardised procedures enable the exercise of discretion at Kela. Moreover, these changes undermined the original efficiency goals of centralisation.

Initially, the reduction guidelines outlined a straightforward process, but implicitly left room for adjustment and mentioned discretion. Many instructions remained non-binding, such as those concerning reasonableness, for instance: “If a reduction of 20% or 40% is unreasonable, the reduction may be lower” (Guidelines, 2017, p. 57). These permissive formulations and the lack of clarification of vague expressions did not mandate specific actions or encourage discretion, meaning that application remained ambiguous.

Furthermore, the guidelines did not clarify the legislative requirement to consider reasonableness, instead stating: “Additionally, it is considered whether the reduction could endanger the minimum living standards required for a dignified life and whether the reduction can otherwise be deemed reasonable. If the municipal social worker deems the reduction unreasonable, it is generally not applied” (Guidelines, 2017, p. 57). This phrasing obscured the significance of the legislative requirement and left room for interpretation, even when social workers considered it unreasonable. Moreover, by framing the reduction as reasonable rather than unreasonable, the guidelines subtly shifted the normative emphasis.

These early guidelines suggested a clear-cut process. Discretion was clearly granted only in the instructions for 40% reductions, which required “particularly careful consideration” when assessing a valid reason for neglect or refusal (Guidelines, 2017, p. 60). Although two illustrative examples followed, substantial room remained for workers to determine what qualified as a valid reason.

Instead of emphasising the exceptional nature of reductions, the guidelines implied their continuation over several months. In 2018, they stated that the reduced amount would be copied by default into subsequent decisions (Guidelines, 2018, p. 61; see also Veilahti, 2020). This conflicted with the two-month limit set by legislation for reductions. The requirement for individual assessment in follow-up decisions was unclear, obscuring the legislative intent while simplifying the process. Overall, vague instructions required interpretation, which could make the guidance for individual assessment ambiguous.

After four years, the reduction guidelines appeared more structured, listing four mandatory preconditions for assessing reductions and emphasising discretion. The process was foregrounded by the following statement: “reduction always involves especially careful overall consideration of each case” (Guidelines, 2021, p. 194; Guidelines, 2022, p. 10).

Despite this emphasis on discretion, new ambiguities emerged due to open-ended expressions and contradictions within the preconditions. For example, the first precondition for assessing neglect states: “Neglect by a person can be either active, involving a deliberate refusal of an offer, or passive, involving a failure to make contact or take other similar actions” (Guidelines, 2021, pp. 193–194; Guidelines, 2022, p. 12). This wording allows almost any action or inaction to qualify as neglect. However, the third precondition, assessing whether the reduction endangers minimum living standards or is otherwise unreasonable, requires consideration of the recipient’s efforts and the means available to remedy the refusal or neglect (Guidelines, 2021, p. 195; Guidelines, 2022, p. 12). These instructions lack coherence: one applies a broad definition of neglect, while the other implies assessing the recipient’s impediments. This inconsistency creates ambiguity and may lead to uneven interpretations.

The recognition of vulnerable social assistance recipients was increasingly acknowledged in the guidelines, signalling that these groups required explicit mention to ensure that they were given due consideration. Vulnerable groups listed for special scrutiny included families with children, young people aged 18–24 without vocational education, and people in particularly difficult circumstances (Guidelines, 2021, p. 194; Guidelines, 2022, p. 14). While these instructions required attention to individual circumstances, open-ended terms such as “particularly difficult circumstances” left room for interpretation.

Overall, the guidelines reflect a gradual shift towards consideration of individual circumstances by explicitly granting discretion. However, the combination of detailed instructions and expectations for interpreting open-ended instructions may obscure how such instructions should be applied in practice (see Baker Collins, 2016).

The August 2023 guidelines, revised after the 2022 reform, became less detailed but required discretion more explicitly than before. Specific instructions, such as identifying vulnerable groups and case examples, were removed. The guidelines strongly emphasised the need to exercise discretion in reductions, potentially signalling an increased reliance on workers’ judgement. Earlier instructions implying that the reduction could be continuous were also removed, aligning the interpretation with its intended exceptional nature.

Several previously vague instructions were simplified, thereby clarifying the reduction process. For instance, hearings that could previously be conducted by phone were now restricted to written form only (Guidelines, 2023, p. 14). Instructions concerning reductions less than the maximum amounts were removed, suggesting fixed percentages. While this increased consistency, it limited the possibility of adjusting reductions to individual circumstances.

Although some instructions narrowed interpretation, others continued to rely on open-ended criteria, leaving significant room for interpretation, for example: “… special attention must be paid to the recipient’s ability to care for themselves and to whether a prolonged reduction of social assistance could lead to an unreasonable situation or endanger the minimum living standards required for a dignified life. This includes, for example, situations where access to essential nutrition, housing, or independent living necessary for maintaining health or viability is jeopardised.” (Guidelines, 2023, p. 14.). These revisions reveal the paradox of promoting discretion while limiting it through a few more standardised expectations. Furthermore, the revisions show the difficulty of formulating instructions that can guarantee assessment through data and ICT systems alone.

Our analysis reveals a persistent contradiction between the legislative intent for individual assessments and the bureaucratic rationalisation processes underpinning welfare delivery. The guidelines for individual assessment varied, leaving considerable room for interpretation and adjustment. ICT systems, introduced in the centralisation bill to standardise decision-making and improve efficiency, narrowed the scope of comprehensive assessment from face-to-face interaction to document-based evaluation. The transition to a centralised administration exposed legislative barriers to data sharing as an unexpected challenge. Inter-organisational exchange could not be structured to support individual assessment. While the digitalised, nationwide system aimed to reduce stigma and promote equal treatment, it inadvertently weakened the ability to consider individual circumstances, a limitation inherent in a system designed for mass processing.

The subsequent reform sought to restore individual assessment by integrating social workers’ expertise and promoting inter-agency cooperation. However, these measures more likely increased administrative complexity. Decision-making authority remained with benefits officers, marginalising social workers’ professional judgement.

The revisions to the guidelines illustrate the difficulty of translating the legal requirements for considering individual circumstances into practical instructions. Variations between simplified guidance and explicit mandates for discretion left ambiguity intact. Thus, while the guidelines implicitly or explicitly enabled case-by-case consideration, they did not guarantee a recipient-oriented perspective, which may undermine consistency. Despite the fact that the 2022 bill emphasises discretion, reductions have increased since 2023 and follow fixed thresholds (Kela, 2025), suggesting that discretion may remain largely symbolic.

Finally, the highly digitalised system caused difficulties for recipients. Centralisation increased applicants’ responsibility to articulate their needs in writing, often through digital interfaces (Blomberg and Kroll, 2020). The findings question the feasibility of screen-level bureaucracy in delivering individually tailored support.

We asked how institutional provisions shape individual assessment in social assistance within screen-level bureaucracy, using Finnish social assistance as the empirical case. Drawing on the findings above, screen-level bureaucracy was introduced to deliver consistent and accountable individual assessment, but clear-cut steering through legislation may not be achievable, because assessment inevitably requires flexibility. In Finland, persistent ambiguity in the guidelines suggests that these goals have only been partially achieved. Nevertheless, extensive data alone appeared insufficient for the comprehensive understanding required for a careful assessment. Consequently, consistency may be pursued at the expense of genuine consideration of individual circumstances. This can lead to decisions that are neither consistent nor responsive (Ranerup and Henriksen, 2022), creating risks of digital exclusion for vulnerable recipients (Hofmann et al., 2024). We identify three key challenges for individual assessment in screen-level bureaucracy.

First, reconciling the flexibility needed to take individual circumstances into account with prescriptive guidelines remains inherently problematic. In the Finnish case, requirements such as assessing reasonableness and minimum living standards were left open-ended, leaving interpretation to workers (Wittberg et al., 2024). These legal requirements underscore that the implementation accuracy involves not only replicating the letter of the law but also embodying its spirit. This means that decisions remain accountable when individual circumstances are also taken into account. Such flexibility, grounded in human interpretation, conflicts with rigid guidelines. In the future, this tension may be further intensified when automation is applied, as it requires exhaustive rules, which in turn entails making value judgements (Ranerup and Svensson, 2023). Nevertheless, scrutinising guidelines is crucial for detecting shortcomings in implementation, identifying misalignments between intended objectives and practice (Sossin and Smith, 2003), and shedding light on how organisations interpret legislation (Bovens and Zouridis, 2002).

Second, decision-making based on digitalised data limits comprehensive individual assessment in social assistance. Digital interfaces constrain recipients’ ability to explain their situation, while data structures do not capture contextual information beyond digitalised data (see Halliday et al., 2024). This may constitute a critical deficiency, because accountable individual assessment depends on complete and accurate information about recipients’ circumstances (Jewell, 2007). In the Finnish case, the assessment of reasonableness illustrates this deficiency: neither the guidelines nor the bills explained how such evaluations should be carried out using digitalised data within ICT systems. Moreover, a lack of recipients’ information may hinder the exercise of discretion, which may lead to decisions to default to rigid practices within screen-level bureaucracy. Consequently, reductions may potentially occur when full contextual information is unavailable (see Marienfeldt, 2024).

Third, accountability and transparency may be compromised when decision-making authority is concentrated among administrative workers operating under ambiguous guidelines and ICT constraints. Delegating open-ended legislative principles to professionals has long been criticised for granting wide authority over eligibility (see Wittberg et al., 2024). In screen-level bureaucracy, this criticism intensifies: administrative workers, unlike professionals, have limited training and are bound by system rules, yet they face mass processing pressures that further increase the risk of inappropriate or inconsistent assessment (cf. Jewell, 2007; Nordesjö et al., 2022). Consistent with our findings and previous research, screen-level bureaucracy may still allow for considerable discretion (Marienfeldt, 2024). Thus, screen-level bureaucracy does not necessarily reduce accountability concerns, but rather transforms them, creating new blind spots (Bovens and Zouridis, 2002). Accountability mechanisms within ICT systems would therefore be needed to safeguard decision quality and ensure recipient-oriented decision-making in social assistance.

Taken together, these challenges reveal that individual assessment in social assistance cannot be fully codified into rules and ICT systems. In practice, discretion remains indispensable yet difficult to capture consistently in guidelines. When discretion is left largely unstructured, the gap between legislative requirements and implementation persists, undermining the goal of digitalisation. ICT systems could, in principle, enable retrospective checks on consistency and decision quality, but we found no evidence of such use. This makes transparency essential to understanding how discretion operates in digitalised systems (Bovens and Zouridis, 2002). These deficiencies in screen-level decision-making have direct consequences for recipients: when systems cannot accommodate contextual information, vulnerable recipients face reduced access and the risk of unequal treatment, eroding the purpose of a last-resort safety net. These findings underscore the need to clarify recipients’ role in digitalised welfare reform and to design interfaces capable of capturing individual needs effectively (Halliday et al., 2024; Meers et al., 2025).

Several limitations should be noted. First, this study examined a single centralised system, which may limit its generalisability to decentralised or hybrid governance contexts, even though similar standardisation goals are evident elsewhere (cf. Lunder, 2016; Petroons et al., 2025; Wittberg et al., 2024). Second, although our analysis focused on guidelines and ICT systems, we did not examine how these interpretations translate into actual practice, where decisions are shaped by other factors, such as leadership, managerial practices, workers’ educational background and work experience, in addition to written guidelines and rules (Petersen et al., 2020). Third, we did not explore the interaction between workers and ICT systems, which merits closer scrutiny to understand how technology influences the decision-making process (see, e.g. Høybye-Mortensen, 2019).

Future research should build on these findings and examine how organisational culture and work experience shape the interpretation of guidelines and influence how workers use the discretion granted to them within screen-level bureaucracy. Comparative studies could contrast interpretations of guidelines and written justifications for decisions between centralised and decentralised systems (cf. Hölsch and Kraus, 2004; Minas et al., 2012). Such comparisons are important because centralisation can significantly reshape decision-making. For example, in Finland, social assistance reductions became much more common than in municipal social services prior to centralisation (Blomberg and Kroll, 2020). Furthermore, the effects of sanctions on recipients in the digitalised welfare administration warrant closer analysis. As welfare administrations increasingly integrate automation and AI, research should examine how these technologies operationalise legislation, reconfigure individual assessment, and affect social inclusion and welfare delivery (Carlsson, 2025).

This study examined how institutional provisions shape individual assessment in social assistance within screen-level bureaucracy. Using the Finnish case, we identified challenges that hinder accountable individual assessment in social assistance. The case showed enduring ambiguity in guidelines, despite reforms efforts to standardise processes with the support of ICT systems.

Recurring legislative efforts indicate persistent problems in ensuring such assessment within highly standardised and digitalised system. This system may curtail discretion and weaken accountability when decision-making relies on standardised procedures while still requiring workers’ judgement. Our findings advance theory concerning screen-level bureaucracy by showing that formally granted discretion can remain largely symbolic under ill-suited conditions, thereby undermining recipient-oriented decision-making. Moreover, guidelines and ICT systems intended to ensure consistency can inadvertently create new accountability and fairness issues when safeguards for individual assessment are lacking, particularly for vulnerable recipients.

These insights are becoming increasingly relevant as welfare systems adopt automation, AI, and data-driven forms of reasoning. Integrating these technologies requires careful attention to the role and purpose of individual assessment, including distinguishing routine tasks from those requiring professional judgement and designing interfaces that capture nuanced needs and recipients’ perspectives. Further investigation should continue addressing normative questions concerning the safeguarding of individual assessment, while promoting consistency in digitalised systems. By highlighting organisational challenges within screen-level bureaucracy, this study supports research on administrative design and offers guidance for building welfare systems that balance efficiency and fairness.

Ethics approval was not required for this study.

We have no material from other external sources.

The first author thanks colleagues and reviewers in the Sociology Unit at the University of Turku and the anonymous reviewers for their valuable comments, and the Turku University Foundation for the personal grant providing financial support during the research period.