Reducing social security spending

Driven by the perception of growing ’welfare dependency’, the design of Universal Credit as a ‘digital first’ benefit has its origins in the austerity-driven imperative of the Coalition Government to reduce taxpayer spending amid growing disquiet over what was seen to be an unsustainable social security bill (Taylor-Gooby and Martin, Reference Taylor-Gooby, Martin, Park, Curtice, Thomson, Phillips, Johnson and Clery2008). With deficit reduction taking precedence over all other manifesto commitments, digitalisation was intended to cut costs by streamlining claims and payment processing, reducing Jobcentre footfall and freeing up staff time for claimants with the most complex needs. Between 2016 and 2018, 100 Jobcentres – about 15 per cent of the DWP estate – closed their doors to the public (Finn, Reference Finn2018). Yet despite a swingeing programme of Jobcentre closure and job cuts that has seen thousands of staff in customer-facing roles made redundant, during the past decade, UC’s costs have in fact risen to more than six times the original budget (NAO, 2018). Software and systems for an early version of UC – known as the ‘live system’ – were written off at a cost to the taxpayer of around £40 million (Ballard, Reference Ballard2013). The heavy reliance on outsourced expertise based on multi-million pound contracts with a small number of IT suppliers also proved to be costly and inherently risky. In a highly critical report, the National Audit Office was sceptical as to whether any of the promised efficiency savings would ever be realised (NAO, 2011).

In response to such criticism, in 2013, the entire project was re-engineered using ‘agile’ techniques ^{Footnote 9} and ‘Full Service’ Universal Credit was launched as a ‘digital first’ benefit. As part of the Government’s wider digital transformation strategy, the DWP began to re-focus its IT strategy towards developing greater in-house capacity and since this time has hired more than 1,000 data scientists (Clement-Jones, Reference Clement-Jones2020). In 2020, it was announced that the DWP’s main external IT provider ^{Footnote 10} will join with DWP staff to create a single digital function (Say,Reference Say2020). However, the government’s ambition to have ‘one of the most digitally skilled populations of civil servants in the world’ (DWP, 2020a) has come at a cost. The DWP is now the government department with the highest spending on digital services and, by 2018, UC was four times more expensive to operate than originally forecast (House of Commons Work and Pensions Committee, 2018). OBR forecasts indicated that UC will probably end up costing significantly more to administer than the legacy system (OBR, 2020) ^{Footnote 11} .

Driving down fraud, error and overpayments

Another stated goal of automation was to drive down the error, fraud and overpayments that were seen to characterise the legacy system of tax credits. Claimants of Working Families Tax Credit (WFTC), introduced in 1999, completed a form based on their forecast earnings for the year ahead, giving rise to significant overpayments (Godwin and Lawson, Reference Godwin and Lawson2012). With up to 14 per cent of expenditure also believed to be fraudulent, WFTC was withdrawn in April 2003 and replaced with Working Tax Credit and Child Tax Credit. Using a system of annual retrospective reconciliation based on actual earnings, the prevalence of overpayments, underpayments and fraud reduced; but concerns remained. By replacing the system of self-completed form filling with automated monthly assessment in UC, the aim was to improve payment accuracy and lessen the financial hardship that recovered overpayments (and underpayments) could cause for people on low incomes (Millar and Whiteford, Reference Millar and Whiteford2020). Risk-based ID verification and data analytics in UC now cross-references claimant data with a range of third parties for checks of identity, address, income and occupancy (Department for Digital, Culture, Media and Sport, 2019). Data matching technologies also help identify fraudulent and erroneous claims arising from undeclared capital, income and partners (Public Accounts Committee, 2020). Automated systems to confirm the existence of dependent children and to verify childcare and housing costs are now central to the online application. A new ‘transaction risking’ system based on algorithms and profiling tools is also under construction (Public Accounts Committee, 2020: Q24).

But here too, UC’s performance to date has singularly failed to match the ambition. The DWP’s own research showed that only a third of applicants were able to prove their identity using Gov.UK Verify’ – an online system for authenticating personal identity credentials – obliging them to present documentary evidence in person at a Jobcentre (Glick, Reference Glick2018). Delays in authorising ID mean that some applicants wait significantly longer than the stipulated five weeks before receiving their first payment (Griffiths et al., Reference Griffiths, Wood, Bennett and Millar2020). Because ID verification is required before the claim becomes live, such data never find their way into the official statistics recording the timeliness of payment. ‘Verify’ also proved to be susceptible to fraud. The ability to verify a person’s identity remotely meant that fraudsters posing as applicants, or who duped applicants into disclosing their personal details, were able to apply for an advance loan – worth up to a full month’s predicted entitlement – without ever meeting Jobcentre staff face to face. Around 100,000 claims, worth between £98 million and £147 million, were suspected to have been made fraudulently (NAO, 2018: 7). In September 2019, the DWP tightened its policy and an advance loan request was only granted following a face-to-face interview. Fraud subsequently dropped (NAO, 2020), but when this restriction was removed in response to the Covid-19 pandemic, levels surged again. Late in 2020, a new ‘Confirm Your Identify’ service was introduced. Its ability to reduce the incidence of fraud is as yet unknown.

Overpayment and underpayment of UC is also highly prevalent. In 2019/20, 9.4 per cent of total UC awards (worth £1.7 billion) were overpaid, the highest rate for all benefits (DWP, 2020b), and only a little under the highest ever recorded overpayment rate of 9.7 per cent for tax credits (Public Accounts Committee, 2020:Q15). HMRC and employer error is also rising. Between 2017 and 2018, there was a fourfold escalation in earnings disputes (NAO, 2018), with up to 5,700 people per month affected (Booth, Reference Booth2019). HMRC have also acknowledged that around 10 per cent of PAYE schemes have data quality issues affecting the accuracy of individual tax records which could potentially result in an underpayment or overpayment of UC ^{Footnote 12} . Unlike the legacy system, overpayments due to official error must now be repaid, with the amounts owed automatically deducted from a claimant’s UC payment. There has also been a higher than expected proportion of claimants with ‘atypical’ forms of employment whose earnings are not captured by HMRC’s PAYE system including self-employed people and those working a small number of hours. Here, earnings must be self-reported monthly using the online portal. Failure to do so within strict timeframes can result in a reduced or nil payment.

Behaviour change

Automated processes are also central to UC’s goal of promoting behaviour change. Monthly adjustment of the award in response to changes in earnings in real-time is intended to make the financial rewards of working and of earning more, more visible and motivating to claimants (DWP, 2014). Less explicit was the underlying moral intent. Framed within a set of normative values about people ‘who support themselves solely through work’ (DWP, 2014: 9), behaviour modification goals designed to encourage personal responsibility and reduce ‘welfare dependency’ have literally been coded into UC’s design. Designed to mimic a monthly salary, a single, integrated payment in arrears is intended to make claimants responsible for paying their rent, as people who ‘manage without support from the State’ do (DWP, 2014: 12). Housing Benefit, by contrast, was often paid direct to the landlord. Automated deductions enforce ‘social and financial responsibilities’ by obliging claimants to repay debts, fines and child maintenance based on fixed, generally stricter and less negotiable, repayment terms than under the legacy system. Not simply a convenient repository of messages and payments, the journal and ‘to do list’ play a central role in the enforcement of work conditionality. Formally accepting the online ‘claimant commitment’ is a mandatory requirement of benefit receipt. Failure to do so can cause the payment to stop, requiring the claimant to request a ‘mandatory reconsideration’ to have the payment reinstated. The journal is also key to evidencing the job search which many unwaged and low-earning claimants are required to undertake.

Although never officially articulated, the ceding of human agency to automated assessment and payment systems may also have been motivated by the desire of policymakers to change the culture and behaviour of DWP staff who may have been seen as contributing to claimants’ ‘welfare dependency’ and to programme deadweight. Virginia Eubanks speculates that a key driver of digitalisation in U.S. state welfare systems was to undermine the personal relationship between caseworkers and their clients (Eubanks, Reference Eubanks2018: 52). Reducing the scope for ‘street level bureaucracy’ ^{Footnote 13} may also have informed the decision to replace personal advisers with work coaches in UC, many of whom were recruited from outside the DWP, and to replace local offices with telephone contact and service centres whose staff are located remote from their local communities.

Given UC’s explicit behavioural intent, a deep understanding of the people whose behaviour the benefit was intended to change might have been expected to inform the system design from the outset. However, in the early phase of ‘reimagining the customer experience’ (McKinnon, Reference McKinnon2020), little attempt was made to test the assumptions on which UC’s digital design were based, the effects they had on claimants, or whether they actually worked in the manner intended. Local authorities and third sector bodies with long experience of supporting benefit claimants and disadvantaged groups, though involved in a series of open consultation exercises, were largely excluded from the design process. Early warnings that the default online application was not ‘grounded in reality’ were not acted upon (Hitchcock, Reference Hitchcock2012). Making and managing the claim requires an ‘ecosystem’ of resources including a functioning, internet-enabled device and reliable internet access (Coles-Kemp et al., Reference Coles-Kemp, Ashenden, Morris and Yuillel2020). Home broadband and mobile data plans are expensive and the number of libraries and jobcentres where claimants could secure free internet access had decreased significantly in the wake of the government’s austerity cuts (Alston, Reference Alston2019a: 14). However, concerns over digital access, together with other potentially problematic aspects of Universal Credit’s design, went unheeded.

In a rare display of frankness, when asked why government decision makers frequently fail to talk to the people who are the intended beneficiaries of new policies, the senior civil servant responsible for the digital aspects of UC admitted, ‘the really, really terrible or brilliant answer is that it would just never occur to someone to do that’ (Trendall, Reference Trendall2017). Somewhat presciently, early DWP research examining the attitudes of potential customers towards UC’s proposed design – including the online application and single, integrated monthly payment in arrears – pinpointed many of the risks and challenges that have since proved to be problematic, including potential budgeting difficulties, increased scope for fraud and error and insufficiently compelling financial work incentives (Rotik and Perry, Reference Rotik and Perry2011). A more user-centred, ‘agile’, ‘test and learn mind-set’ based on small scale, ‘iterative testing’ was subsequently adopted by the DWP (DWP, 2012b: 6), but a key recommendation, for ‘using ethnographic methods to track actual day-to-day behaviours and barriers’, was never acted upon (Rotik and Perry, Reference Rotik and Perry2012).

A more constructive dialogue between the DWP and key stakeholders began to emerge after UC was extended from single unemployed claimants to families and people with more complex needs. The DWP’s own research found that more than four in ten applicants needed help to complete an online claim for UC and three in ten needed on-going support with using their digital account (DWP, 2018a: 3). ‘Universal Support’, a local authority-delivered initiative to provide digital and personal budgeting support for claimants, proved to be expensive and geographically variable (DWP, 2016). It was replaced in April 2019 with a new ‘Help to Claim’ service delivered by Citizen’s Advice. The service provides hands-on help for would-be claimants who lack the digital skills or resources to complete the online application unassisted. However, help is only available up to the point that claimants receive their first payment; thereafter, they are expected to self-manage their claim. Telephony and face to face support is available for claimants unable to use or access digital channels, but official policy is to maximise use of the online journal. Consideration of the effects of digitalisation on claimants, moreover, remains narrowly framed in terms of access issues and according to strict technical parameters which focus almost entirely on the end-user interface. There has never been any official recognition, for example, that, regardless of an individual’s technical capability, automated systems in UC might not always operate smoothly or in the best interests of claimants.

Automated monthly assessment

By ensuring that claimants are paid an amount that more accurately reflects the needs and income of the household in ‘real-time’, assessing entitlement monthly in UC is designed to reduce the risk of overpayment. With only one agency to apply to and notify about changes of circumstances, no need to produce wage slips if earnings increase or decrease, and no need to close claims and reapply for different benefits when earnings and personal circumstances change, a key assumption is that automatic adjustment of the payment is also less administratively burdensome for claimants. A ‘smoother’ transition from benefits to work, and greater transparency between the payment amount and earnings, is also central to the policy aim of incentivising work and higher earnings. ‘The design of Universal Credit will make the payoff from taking up work clearer to individuals … than they were on legacy benefits. Also, the transition into work is “smoother” due to reduced transaction costs of closing and reclaiming benefit’ (DWP, 2018b).

This research found that, in practice, many of these assumptions did not hold. For working families in particular, the administrative burden of UC could be greater than the legacy system and the incentive to work reduced. Many of the issues centred upon the automatic calculation and adjustment of the monthly payment and the use of the RTI earnings feed from HMRC. A key issue was that, unlike legacy benefits which generally achieved a steady state once the claim was in payment, UC is designed to be highly dynamic and responsive to change of earnings and household circumstances in real time. It is precisely this volatility that working claimants, in particular, could find challenging and onerous.

A specific problem was the interaction of wages with the fixed monthly assessment period. People with irregular earnings, such as those with erratic shift patterns, on zero hour contacts or working overtime, found that the UC payment could fluctuate in unpredictable ways. Those paid weekly or four weekly, even some of those paid a regular monthly salary, could also experience large, month to month fluctuations in the payment. During certain months of the year, or if wages were paid close to the start or end of a claimant‘s assessment period, or paid early (due to a week-end or bank holiday, for example), two sets of wages would be counted in the monthly assessment period. Even though they had not actually earned any more money, this often resulted in a reduced UC payment or a nil award. Although a higher payment would generally be made the following month, many lost the work allowance to which they would otherwise have been entitled. Often, these were not one-off episodes, but a regular monthly occurrence.

Large oscillations in the award from one payment cycle to the next made it hard for claimants to predict when, or by how much, the UC payment would rise or fall, or if the amount received was correct. With notification of the award posted to the online account only a week before payment was due, claimants with awards that unexpectedly fell or stopped could find themselves with significantly reduced income for the coming month. Income uncertainty, in turn, caused budgeting difficulties, anxiety and stress. In dual earner-families, where the partners had different wage frequencies and pay days, these effects could be multiplied. Sometimes multiple sets of wages were included in the monthly assessment. The aggregation of earnings in couples also made it hard to tell how many sets of wages had been used in assessing entitlement, and therefore whether the payment was correct. If recorded earnings rose above the eligibility threshold for UC, this would automatically end the claim, requiring affected couples to reclaim UC the following month. Another area of difficulty was automated deductions for debts. Couples found that benefit and tax credit overpayments and historical loan repayments, even from before their relationship began, were aggregated and automatically deducted from the UC payment. In working couples, the amounts deducted often fluctuated month to month. With no advance notice given of the amounts that would be taken, this made budgeting challenging, regardless of employment status. By contrast, the earnings rules and annual assessment of tax credits (which generally fixed weekly or four-weekly payments for a year) were said, by many participants who had claimed them in the past, to be easier to understand and manage than the monthly recalculation of UC entitlement.

Accessing contributions towards childcare costs was also said by participants to be more complex and demanding in UC than under the system of tax credits, with payment required upfront and the reinbursement of their outlay subject to serious time lags ^{Footnote 15} . Having to upload evidence of childcare fees and expenditure to the online account every month in the form of validated invoices, receipts and bank statements was a significant burden on the working parent with responsibility for organising and paying for child care; typically the female partner. Claimants often did not know whether the information had been received or was being acted on. The integration of childcare contributions in the single payment, tapered away with earnings, also caused budgeting difficulties and made hard to tell if the amount refunded was correct ^{Footnote 16} .

Complex algorithms underpinning the payment calculation meant that DWP staff were frequently unable to explain why claimants had been under- or over-paid, why their entitlement had reduced or why the payment had unexpectedly stopped. More information is now included on the statement, but the complex way in which entitlement, deductions, earned and unearned income interact to produce the monthly payment was said, by many participants, to be abstruse and impenetrable. Lack of transparency about how the payment is calculated made challenging decisions and resolving errors more difficult than under the legacy system. Dispute resolution was also frequently long and involved, sometimes taking many months to resolve, with claimants passed between work coaches, case managers and staff from debt management. Monthly assessment of entitlement meant that issues resolved in one month could subsequently recur.

For working families in particular, the weight of personal responsibility in uploading information, making calls to the service centre, responding to messages, filling in the journal, checking the monthly statement and challenging decisions could be a heavy administrative burden. Budgeting in the context of a variable and often unpredictable UC payment could also increase the amount of time involved in monitoring household cash flow. With typically greater responsibility for household budgeting and for managing the UC claim, responsibility for these tasks, together with the stress and worry that often accompanied them, fell disproportionately on the shoulders of women. The fear of not being able to cope, or that the family’s difficult circumstances would come to the attention of social services, were further source of anxiety that were more typically voiced by women. For those who bore the brunt of these additional compliance costs ^{Footnote 17} , the extra effort and emotional heavy lifting meant that UC was not just ‘like work’, it was work, and frequently onerous, stressful work at that (Griffiths, et al., Reference Griffiths, Wood, Bennett and Millar2020: 205).

The all-consuming nature of managing the UC claim sometimes spilled over into couples’ relationships, culminating in disagreements about money and debt. With no let up, the demands of managing the claim while also juggling childcare and paid work led some working mothers in this research to reduce their hours or give up employment altogether, contrary to the policy intent. The trade-off of a more responsive, digitalised, single, monthly payment thus appears to be increased financial insecurity, administrative burden and compliance costs, with greater risk and uncertainty transferred onto claimants.

Legal challenge to the treatment of earned income

Intelligence collected by the UK Child Poverty Action Group (CPAG) lends strong credence to these research findings (CPAG, 2019a; CPAG, 2019b). A legal challenge brought by CPAG on behalf of four lone parents argued that monthly assessment in UC incorporated an unlawful approach to the calculation of earned income (Johnson and Others v SSWP, 2020). Defending, the Secretary of State for Work and Pension’s argued that, although counting two sets of wages in one assessment period may be ‘unfortunate’ and ‘arbitrary’, redesigning the IT system ‘essentially from scratch’ to accommodate an adjustment would cost in excess of £7 million (Johnson and Others v SSWP, 2020: 78). This defence was rejected and the challenge succeeded, but not because the automated assessment system was deemed to have infringed claimants’ legal rights or could create income insecurity, but because the effects, in these instances, were judged to run counter to UC’s policy aim of ‘making work pay’, so were ‘irrational’. The relatively large number of claimants said to be affected also played a part in the ruling (Meers, Reference Meers2020). A series of similar legal challenges in which the fixed monthly assessment period resulted in adverse outcomes for claimants have since failed because the public law irrationality threshold was judged not to have been met. Furthermore, although the DWP has since amended the regulations, and is intending to automate the correction, the new rules only apply to claimants paid a monthly salary, not to those paid four weekly or weekly, even though many are similarly affected and may also have reduced hours of work or withdrawn from the labour market (as our research showed), contrary to the policy logic.

In spite of continuous lobbying, the DWP has repeatedly insisted that monthly assessment and the use of RTI feeds from HMRC are hard-wired into UC’s design and cannot be changed. Will Quince, Parliamentary Under-Secretary of State for Work and Pensions, stated that changing the system of monthly assessment was ‘not operationally deliverable’ (House of Commons Work and Pensions Committee, 2020). Neil Couling, the senior civil servant responsible to Parliament for delivering Universal Credit, similarly claimed that it would need human beings to manually operationalise such a fundamental design change, with serious knock-on effect in terms of the timeliness of benefit payment: ‘If you play about with the architecture of Universal Credit you won’t be able to pay the vast millions we have to pay every month … I’ve explained … many times that the month-long assessment period is integral for how it works’ (Cauldfield, Reference Cauldfield2020). The government’s position is thus that any hardship claimants may experience as a result of monthly assessment, or arising from automated processes in UC more generally, though regrettable, is worth it for the greater good of administrative efficiency in the processing of benefit payments. This trade-off involving the subordination of claimants’ rights and greater compliance to the technical constraints of automation thus continues to hold sway within the Government and among DWP policymakers.

Universal Credit’s ‘moment in the sun?’ ^{Footnote 18}

In the wake of the huge surge in claims due to the Covid-19 pandemic, some have claimed that UC’s digitalised features may be helping to restore its tarnished reputation (Timmins, Reference Timmins2020b). Following the UK lockdown in March 2020, an unprecedented 3.7 million people applied for UC; around seven times the usual volume (DWP, 2020a). A surprise to even the harshest critics of UC, more than nine out of ten eligible claims, together with around 1 million advance loans, were paid in full and on time (DWP, 2020a). By December 2020, 5.9 million people were in receipt of UC (DWP, 2021), compared with 1.9 million in March 2020 (DWP, 2019). That the system was able ‘to work at great volume … through [an] unprecedented claims spike’ while achieving operational performance levels apparently in excess of pre-Covid rates was attributed by Neil Couling to the benefit’s automated features (Public Accounts Committee, 2020: Q12). He claimed that in the face of such huge and unforeseen demand, Jobcentres would have been inundated and the legacy system would have collapsed within days. ^{Footnote 19} Iain Duncan Smith, former Secretary of State for Work and Pensions and UC’s principal architect, concurred: ‘I shudder to think how many people would have put themselves in harm’s way, in queues snaking away from job centres as they tried to sign on in the middle of a pandemic’ (Duncan Smith, Reference Duncan Smith2021). Commentators and think tanks from across the political spectrum tended to agree – that only a fully digitalised system could have processed such a large volume of applications and payments in so short a period of time (Brewer and Handscomb, Reference Brewer and Handscomb2020; Timmins, Reference Timmins2020b).

Others argue that had the legacy system received the same level of investment in IT as UC, it would have coped equally well (Bennett, Reference Bennett2020). In fact, online application procedures for several legacy benefits pre-date the roll out of UC. In 2012, almost 40 per cent of claims for JSA were made online (DWP, 2012a). Claimants could also view information about their JSA, ESA or IS award through a (now defunct) personal benefit account called ‘My Benefits Online’ (DWP, 2012a). ‘New style’ ESA and JSA, contributory benefits which operate alongside UC, and which also have an online application, proved to be equally resilient in the wake of the recent unprecedented increase in claims (Roberts, Reference Roberts2020).

Furthermore, while the digital application for UC may have avoided long queues forming outside Jobcentres, they snaked out of view, online, instead (Glick, Reference Glick2020). The DWP’s ‘don’t call us, we’ll call you’ policy, rapidly implemented to handle the unprecedented surge in claims, was largely a result of shortcomings in the online system. The ability of the DWP to handle, at the peak of the pandemic, more than 4 million telephone calls in forty-eight hours was, moreover, due in no small part to the Herculean efforts of 10,000 UC staff, supplemented by thousands of civil servants drafted from the furthest corners of the DWP and other government departments, to manually expedite the processing of new claims (House of Commons Work and Pensions Committee, 2020), a process which involved halting virtually all other DWP business (Butler, Reference Butler2020). It has also since transpired that, among the 1 million advance loans approved since the start of the pandemic, many have been fraudulent. This recently came to light when thousands of victims of identity theft who had never claimed UC began to have advance repayments automatically deducted from their wages (Brignall, Reference Brignall2020).