There has been increasing discussion and interest, recently, in the idea of using existing data, particularly from benefit claims, to determine or automatically award other benefits for people who are entitled.

It would be a very attractive way of solving, or reducing, the enormous under-claiming of many benefits. After all, its proponents say, if we already know something about people’s circumstances, why make them give them to us again. Even better if we could use that data to give them money from benefits that they didn’t even know existed. How grateful they would be for such unexpected windfalls.

With much data, and records, having moved from paper into electronic forms, it’s now very feasible, technically, to automate such processes.

Sadly, it isn’t quite that simple.

Grateful recipients

Before getting a little technical, lets have a look at the results of trying this before. Between November 2010 and March 2011, the DWP ran a small study where a sample of 2,000 people were paid their estimated Pension Credit entitlement without them first having made a claim. A similarly sized comparison group were visited by DWP staff to help people understand their entitlement and to assist them with the claims process.

At the time it was estimated that non-take-up of pension credit was between 32% and 38% (probably lower than today). As one of the reports on  the project says[1]

Non-take-up of benefits remains a significant reason why a proportion of pensioners remain in poverty. The DWP wished to assess the effect on take-up rates of paying an estimated entitlement.

The Department for Work and Pensions (DWP) has been investigating whether the data it already holds on individuals – from its own administrative records and those of HMRC – might be effectively and appropriately used in the administration of Pension Credit, both to identify and pay those people who appear eligible. It is not possible to pay Pension Credit automatically with sufficient accuracy, due to the level of information required for each individual at the point of claim

The data used was taken from DWP, local authority benefits systems and HMRC PAYE tax returns. This was matched to so that only people whose state pension was being paid into a bank account were included.

After the 12 weeks of automated payment, the study looked at the number of recipients (the payment group) who went on to make a formal claim for pension credit.  If they didn’t claim then the benefit stopped.

The figures weren’t good.

By the end of August 2011 (approximately 5 months after the end of the study), 8.6 per cent of the Payment group had made a successful claim to Pension Credit. This is compared to 13.1 per cent of the Visits group and 2.9 per cent of the remaining eligible non-recipients. So the Payment group had a take-up rate of around two thirds that of the Visits group, but around three times that of the remaining eligible non-recipients.[2]

 

So the human involvement, more akin to traditional take-up work, was about 50% more likely to generate a claim than actually giving people money and asking them if they wanted to continue.  Both were much more effective than doing nothing.

Paying customers their estimated entitlement without them first making a claim cost around £3,800 for each successful award of Pension Credit. In comparison, the service provided by DWP Visiting to the Visits group cost around £600 for each successful award of Pension Credit.

This would seem to indicate that the more traditional method of encouraging take-up did much better and cost much less.

Why?

Why didn’t people who actually had a benefit paid to them not continue?  We know from much other research that the main reasons why people don’t claim means-tested benefits is because they don’t know about the benefit or they don’t think that they are entitled.  Both these reasons seem to disappear here.  The reasons summarised in the study were:

• poor understanding of the study or letters explaining how to claim;
• feeling they did not need Pension Credit payments;
• feeling they were not entitled to Pension Credit;
• attitudes towards claiming benefits;
• views and expectations of the claims process;
• personal circumstances at the time. [3] 

In more detail, the report summarised the thinking of recipients:

Respondents reported feeling uncertain about their eligibility for Pension Credit when they worked or their partners worked. This view was particularly prominent amongst those who were self-employed or had difficulty predicting their annual income. These respondents would have benefited from receiving further information about how employment impacts on their eligibility for Pension Credit.

Respondents were also deterred from claiming if they were uncertain about income thresholds. They felt it would have been useful for this information to be provided before they made a decision about claiming.

In some cases respondents did not realise that an assessment had been made and thought that all study participants had received the same amount of Pension Credit. These people would have been encouraged to claim if they were told more clearly that a calculation had been made.

Inertia and memory problems were a barrier in some cases, as well as people feeling they did not know how or where to claim. These respondents would have benefited from more prompting and reminders to claim but obviously the benefits of doing this need to be weighed up against the costs.

However, there were some respondents who appeared resistant to claiming through strong underlying attitudes towards claiming benefits. These were underpinned by past experiences, as well as feeling that they did not need the money, and did not want to ask for it. Concerns about feeling they were ineligible were amplified by anxieties about the upheaval of the claims process itself, and a worry that they might have to give the money back if they were wrongly paid, leaving them feeling that it may not be worth applying.[4]

 

Interestingly, there did not seem to be any strong feeling that the automatic process had identified a real entitlement to benefit, which might have addressed the identified established barriers to claiming.

The real entitlement uncertainty is understandable as the estimated Pension Credit entitlement that was paid may have been different from the actual entitlement, as the report pointed out “due to the level of information required for each individual at the point of claim”.

Could things be different, and more accurate now?

Automating Take-up 2023

The argument underpinning proposals to increase automation of benefit claiming is simple – if we already know something about a person’s circumstances, then we don’t need to ask it again and we can use it for another purpose.

That assumes a few things that deserve fuller consideration.

• Is the information accurate now?
• Is the information identical in the two cases?
• Do the two benefits have identical data requirements?

All means tested benefits depend upon two core pieces of data, which in turn may have many common elements. They need to determine the needs of the claimant and they need to determine their resources. When assessing needs, there will be a great amount of commonality: household members, ages, relationships, children, health, et cetera. Looking at resources, again there would seem to be a great deal that should be mutually useful, earnings, other incomes and capital will all be relevant.

At a superficial level then, this should be a worthwhile and relatively simple exercise using existing data. The problems sadly become apparent when the issue is looked at in more detail. Despite various efforts over many years, we do not have a coherent, structured and common data set which can be used for means tested benefits. We have almost 70 years of regulations which have just ‘growed’ and had amendment piled upon amendment.  The result means that, for different schemes, which look superficially as if they used the same data, in detail they often do not.

For example, a quick skim through the Council Tax Reduction regulations for pensioners and the pension credit regulations produces numbers of differences; for example:

Council Tax Reduction Regulations disregarding capital value of home.

“Any premises occupied in whole or in part—

(a) by a person who is a relative of the applicant or his partner as his home where that person has attained the qualifying age for state pension credit or is incapacitated;”

Pension Credit regulations define relevant occupiers differently

4.Any premises occupied in whole or in part—

(a) by a person who is a close relative, grandparent, grandchild, uncle, aunt, nephew or niece of the claimant or of his partner as his home where that person has attained the qualifying age for state pension credit or is incapacitated”

The Council Tax Reduction regulations concerning disregards from earnings state

5.—(1) £20 is disregarded if the applicant or, if he has a partner, his partner—

(a )is in receipt of—….

(ix) main phase employment and support allowance

while, for Pension Credit, that becomes

4.—(1) £20 is disregarded if the claimant or, if he has a partner, his partner— (a) is in receipt of—

…

employment and support allowance

There will not be a rich descriptive attribute in existing data, the best you’ll get will be a value of ESA, without knowing which phase, or the capital value of the home , that may have depended upon the relationship of the occupant.  Using that in an automatic award for the other scheme may be an error.

This is even more difficult when schemes other than benefits get involved.  Some schemes will average incomes over different periods, 4 weeks, or 6 weeks, others will disregard different types of income, so there will be no record of them, and so on.

Another major stumbling block with sharing data across applications and datasets is the fact that most means tested benefits are determined on the basis of a snapshot situation at a particular date. Sometimes this data is, at least theoretically, assessed weekly or at the end of an assessment period, while many systems depend upon an annual reassessment. If all the dates for the final dataset do not correspond then it is not possible to have confidence in the results.

It’s been suggested, looking at data held by Universal Credit, that this could be easily used for assessing local authority benefits such as Council Tax Reduction.  In practice this will have some difficulties. The Universal Credit Digital Service (UCDS) is widely used within local authorities HB and CTR processes to receive notification of Universal Credit awards.  This data is sent daily and, where Universal Credit awards are the data required, then this is extremely useful.

Where, however, underlying data is being considered then it’s important to recognise that the data used within a Universal Credit assessment is applied as it stands on the last day of the assessment period. That may not match the relevant date for local authority benefits assessments.  Additionally, when circumstances change, it is in practice moving the weekly based local authority benefits to the Universal Credit monthly cycle, with potentially serious consequences. This data is sent daily and, by Universal Credit awards are the data required, then this is extremely useful. Where, however, underlying data is being considered then it’s important to recognise that the data used within a Universal Credit assessment is applied as it stands on the last day of the assessment period. That may not match the relevant date for local authority benefits assessments.  Additionally, when circumstances change, it is in practice moving the weekly based local authority benefits to the Universal Credit monthly cycle, with potentially serious consequences.

DWP have shown absolutely no interest in using local authority data as an input for Universal Credit, or other benefits, which may indicate their own feelings about data sharing and accuracy.

I’ll avoid detailing the issue of data accuracy, which is of course a major problem across all systems; what happens where the same data item has two different values in the two systems?

Indicating possible entitlement

This is not to say that data matching is useless, quite the contrary, but it does seriously put into question the possibility of automating actual awards.

In Ferret, we have looked at using data in support of benefits take-up in a number of ways over many years. We separate the use of data into a number of areas

• Awarding – automatically determining entitlement
• Filtering – excluding those who can be shown not to qualify
• Targeting – searching for those who seem most likely to qualify
• Forecasting – potential future entitlement or changes in entitlement.

Awarding

To assess entitlement, exactly the same data, as described earlier,  is needed as would be required for a claim. That seems extremely difficult in practice.  However, partial population of claim data is much more practicable but would require a ‘completion’ stage, or a reconciliation exercise, but this removes much of the attraction of automation.

Filtering

Using what we already know about people, lets us exclude them from our communications or other interventions, if we can be confident that they are not relevant to our work. In the simplest example, if we know that somebody is below pension age then we can remove them from our working set.

Targeting

Identifying potential people for a more personal approach. In a targeted approach, we can define the characteristics of those people that we want to concentrate on. We may want to focus, for a variety of reasons, on people living in a particular area, in a particular type of housing, with particular types of income or savings, or any other characteristics.

Take-up campaigns have been local in scale, in the main, as they have been the result of local initiatives. This has meant that it has been common to target particular local groups or needs. Targeting also has many attractions for research or pilot projects, where alternative approaches can be tested.

Forecasting

Benefit assessment tools, and advisers, have always been snapshot based. They looked at the current situation and assessed entitlement under the current rules. It is understandable, as the focus of work in the advice sector has always been helping people who face problems now. The resources have never been there to plan and advise for the future, in a way which might mitigate problems in the first place. For Pension Credit, for example, there is a strong argument in favour of using existing data to look ahead.

We have developed a proof of concept advice tool which uses data in different categories to build information which can be ‘pushed’ to the customer, to help them plan and consider their options. We separate the data into:

• Individual – items such as date of birth, health conditions, housing type and costs etc.
• Planned – current intentions such as retirement date, mortgage completion, children leaving home etc.
• External – factors which are generally applicable to the population, such as changes in state retirement age, preannounced rule changes in benefits etc.

Data richness

For all of these uses of data, precision is improved as the richness and accuracy of data about the individual grows. The likelihood, however, of achieving this is that it will mean increasing human and organisational effort to reach the better representation of the individual. The practical effect of improving the quality of individual data is that the individual result will need more resources and, consequently, there will be a limit to the numbers of people where this can be achieved.

An alternative approach is to think instead about what can be achieved using a more limited set of data, which can be gathered more easily for a larger number of people.

For example, benefit calculators, which we have been producing for over 40 years, can generate accurate assessment of entitlement to benefits, but require answers to a detailed set of data collection questions.

It is possible to collect much less data and provide less comprehensive but still extremely useful results.  One approach, we have used, is to provide interim or partial assessments of entitlement.  With these, we only use the data which is required to determine the ‘needs’ part of the assessment. This substantially reduces the amount of data required and tends also to be the data which is most likely to be identical and consistent across systems. Once we know the requirements figure for that benefit it is possible to reverse calculate the resources that would be needed to extinguish entitlement. That means that we can accurately produce a figure, below which there is likely to be ‘some’ entitlement to benefit. For take-up purposes, we can then allow people to do a comparison with their own income to assess the likelihood of a successful claim

You can access our demonstrator of an interim PC entitlement guide at:

Ferret’s PC Guide  (https://www.ferret.co.uk/reckoners/internal/rec2022/Pension Credit Entitlement Guide 22-23/Pension Credit Entitlement Guide  22-23.htm)

Blended approach

I do not believe that a pure data driven approach is either feasible or likely to prove successful.  Instead, I suggest a blended approach.

Make use of existing data to identify likely entitlement and then work with local government, advice services and other organisations to approach these targets with the personal support and information that they may need.  We know that many people are reluctant to claim or approach the DWP because of previous refusals or other experiences. Encouragement and support can help overcome such barriers. Others may almost need persuasion that they have an entitlement.

Identifying those most likely to qualify will enable local support to be concentrated on interventions where they are most likely to pay dividends.

Mine the information held by local authorities in their benefits systems, and also in other similar systems such as social care or disabled facilities grants (where we already have such a system in place).  Create partnerships to make use of data held by other organisations, where data protection permits, as this is indicative and will not be used for claims.

Delivering the blended approach

Many local authorities are prone to see their third sector partners purely as a delivery mechanism. Organisations are tasked with projects that may not be the most appropriate for their skills, experience, and knowledge of the community. Partly this is because organisations have often over promised their capabilities when competing for funding.

Where local authorities partner more productively with organisations in their area to identify the groups that are most appropriate to work with, then using LA data to identify targets in that group should mean that the existing expert and trusted relationship will be most effective.

Research and experience agree that one crucial factor in increasing take-up is trust. Central and local government are not seen positively, by many people, as sources of help and information. Active approaches from official sources are often viewed with suspicion.  This is particularly true, it is believed, where previous experience of failed applications becomes a powerful disincentive for further claims.  Advised workers are familiar with the conundrum that the least knowledgeable and aware are often the most trusted.

Organisations in the community spend a great deal of time and effort to establish a trusted relationship with their clients. This is not a quick process and often a painful and difficult one but once trust has been established, working with clients becomes more constructive.

This means that a blended approach can be the most effective method of encouraging take-up. Targeting data is sourced from the richer sets available to central and local government, possibly making use also of more local, internally held, data.  Those identified are then approached by people they already know and trust, who are able to encourage and guide them through the claims process.

The ‘official’ approach would have been less successful because of the issues of trust and understanding while the ‘local’ approach would not have been able to identify as accurately potential claimants. Together they become more effective.

To increase the effectiveness further, there is a very strong business case for providing extra resources for the local delivery partners. Many will be currently limited in the work they are able to do because of a shortage of resources and would be able to scale up their operations quickly and effectively. They are also likely to be able to deliver more with less than increasing resources within more official operations.

There are, of course, numbers of practical issues which should not be underestimated. GDPR requirements around the purpose for which data has been collected is one major issue. Another is the question of ‘data ownership’.  By this I mean the responsibility and authority to check and correct data.  Whenever datasets containing common information items are compared, there will be differences. Sometimes this will simply be the result of information in one set having been collected after a change in the circumstances which were captured in the other set. Data quality issues and errors in collection or entry may be another cause of difference. While these differences may not invalidate the use for targeting it does mean that someone is holding, and potentially using, incorrect data on the individual.  This should be corrected but who has the authority or capacity to identify which is the accurate data and to correct it?

Summary

While automatically awarding benefit is currently not feasible, using data to target those who are likely to be entitled to unclaimed benefits is a proven and valuable approach. Much of the difficulty, as research has shown, is then to persuade those people to actually make claims. Here trusted intermediaries will be more effective in encouraging and supporting that process.

Together that can make a difference.

[1] Quantitative Evaluation of the Pension Credit Payment Study, Lucy Radford, DWP Research Report no 796,  2012

[2] ibid

[3] Qualitative evaluation of the Pension Credit payment study, Natalie Maplethorpe, Mehul Kotecha and Sue Arthur DWP Research Report No 795, 2012

[4] ibid