Welcome to the Public Accounts Committee on Thursday 4 December 2025. Fraud and error in benefit expenditure remains one of the Department for Work and Pensions’ most persistent and pressing risks. At the start of this year, the Committee reported that levels of fraud and error in benefit expenditure had remained unacceptably high. We noted that it was DWP’s job to improve its defences to ensure benefit claimants receive the correct amount of money to which they are entitled. Looking at the 2024-25 report on DWP’s accounts, benefit expenditure payments fell slightly from £9.7 billion in 2023-24 to £9.5 billion in 2024-25. However, this was the 37th year in a row that successive Comptroller and Auditor Generals have qualified their audit opinions on the regularity of DWP’s accounts due to the level of fraud and error in benefit expenditure. To generate savings and help detect and correct overpayments, DWP has rolled out a range of counter-fraud interventions, including the use of data analytics and reviews of universal credit claims. Today’s session will be an opportunity to examine our DWP witnesses on the current strategy and funding for tackling benefit overpayments, scrutinise DWP’s prevention and detection of benefit overpayments, and challenge DWP on benefit underpayments and unfulfilled eligibility. To help us with that, we have the permanent secretary, who is a regular attender before the Committee, and two of his key people who work with him. Sir Peter, would you be very kind and introduce yourself, followed by your colleagues?

I am Peter Schofield, the permanent secretary at the Department for Work and Pensions. I am here with my colleagues Neil and Vikki.

Good morning, Committee. I have been here once or twice.

You have. You have also got a season ticket.

I think I have, I’m afraid. I am Neil Couling, director general for services and fraud. Since 2021, I have had the executive team lead on tackling fraud.

My name is Vikki Knight. I am the policy director in DWP for fraud, error and debt.

You have not been in this Committee in this Parliament, but you have been before the Committee before.

I have.

What capacity was that in?

That was under my responsibility for housing and supported housing. I also appeared with Peter for that.

Was that a joint hearing with MHCLG?

It was, yes.

Well done and welcome back. Thank you all for coming. I shall start the questions. Neil, how do you plan to make better use of customer data to prevent fraud and error from occurring in the first place?

Could I possibly come in first, Sir Geoffrey?

Of course, absolutely. Why not?

This is a key part of our whole approach to tackling fraud and error. This has been quite a journey for us over the years. Tackling fraud and error is a massively important challenge and a key priority for us. As we came out of the pandemic, we produced a plan that set out our intention to drive down fraud and error by detecting fraud and error that was already in the system and preventing fraud and error coming in in the first place. I remember coming to this Committee soon after we produced that plan in 2022. I think the Committee welcomed the plan but was very challenging, quite rightly, on whether we could implement it and see it through in terms of our ability to recruit key staff, get legislation through and get co-operation from across Government. We are seeing real progress at the moment—I would say that—as is set out in the NAO Report. I hope the Committee did not mind this, but I wrote to the Committee at the beginning of this week to reflect some of the latest forecasts coming out of the OBR, which showed fraud and error, going forward, coming down below pre-pandemic levels and, in fact, by 2028-29, universal credit being at 7.5%, which is well below pre-pandemic levels. If you then do the arithmetic, that would be a cross-welfare rate of 2.8%, which is lower, on that measure, than we have seen ever since tax credits were introduced in the first place. I want to give a sense that we are driving down fraud and error, we are seeing progress, and we are seeing progress going forward. I wanted to give the Committee that sense that we are doing what we set out to do. There is a lot more to do, clearly, and it is set out in the Report, but we are determined as a Department to drive down fraud and error. The OBR is setting a real sense of confidence that we will see fraud and error not only continue to fall, but fall to levels that we have not really seen before.

Thank you for the letter. We have all seen it. It is good news that you have reduced the percentage in the 2025 forecast. In the Report, the percentage was 8.1%[1]. This presages 7.5%. Is this a new target? Is this the new normal, as it were, that you will be aiming at?

This is what we are going for. It reflects what the OBR thinks our measures will achieve if we do them effectively and in the way that we think we should. It does not set out new things that we might develop going forward, but it is confidence in the things that we have already set out that we are funded for or have legislation for. In the Budget last week we also got additional funding to continue the targeted case review for another year and to roll it out a bit more widely to pension credit in particular. That is reflected in this as well. This is where we want to get to. The NAO would want to audit this number, of course, but I think that this is effectively, at 2.8%, lower than it has ever been since tax credits were introduced. If we can achieve that, that would be a fabulous achievement and would be great for us to be delivering.

We will examine how you are going to achieve that in various aspects during the Committee today. We would like to have a conversation with you and the C&AG as to whether we can get to a level that you could achieve and how you might achieve it, so that we begin to end the succession of your accounts being qualified. If we could get to a position where your accounts were not qualified, that would be a huge win. If you would undertake to have that conversation at some stage, I would be very grateful.

I would be very keen to. We take our accounts incredibly seriously and qualification is a massive deal. The problem is that, as you say, after 37 consecutive years, it maybe starts to lose its effect. Obviously, Parliament, and this Committee in particular, would want to have confidence that we are absolutely on the way and driving this down relentlessly. I hope that that forecast gives some sense of that. If we could get to that point, lifting the qualification would be great, and then enable us to have the qualification hanging over us to keep us driving it out, rather than having it as a given after 37 years now.

What we want to hear today—various Committee members will ask you various aspects—is how you are going to get to this target. Then, bearing in mind what you are hopefully going to do, we might have that conversation to see where we might get to by, say, the end of this Parliament in order to try to lift that qualification.

That would be great.

That is very kind. That is a very helpful beginning. That is probably enough from me for now. Sarah, I am sorry; I have not given you any warning. I have been so busy thinking about things. Would you like to ask your question now?

Yes, of course. Ms Knight, at the time when the Report was being put together, the Department was in the process of putting together a road map of how it was going to deliver the new strategy. That includes how success will be measured. This is particularly looking at preventing overpayments. I wonder whether you could give us more information about how you will know you have been successful in preventing overpayments, on the basis that, if something did not happen, it is quite hard to measure it.

I will bring Vikki in, but can I answer the first level of that, which is about how we measure prevention overall. I brought along the full annual report and accounts for DWP, which I am afraid is 400 pages. Within that, you will see that, for the third year running, we have set out our assessment about the overall impact of our controls to detect, but also, crucially, to prevent. That sets out that we think that, under this measure, we prevented or detected £25 billion of fraud and error in the system over the last year. That is up from the year before. It is something that we will continue to measure. They are experimental statistics, but it gives us a sense of what is so important for us. Vikki can talk more about the strategic controls framework. It is really important for us, and to give the reassurance to this Committee as well, to be able to measure these things. We can measure fraud and error coming in the system through our normal sample. We also want to measure the effectiveness of our controls, not just in detection, by driving through things such as the targeted case review, but what we are doing to stop fraud and error coming into the system in the first place. That is what this set of measures sets out.

Specifically on the point about overpayment, clearly preventing overpayment is one of the most effective things you can do. I was interested in your use of the phrase “experimental statistics”. Specifically, how are you measuring overpayment that you have prevented?

We look through the various controls we have on the system. They can include controls on eligibility—when someone applies for a benefit, checking their eligibility and looking at the impact of those controls that identify people who are not eligible and who are claiming. We score the impact of that. Claims that otherwise would go through but where we have identified that they were not eligible would be part of the prevention costs that we are identifying. We add to them the things that we are already doing in terms of detection—all the various different measures added together. We set it out in quite a lot of detail in the report that we published in July. Vikki, do you want to say more on that?

Yes, just a couple of things on controls. As we said, page 99 in the annual report and accounts sets some of these out. It is the third year that we have used them. There are a couple of things to add to what Peter has said. We are working with the NAO really closely on the work on controls, because we want to make sure that we are demonstrating that effectiveness as part of lifting the qualification on the accounts. We are doing some work around that. We use many different experimental ways to do that. We will use our national statistics that we release every year to give a sense of where those overpayments have come into the system. We will look at the effectiveness overall of our controls. Examples of areas where we look at the controls include the functional health assessments or where we have our counter-fraud and compliance directorate putting in interventions. We will evaluate and start showing how much we have prevented. If we did not have those controls, what would the impact on it have been? We are doing quite a lot of work beyond the controls as well to learn from where overpayments have come in in real time. I am sure you will want to explore later targeted case review. When we look at that, we learn an awful lot from what is coming into the system. Where can we tighten those controls? For example, one area that we have looked at quite closely recently is self-employment. Can we make improvements on that? It is not one single area that we rely on. Our analytical professionals spend a lot of time looking at this, and have worked with the NAO and continue to work with the NAO on this. As I say, the annual report and accounts is an attempt to show and demonstrate that. That is something we have done for three consecutive years now.

Sir Peter, the Department will be receiving £6.7 billion of dedicated funding to tackle fraud and error. Can you give us more of an outline of how you plan to use that, particularly in the three years going forward from 2026-27?

They cover a series of measures. There are funding elements of the plan that we initially set out in 2022, but they are a continuation of some of those. They include a continuation of the targeted case review, which is addressing those claims that have come into universal credit that we can identify as fraudulent. There, we are imagining continuing to spend between £300 million and £400 million a year on the targeted case review, but that is driving around £3 billion to £4 billion of savings every year coming out of the system. That is one element. We are also extending the targeted case review to other areas, particularly pension credit, where we are looking to spend between £60 million and £70 million a year, bringing out between £330 million and £190 million of annual savings from there[2]. There are a series of other measures as well, including the implementation of the powers that came through the fraud and error recovery Bill, which I am glad to say got Royal Assent on Tuesday, so we are rolling that out as well. The Report sets out our intention. The intention set by the then chief operating officer of the civil service is that fraud and error measures should deliver a 3:1 return. We are looking to always beat that, and our measures largely do that. The targeted case review in the past delivered, I think, a 1:7 return initially[3]. There is a whole series of those measures. They are set out in some detail in the NAO Report. What we will do, and what we do every year in our annual report and accounts, is to set out what we are spending and what we are delivering in terms of AME returns and all of those.

This is the plan we set out for the Committee in 2022 and this is what is driving the numbers down now. The OBR forecast suggests that it will go lower. We have been augmenting the plan as we have been able to prove cases, as we proved the case on targeted case review. It got expanded. Now it is being extended into pension credit. The measures coming through the new Act will bear further weight down and drive fraud down even further.

When I was here in the Committee, I think in October, we talked quite a lot about innovation, didn’t we, and test and learn? That is what you are seeing in elements of the fighting fraud and error plan. Targeted case review is a great example of that, where we tried it at a small scale. We worked out how to make it work and deliver a return. We rolled it out further in terms of making it a full case review, rather than just a limited focus on certain elements, and then building it to scale. As we built it to scale, we have shown the return I was describing in my answer to Ms Olney. That enables us to get further investment from the Treasury, as we saw in the Budget last week, so it is a virtuous circle. Obviously, there are things that we try that maybe do not work quite so well, but, as we were talking at the Committee the other day together, those are the things where you try, stop, move on and emphasise the ones that work. That test and learn, innovation and then scaling up the things that work is a feature of how we are implementing this plan.

You sent us this letter at the beginning of the week with the revised forecast from the OBR. Is the OBR forecast based on a detailed analysis of your plans and its estimate of the impact they are likely to have, or is it just a reflection of the fact that greater investment is going into tackling fraud and error?

It is very much the former. We have been through looking at the likely impact of all the things that we are currently doing and the things that we intend to do. We have had to demonstrate it in such a way that the OBR feels reassured and reflects that in its forecast.

It is a central forecast, so it could be higher or lower. The last two years have shown that the central forecasts have been too pessimistic and fraud has actually been lower than the OBR forecasting. There has been a long approach here. I am the person who implemented universal credit and got a lot of stick for how long it was taking to implement, but I knew, at each stage I was going at, that the things we were trying to do would work. I think that is being replicated here in our fraud plan. The NAO reflects that in its Report quite faithfully.

We will move to Clive now on targeted case review.

No, it is the—

It is continuous improvement. Targeted case review will be later on.

Thanks for the warning.

You have been pre-warned. You will give us a better reply.

Some Committee members went to the Department and looked at the continuous improvement activity with regard to universal credit. That seems to be working, or at least moving it forward. One of the challenges we had was, “Which other benefits are you rolling it out to?” I think that the answer was that that is not easy because the digital systems for other benefits are not in the same state and are not as modern as those for universal credit. How are you going to address that challenge going forward?

It is a really good question. The reason we have focused on universal credit is, in part, as you say, Mr Betts, that it is a modern platform that was introduced in recent years and the data is easily accessible. Also, it is by far the biggest area of overpayment, so it was the right place to start and somewhere we have made good progress. You have already seen some measures around—it is not quite under the heading of the continuous improvement programme—the targeted case review, which is working so well in universal credit that we now have the funds to roll that out to pension credit. That is an area where we have seen an increase in fraud and error over the last year or two. Likewise, we should be thinking about rolling out other elements of the continuous improvement programme elsewhere, such as the periodic redeclaration measure. Remember, one big area of fraud and error in the system is that someone might make a claim and their claim might be accurate at the point they make their claim, but something has changed over the time they have been on the benefit in terms of their circumstance and they might forget to tell us or let us know, or intend not to tell us and let us know. Periodic redeclaration is that prompt that says, “You know what, you really need to let us know.” We are rolling that out for universal credit. That covers some of the things that are more difficult to use data for, such as living together or household composition, for example. Those could have a similar benefit to roll out elsewhere. As I say, we are in the first fruits of that, but the focus has been on universal credit because of its scale.

Have you an aspiration or a plan to roll this approach out to other benefits? Does that mean you have to have a plan to improve the digital technology and the platforms on which other benefits are operating?

There are two elements to that. There are specific plans that are costed, in the plan and reflected in the forecast. Yes, we are rolling out targeted case review to pension credit. More broadly, everything else is an aspiration and part of test and learn. We will try new things elsewhere. Partly, it depends, in some cases, on a service modernisation programme to re-platform things to have them on a footing that enables some of these measures to work. This would very much be the plan. Vikki, do you want to say any more about how we are doing this?

I would mention pension credit, but we have done something similar on targeted case review in housing benefit and specifically given some funding to the local authorities to look at the housing element. There are also bits that go beyond technology. We set out again in the annual report and accounts, on page 110, some of those other activities that we are doing around communications. Where we can use data, such as VEP alerts, we are doing so. Also, as Peter said, universal credit has a number of controls. It is our largest benefit. We will want to test stuff in there. Equally, let us look at what we learn from that to bring out to other areas. Pension credit is now our highest percentage of incorrect payments, so we are absolutely focusing on pension credit now and the things that we can do there on strengthening those controls. Those areas will also be supported by the eligibility verification measure that is now in an Act, on the data that we will get from banks. The largest incorrect payments that we have there are about abroad and capital. There are a number of things that we are taking forward.

I have one more point. You are clearly trying to look at other benefits, but have certain challenges because of the IT systems. How far are you able to do that with other Departments? You just mentioned there the household composition. I think that we can all see that that sometimes is a challenging issue. There is information that DFE will have, to a degree, about that. There is information that councils are using on council tax with regard to that. How far are you looking across the piece, rather than just internally in your own Department?

It starts from the relationships that work best, particularly with HMRC. Our track record here in terms of earnings data is really effective. To give you an example, the biggest area of loss in universal credit is earnings, but that earnings loss is all in self-employed or self-reported earnings. Earnings from employment are virtually zero in terms of overpayments. That is because we have a real‑time feed from HMRC, through the PAYE system, that captures changes in earnings in real time. The payment of universal credit is related to that. That is our gold standard, if you like. That is what you would aspire to elsewhere. Why it works so well with that one is for two reasons. First, it is real-time information, because employers are uploading that information through the PAYE system every month. Secondly, the policy was developed around the data. The definition of income or earnings in universal credit is directly related to the data that comes through the PAYE system to HMRC, so you can do that one almost automatically. Elsewhere, the challenge you have is that you might have data, but it is only ever going to be an indication. Even on benefits that we run where there is an earnings element, such as pension credit or carer’s allowance, the feed from HMRC is not comparable to the actual definition of income or earnings related to the benefits, so you need a human to be involved in that. It is only an indication. For most cases, data is only an indication, rather than a definite sign that something is going wrong. Having said that, there are areas where we are doing more work, for example with the Home Office, on data about people leaving and entering the country. Neil will have some other examples.

When I designed universal credit in 2010-11, I knew that there would be about 2 million people on it eventually who would be in work. Given the need to adjust 2 million people’s benefits, if we could not do it automatically, we could not make a universal credit work, so we built the whole system around the data, and that is why. We do not have that luxury with other benefits. I think that it was the Chair’s first question to me that you took, Peter .

You never answered it.

I never answered the question. Data is absolutely key. Exploiting forms of data to keep claims accurate is the lesson we have drawn from universal credit and something we want to do more with. There are sources, as you say, Mr Betts, from across Government and beyond that, potentially, if we could ingest that data into a form that our systems can use, would further drive down fraud and error. We are very keen to exploit such opportunities.

All the controls have an effect. They are all in place because Parliament has decided that that is the eligibility criteria for an individual benefit. Our job is to administer and make sure those controls are effectively put in place. I go back to what, quite interestingly, we talked about during the passage of the Bill. The controls are in place. We attempt to administer them as effectively as we can. We can see within the £25 billion that, by virtue of having those eligibility criteria, you prevent those incorrect payments from happening. You have the checks and balances in place. Some are harder to control than others. You appreciate that I probably do not want to go into the detail of which ones specifically. We have talked about data. That is why it is so important and why we took the Bill that we did and have the Act in place. That brings in another source of data in areas that we have identified as high risk. Peter talked about the employment and data, and eradicating that. We talked in the annual report and accounts, and we talk generally, about the challenges of capital and abroad fraud or incorrectness. That is why we have looked for a different source of data. The only place that we can get that from is financial institutions, which is why we have taken that through. We absolutely need to make sure that we administer the benefit system and the controls that Parliament has given us. We ask for, and have asked for through the Bill, the means to be able to do that. The NAO, when it gave evidence prior to us going into the Committee stage of the Bill, set out really nicely that this is about ensuring that Parliament also enables us to use data where possible, so that we can use those controls and administer the benefit effectively. I have one last point to make. Going back to the continuous improvements, that is an area where, because we are monitoring these controls, we are able to identify where we possibly have further weaknesses in them and bring in some of those continuous improvements. It is a constant moving picture.

I will bring in Neil in a second. You are right that it is the second biggest cause of fraud and error. I think that is about 1.8% of the 9.7% in universal credit. It is a difficult definition. We are trying to make it easier for people to report changes of circumstance. There is a campaign we are launching next month: “Tell DWP”. Within that, it takes you to a website where we have tried to give people some explanation about how you would assess this in your own life. As you say, family life is fluid in many cases and it is a difficult definition, but we have tried to give people advice as to how to declare that. The starting point with all of this—the starting control—is to encourage people to tell us about their circumstances and changes in their circumstances. One thing I take really seriously, for those people who are honest but might make a mistake, is how we help them to declare what they need to declare. The campaign we launch next month, and the website that goes alongside that, will help people to do that. That is one element.

Stepping back, the household test in the income-related benefits has been there almost since their inception. It is worth about £40 billion in terms of public expenditure prevented. For example, my wife is a self-supporting Church of England vicar. She could claim if there was not a household test on us, despite the salary that I am paid as a civil servant. That household test saves £40 billion. The problem with the household test is that it is not easily susceptible to the data-driven solutions that Mr Betts alluded to in his previous question, because there is no data that proves that somebody is living with somebody else in a couple. It is much more of a subjective thing, which is why you will see that it is an area of friction between citizens and the Department. This subjectivity means that we might decide something that that claimant thinks is completely wrong, in that they are not in a household relationship with an individual. It is quite a contested area. It shows up in the fraud and error statistics as a large amount, about £1 billion, because of that. That £1 billion that looks like it is lost to the taxpayer is, effectively, some kind of down payment on the £40 billion that would otherwise be expended because there was no household test inside the benefits system.

The policy is that just getting a job should not trigger a work capability assessment review, because the work capability assessment is used inside of universal credit to determine entitlement to the LCWRA component. It should not automatically trigger. Sometimes it does, which is more of a mistake on our side. It does not drive our overpayment figures. If it is correctly done, the work capability assessment should not trigger that, unless there is been a change in the physical or mental impairment of the claimant concerned.

Neil, I should know the answer to this question. Given the very big figure you have talked about with this household benefit test, we have had some evidence, as Anna says, that just because somebody starts to live together with somebody else, it does not necessarily mean that they are going to put any money into that household. Does part of the household benefit test include whether the new partner, whether they be male or female, is actually putting money into that arrangement? Otherwise you get a situation where no new money is put into a situation where they were previously entitled to benefit, but may not now be entitled to it.

This takes me back about 40 years when I used to visit cases like this and try to do it, when I was a lad. You try to assess whether there is a common household. Are bills being shared? Is there some contribution going on? It is not as black and white as, “There is no money coming in.” The reason for that is that because Parliament assumes that a couple can live more cheaply, it pays a lower rate of benefit than two single-person rates of benefit. That is the logic behind the test. Back in those days, we used to look for things such as, “Do you share a salt and pepper pot?” I think that the test has moved on since then, but it is this concept of a common household. Is a common household being kept? If a common household is being kept, you are living together as a couple under the legislation. If it is not—if it is like a flat share—you would not be determined as being a couple.

It is an example of the complexity in the benefits system. On earnings, things are black and white. On other aspects, there is a degree of discretion. Our hard-working frontline folk are taking those decisions and doing a brilliant job, but we do not make it easy for them.

I see that the spring statement forecast for 2025 says that overpayments would fall to the pre-pandemic level of 3.1% by 2028-29, which you have just referenced. That does not sound to me to be very ambitious. In fact, it seems to be rather unambitious. I am wondering how confident you are that, in due course, you will be able to reduce the payments even to that unambitious 3.1%.

That is where my letter that I wrote to the Committee on Monday comes in. Don’t worry: a more recent forecast was produced by the OBR alongside the Budget last week. That shows it going further down to 2.8% equivalent by 2028-29, which, as I say, would be the lowest—

Yes, you did say that.

I appreciate that the NAO has not audited this, but our calculation is that this would be the lowest level since tax credits were first introduced. We think that is impressive. It goes alongside the challenges in a society where sometimes it feels like fraud and error is becoming more acceptable. We operate against that. Against those headwinds, we are driving fraud and error down, we think, to potentially 2.8% in 2028-29. As I said in my answer to the Chair earlier, that is absolutely something I would be gunning for and pushing for. That would be an objective for me, absolutely.

I hope you do.

We have actually discussed this.

We did, right at the beginning.

Having read all this material, it seems to me that official DWP calculation errors are causing nearly as much loss as claimant errors. Why does the Department persist in framing the problem as primarily the claimant, rather than acknowledging serious failings within the DWP? That seemed to me to be a question. After 37 years of successive qualified audits, there seems to be a systemic problem. I wondered whether that might be it.

You are right to point out that official error is a component part of overpayments. The biggest factor by far is fraud in the system. That is what the numbers set out. Three-quarters of it is fraud. That is why we are focusing on fraud. In particular, fraud had been going up prior to producing the plan that we described at the beginning of the hearing, so that felt like a really important area to drive down on. We have gradually been reducing official error as well. The challenge of official error relates to the conversation we were just having with the Chair about the complexity of people’s lives, alongside the “Yes or no” factors in the benefits system. You are right. That is a continuous focus for us in terms of frontline training. We have just rolled out a new set of frontline training in fraud and error for everyone to complete across the Department. We have regular audits in terms of line management regularly reviewing progress. We have second and third lines of defence. We have to get our house in order, Mr Lowe. I could not agree more, but this has to be seen alongside where the big overpayments have been coming in the recent years and where Parliament, I think, has been focusing in on us: “Get your house in order, but also make progress on the three-quarters of the overpayments that are fraud and error.”

Listening to your answers, which have been very enlightening, I think I am right in saying that the DWP is ramping up surveillance of claimants and issuing sanctions at record levels now. How does that explain the Department simultaneously losing £7.4 billion to fraud in 2023-24? Does that not play to serious issues within the DWP itself? If you are ramping it up and we are still losing £7.4 billion, does that not tell you that there are systemic failings that need to be addressed? I do not know. If I was running a business that had that issue, I would probably say that that was a problem.

That is why we have been bearing down on fraud. Fraud is committed by individuals stealing from the state and we give that no quarter. It is a large-volume activity. There are a lot of people engaged in this. If I extrapolated from the last fraud figures, it would be a thousand people in each of your constituencies who would be committing some kind of benefit fraud, if it was evenly spread across the country and you had equal-sized constituencies. I know that you do not. This is a large-volume challenge. It is not the Department failing to do things. It is about the Department trying to find new ways to tackle these challenges. The recent Act is one of those, by getting information from the banks for people who are not telling us that they have savings. That will help. It will allow us to see who has gone abroad and is encashing benefit money abroad. That is one of the causes of fraud. We are tackling this issue by issue and trying to come up with responses to it. As Peter’s letter to the Committee shows, we think that it is bearing fruit. It is still too high. We are still striving to drive it down further, as we said in our answers to Ms Gilmour. This is a big problem in society. Fraud is going up elsewhere. We have it coming down in the benefits system. We want it to come down more. This is a real new 21st-century challenge. Society’s attitudes to fraud, sadly, from my perspective, have got a lot more sympathetic to fraudsters. I would love it if, as a country, we really rounded on this problem.

As a matter of interest—you know this better than I do—is there a problem with the advance payments within universal credit? As a matter of interest, the advance payments seem to be the source of a lot of abuse. Would greater sanctions for those abusing the system, such as removal from any welfare payments if they abuse the system, bear fruit?

Advances fraud is a very small component of the overall total. We estimate that it is anything from zero to £60 million. That is the upper and lower end. We know that it really annoys people when it happens and we have taken lots of steps over the last few years to bear down on that. It is a small component of this. We use sanctions. Ultimately, we will take people to court and see them imprisoned for benefit fraud. That happens at the real top end of abuse. There are a series of other sanctions that we can apply to claimants. Most fraud, though, tends to be people not telling us things and thinking they will get away with it. As I said in answer to Mr Betts’s question, that is where data really helps us. We try to put speed bumps in the road that stop people doing that or remind them, “You should let us know about that.”

You say zero to £60 million. It cannot be zero.

This is the way that that is forecast. It is the range of it. It could be as low as £5 million, say, but it could be as high as £60 million.

Back in 2020 we had a problem and the National Audit Office wrote a Report, which I could refer you to, Mr Lowe. Since then, we have been clamping right down on things such as ID verification and other types of verification through the system, and identifying those cases that look odd as they come through, so that you can stop them before the payment goes out. This is an area also where we have applied, in albeit quite a limited way, machine learning as part of this. These controls are proven to work. It has driven it down to that sort of level, compared to wider issues of fraud and error, which are in the billions rather than the tens of millions.

Thanks for raising that, Ms Dixon. It is a serious and important issue. Liz Sayce’s report, which we welcomed, set out things that we need to do. The Government have accepted the vast majority of those recommendations. We have not been hanging around for this report to come through to take action. We have, over the years, sought to improve the way we serve carer’s allowance customers, for example making better use of information we get from HMRC on earnings. I talked about that as far back as 2019 in front of the Work and Pensions Select Committee. Over the last year, the Government have also taken action on things such as increasing the level of the earnings limit up to about £196 a week. If you recall, carer’s allowance is a benefit that was introduced in the 1970s and has not really been reformed since then. You have this hard limit on what you are entitled to earn before you are not entitled to anything at all in carer’s allowance, but that has been raised. We have also agreed to look at whether there is a way of introducing some sort of taper on that. There is a whole series of things we have sought to improve, as well as the communications issues that Liz Sayce raised. Let me address the very specific and serious issue that you raise about the mistake that we made as a Department. There was guidance related to the way that we deal with fluctuating earnings—customers whose earnings fluctuate from one week to the next. It emerges from Liz Sayce’s report that the operational guidance that we had and were applying, we think, going back to 2015 was incorrect. The law actually gives us discretion. If someone has fluctuating earnings, it allows us to average the fluctuating earnings if those earnings are on a regular pattern that you can discern. It also allows the decision maker to average, if they think that the averaging would be a more effective way of assessing the accuracy of the earnings number, where there is an irregular pattern, and to look for that irregular pattern. The guidance that we put out in 2015 to our operational colleagues did not provide for that discretion in the case of fluctuating earnings where there is an irregular pattern. That is where we made the mistake. I want to reassure carer’s allowance customers. The vast majority of people would not be affected by this. If you think that, at any one time, you have about a million people receiving carer’s allowance, we think that over the last 10 years this would have affected about 26,000 people, but 26,000 people is 26,000 people who I need to sort out. I am absolutely determined. I am sorry for all of those who are affected by this, but I am going to sort it out. To answer your question, we are mobilising the teams to address this. Because we do not know who those 26,000 people are, we have to go through a whole series of cases—probably up to 200,000 cases—to find them. We think that it may take about two years to do, but I am determined to do it in the way that we have addressed underpayments that we found in other areas of the system, such as in the state pension. We have a good track record of mobilising and putting it right. Putting that right for those customers where we have got it wrong is my priority.

We have made progress, but not enough. Liz Sayce has set out a number of actions that we need to take going forward to make a difference. We are doing those, absolutely. We have set a service owner to lead across carer’s allowance, own this problem and challenge. We are doing more in the way of better communications. A really important element here is around the alerts that we get from the HMRC on earnings, where previously we have had the funding to roll out to about 50% of the cases. Over the last year, we have been funded to do all the cases, so we have eliminated the backlog. That means that now, if we get an alert from HMRC that someone on carer’s allowance is earning more than the earnings limit, we can deal with that in real time very quickly and avoid the problem of that person generating an overpayment and a debt to DWP. Alongside that, there is action to look at whether we could, in time, introduce a taper into the system so you do not have that cliff edge, although that is a complex policy and complex delivery. Alongside that, we are taking that action. This is one of the examples where data is a help, but it is not a panacea. The definition of earnings for carer’s allowance is different from the definition of earnings for the PAYE system.

There are really important expenses and deductions that no one would want us to lose for respite care, pension contributions and other such things. It is complex and difficult to administer. You are right that we have got it wrong in some of these aspects. We got the operational guidance wrong. We are going to put it right. We have a whole series of actions coming out of Liz Sayce’s report that we are going to put into practice. I hope that that will make a difference going forward.

A key part of it is helping them know what they can do, in terms of letting us know about a change of earnings, and using what we can do, in terms of the data we get in, to help to address the issue. The majority of carers do not work and the majority of carers, now that we have increased the earnings limit to £196 per week, will be able to do more work without having an issue. I hope that that helps as well, in terms of moving that cliff edge further away from where many people are. Carers do a valuable job in society, in terms of caring for loved ones or friends. We want to be able to encourage that, but also enable them to feel they can work, so they can claim the benefits they are entitled to but also contribute to society, not only as carers, which is so valuable and important, but also, where they can do some work alongside that, to do that too.

We do. We did sampling on the caseload to understand the scale, which gave us an indication that it was around about 26,000 people. To find the 26,000 people, we need to build a team that can do that and work through all the cases that were earnings-related overpayments, which is around about 200,000 cases. That £75 million enables us to build the team, run the team and fund the payments to customers.

Yes.

I am going to call a break now. Sitting suspended. On resuming—

Welcome back. I want to ask you about the advances model. I do not know which of the three of you wants to answer this question. Without an estimate of cost, what assurance did you have that the advances model is value for money?

Thank you for raising that. I will start and Neil might want to come in afterwards. The advances model is one aspect of the controls that we have over advances fraud, which Mr Lowe was asking about earlier. You have to get it into perspective. In the report, we talk about how £4 million to £5 million has been identified and saved through this machine learning model over the period of the three years that we have been running it. That compares with the £25 billion of overall savings from detecting and preventing fraud through everything else that we do. It is a small part of what we do. It is part of what I would put in the category of test and learn. It goes alongside more conventional tools to clamp down on advances fraud, which I mentioned in my answer to Mr Lowe. If you go back to the NAO Report on advances in 2020, it was a serious issue that was raised, which we needed to grapple with. Better controls over identification verification and better conventional controls about matching datasets that suggest that something might look suspicious or be a bit odd and need further investigation have made a massive difference. Fraud through advances is now down into the tens of millions rather than where it was prior to that. Machine learning is a part of it, but it is a small part of it. It is test and learn. It is something that we will keep evaluating over the period ahead.

We have received a little bit of evidence about these machine learning systems that you are putting in place. We have touched on this before, but could you talk about the ethical side of machine learning? First, when you bring these new systems online, how are you making sure that they are fair? Secondly, how close are you to getting them all online?

I will bring in Neil in a second, but I would just say this. We test them to check that they are effective and that they can make a difference. The machine learning tool that we have for advances fraud identifies three times more cases than a random sample would. This is test and learn. We are looking at it gradually, bit by bit. We will not roll them out to scale until we know that they work. We are committed to being absolutely transparent and open. You and a number of other colleagues from the Committee were able to come across in the summer to go through this in detail. The effectiveness report that we produced last year was written in such a way that we did not want to publish it, because we might give too much away, but we were very happy to share with you some of the things that we were not prepared to publish. We wrote the 2025 effectiveness review in a way that meant it could be published. We published that in July, I think. We are keen to be open. There is a lot set out in the NAO Report as well. Neil, do you want to say a bit about where you think we are in terms of the development of this more broadly?

We are at an exploration stage. This is exploratory. It is not clear whether machine learning itself can be used to tackle fraud, because you have to pick up patterning, and the problem with fraudsters is that they do not tend to pattern in an open way. This is why we have kept them as stuff that we are exploring. I have said to the team, “We will explore these. We will probably deploy the models that we have developed.” There is one for living together, for example. There is one for self-employed earnings. We will probably deploy those and see whether they are worth taking to a bigger scale. I would add one other thing on what we used to call the fairness assessment, which I have now renamed an effectiveness assessment. You cannot do a public sector equality duty kind of investigation into these machine learning models, particularly in the fraud area. Why is that? Because the fraudsters hide among the people with protected characteristics. I have a really good example. If you want to get more money as a fraudster, if you claim that you are disabled, you will get more money. That is the way in which the system is constructed. The fraudsters will hide among legitimate claimants who have disabilities. Your modelling will pick up that you have “targeted” a group of people with a protected characteristic to find the fraudsters. You have to find out whether it is better at finding fraudsters than snaring legitimate claimants in the net. It is an effectiveness analysis rather than public sector equality duty kind of work. We do public sector equality duty analyses too but, in this particular regard, this fairness analysis has been misunderstood by many commentators. We are at an early stage with this. We are approaching it very carefully because some other countries have got into real hot water. We always make sure that there is a human involved in the decision-making. We always make sure that we send false positives into the system so that the assessors do not know, “There is a fraud case coming. I had better be on the lookout for that.” We take a number of steps to make sure that we do not fall into the traps that some of our friends and partners across the world have. Notably, the Australians, the Dutch and the Spanish have got themselves into a bit of a mess around some of this work.

That is a really helpful answer. I see that there are some edges there that we might want to examine further, but in the interest of time today we might correspond with you about that. Rupert had further questions on this.

I just had one question in response to the answer on advance payments. One of my colleagues was going to ask a question, but they are not here.

They will be, but I just thought this was an appropriate time.

In 2023-24, I think I am right in saying that £934 million in UC advances was recovered. You have not published how much was actually lost in UC advances. Will you commit to publishing that? It might be quite useful.

The recoveries that we set out there are where we have made a legitimate advance, but we then recover it back through the benefit system over the months after that. If someone takes an advance, they have an advance that they then need to pay back and we need to collect. That is what you will see in the debt statistics that you are probably quoting from there, Mr Lowe.

The range that I gave of nought to £60 million is the answer to that question. The amount that we could not recover from people who fraudulently claimed an advance is somewhere between zero and £60 million.

People who work within the DWP are telling me that it is a big problem. You are telling us that it is not a big problem. The people who work on the frontline think it is a huge problem.

I used to work on the frontline as well. It does grate with people that somebody can dip into the system in this way and disappear without us getting them. They do not see the work on the frontline that goes into making sure that these cases are all verified. They do not see the cases that are turned down. They just see the applications. They will see the applications and think, “Crikey, what’s going on here?” We stop a lot of this fraud now through the controls that we have put in both prior to the pandemic and particularly after the pandemic. It surged again during the pandemic.

Do you publish—it would provide an element of a reassurance—the number of cases that you have prevented?

Prevention is set out in the £25 billion. That is the main one.

Yes, but are there actual numbers attached to that?

We can look at doing that. That would be an interesting add-on. We also declare write-offs. Where there is a debt that we have been unable to recover, we will declare that as part of our overall write-off.

It is like cyberattacks. You only hear about the ones that are successful. The ones that are prevented are equally important.

Yes. It is within the £25 billion. I will very happily add to the way that we set that out.

It is often a modus operandi attached to identity hijacking. That is where a lot of it goes. We publish information on the number of people’s identities that have been hacked, which we have had to repair and sort out. Let us take that away.

That is very helpful. Thank you for that answer.

To go back to the Chair’s previous questions about the model and fairness, Mr Couling, you referred to disability as being something that might be overrepresented. The Report highlights that it is older claimants and non-UK nationals who are being over-referred in terms of what you might expect. First, is that a concern for you? Secondly, what can be done to collect better data? It is disproportionately onerous for those people to have to provide the additional evidence.

I will explain why the over-referrals are occurring. Typically, you cannot get universal credit if you are a pensioner, but you can if your partner is under pension age. In a couple, that would be a legitimate claim. Because the system knows that you should not normally get universal credit as a—

Can I stop you there? It specifically says in the Report “age groups 45 to 54 and above”, which is not pension age.

Yes. Pensioners do not just have relationships with people very close to their age. It can go quite down the age range for both men and women now, I am told. It is a modern world. I need to get used to it. It flags the case for looking at. In the management speak of this, we try to retrain the models. We try to get it not to find so many of those cases, because it is perfectly legitimate for somebody to claim universal credit where one of the couple is a pensioner and one is under pension age, assuming that they meet the other conditions of entitlement. That is why that is happening in the model. Similarly, people from abroad are not allowed to claim universal credit. We get people from other countries, as Mr Lowe was suggesting, attempting to get advances. Again, the model helps to prevent that. There is an over-referral because some people who come from abroad but are resident in the United Kingdom can claim universal credit. The model picks up those sorts of things. We have to retrain it so it does not get so many and we get back into the right sort of proportions.

That work is under way.

That work is ongoing, yes. Each time we look at an effectiveness analysis, we look at what it turns up and then we think, “Is there some change that we need to make here to make the thing more effective?” False referrals are no use to us either. It burns money to look at the cases.

The crucial thing to say is that it gets referred to a caseworker alongside a set of random cases that have also been referred to the caseworker. The caseworker then looks at the actual evidence and makes a decision about whether to allow the advance.

We do that in 24 hours.

Exactly, yes. The median is a day.

Is there not a requirement for the claimant to provide further evidence?

It depends. It depends on the situation. There may be, but by and large it can all be done within a day. That is the median amount. As I say, that includes people who have been randomly referred by the model in order to check across whether there are other cases that are coming through. By and large, that has been very effective in driving out fraud and error in advances.

As you are training your machine learning and getting better at this, will you be publishing the outputs to show the fairness analysis?

We have said that we are going to call it an effectiveness analysis going forward, so people are not misled by the title. We have committed to publishing that going forward. We published it this year. Some of you in the Committee came along to a private briefing on the previous year’s data that we signed up to as a result of one of your recommendations in a previous report. We agree with you that transparency here is really important. We cannot talk about what the model is doing and how it finds cases but, in terms of its outputs, we are very happy to share. We publish the information in a way that does not compromise our ability to tackle fraud.

If the Committee wants another private briefing on that, we are always happy to do that.

That is a very kind offer. Let us just see where we get to today, see what we get to in our report and get the recommendations out. We may well take you up on that, Sir Peter. Thank you for that.

We have mentioned already targeted case review and the fact that it was in the Budget. The time period for it has been extended and pension credits are now going to be included within the framework. On the figures we have, the proportion of cases that was found to be incorrect was less than was forecast. Why is that? You believe this approach is working, but it does not seem to be working as well as you thought it might do.

We are 11% ahead on the savings. The hit rate is lower than we thought at this point, but we are finding a greater amount of money. We are trying to target the higher areas of loss. In order to get this going, as we mentioned, we started it small. We were testing and learning. We were not testing and learning with the full model. We are testing with a partial model just to get to get things going. We are now running the full model. That is turning up a hit rate of about 28%. Why was it down around 20%? That is because the 24% figure is all errors in the system. That might be an error of a penny or a pound a year, so a small error. We were targeting the higher errors, which is why we were 11% ahead on the money that we thought we would save at this point. It is working very well, but it is quite a complex story to get across to the Committee in terms of the way in which we approach this task. This is a completely new thing. It came out of our experience in the pandemic. I ran the pandemic response for the Department, and I asked Peter whether we could suspend some controls. Peter said I could suspend the controls when the numbers of claims were going up and we had to keep people at home during the pandemic, but he said, “You must do a review of the cases afterwards.” When we did the review afterwards, it was puzzling me that we were finding so little fraud. I was talking to the analysts about it and I said, “Why is this?” They said, “Our sampling is much more detailed.” I then said to the team, “What about if we did a targeted case review using the same method as the samplers use to find fraud and error?” That was the genesis of targeted case review. The sample is normally about 20,000 cases. It is a massive task to run millions of targeted reviews. It has been a step-up process. We have now reached full volume. The people doing it are now experienced. They are more productive. As I say, the hit rate is now exceeding our expectations, as we have got good at it.

Just to say, I am glad that the NAO has produced this figure on page 36, which sets out the key metrics. These are the metrics that I look at on a regular basis with Neil and his team to get a sense of whether we are driving this out. We have got to the point where the Treasury is prepared to back it with further investment. That is the crucial thing. This is a massive endeavour. We have 6,000 agents, including the outsourced agents, working on this. Last year, 900,000 cases were looked at. This is operating at huge scale, but it is contributing a massive amount to driving out and detecting fraud and error in the system.

Sir Peter, you have almost anticipated where I was coming to with my next question. On the point about outsourcing, are you referring to Teleperformance?

Yes.

Paragraph 17 tells us that it has 2,600 full-time equivalent agents undertaking reviews and detecting incorrect payments. That is a very different number from the 6,000 you were just referring to. Is part of it being done by your own people and part of it by this outsourced firm?

Yes, very much so. Neil was talking about that. We started by developing the model in-house and working out how to do it. We would have been unlikely to get to a point where we were able to recruit a full 6,000 DWP agents to work on this. We have talked before about recruiting work coaches, for example, and we had other recruitment requirements across the organisation at the time. It is a mixture. It is around about half and half. It is slightly more in-house to get you to the 6,000. As you say, the Report gives 2,600 as the outsourced number. The outsourced contractors cannot do the whole job because some of the decisions have to be made by DWP agents. There is a hand-back as part of the process. It is not quite as efficient, but it felt like that was the right thing to do to get to the scale that we needed in terms of delivering the savings over the period ahead.

I have one other question on targeted case reviews. This comes from the confidential evidence of a person who has been subjected to three targeted case reviews. They are worried about feedback. They say there is a sole survey in existence for benefit customers, but the data in the survey is based on 9,075 interviews, compared with the approximately 23.7 million people who are in receipt of welfare benefits. Are you satisfied that your feedback system is good enough so that people who are targeted with these reviews, whether they are found innocent or guilty, at least have a chance to feed back to you so you can continuously improve the system?

Feedback through the targeted case review is incredibly important. In fact, we found that very helpful in terms of learning what things we need to be addressing. Some customers come through this—it is not many now—and we discover that they have been underpaid rather than overpaid. We can learn lessons from that in terms of how to address that. In terms of the sample, do you mean the sample that we do to collect our official statistics on fraud and error?

I think it is the customer experience survey.

Yes, it is exactly that, Neil.

That is a sample, but it is not the only feedback route that we use. The other benefit of targeted case review, as Peter was saying, is not just that we find fraud and error or underpayments; it is that we learn what is going on out there and how people are presenting to us.

To answer my question, you can learn how to improve the system.

Yes. If people are not happy—I have seen correspondence and input from people who are not happy—we will look at our processes and see whether there is a way of getting the same result without the same impact on individuals.

I would like to focus my questioning, if I may, on the safeguards that you are putting in place to ensure that particularly the new powers that you have around enforcement are being used proportionately and effectively. It would be helpful if you could start by outlining what safeguards there are and whether you feel they will be effective. Do further safeguards need to be put in place?

I am glad that there are safeguards. It gives me reassurance as the accounting officer that there is oversight for the powers that my Department is going to be delivering. These powers and safeguards have been rigorously debated in both Houses of Parliament before the Bill got Royal Assent. Some of the key safeguards—Vikki may want to add to this—are for the eligibility verification. I talked before about a code of practice, which we will consult on. We will appoint an independent person who will oversee the way that we use these access powers, produce an independent report and take a view about whether we are doing that appropriately. You will hold me to account on the basis of that report, I am sure. In terms of the powers around entry, search and seizure, HMRC already has similar powers. In a sense, it is about giving us the same level of ability to take action, to secure evidence and to work in partnership with the police, but to put less of a burden on the police as part of these investigations. These powers will be overseen by His Majesty’s inspectorate of constabulary and fire and rescue services, in the same way as would be the case for the police or for HMRC. The powers were debated at length over the course of the Bill’s passage. It is important that we have checks and balances on this. We think the powers are going to be effective. Being able to have feedback from banks about claimants who have more than £16,000 in their bank accounts and are claiming universal credit will be an incredibly important control for us to address capital fraud, which is one of the three biggest causes of fraud and error in universal credit. Some of the powers will enable us to be more effective with our investigations. Where there are serious fraudsters, we will be able to prosecute them on the back of evidence that we secure to act as a deterrent to all others. There is also the opportunity through the recovery powers. Where we have identified that someone owes money to the taxpayer and we know that they have the money, but they are refusing to pay it—they are not on benefits so we cannot deduct from benefits, and they are not in employment so we cannot deduct from earnings—we have the ability to apply to a bank to get that money in a different way. These are all important powers that reinforce our ability to deliver on the plan that we set out in 2022. I am certainly reassured that Parliament has put safeguards in place. It is right that Parliament does.

Permanent secretary, you have these new enforcement powers alongside the ones that you had before. Could you explain a little more to the Committee the approach of DWP in relation to both criminal prosecutions and civil penalties for benefit fraud? What thinking goes into whether you pursue either criminal prosecution or a civil penalty? How do you reach those decisions as to which you pursue as part of your enforcement?

I might bring in Vikki on this in a second. Prosecution does not contribute much directly in terms of recovering money to the taxpayer, but it is an important thing to bring justice for taxpayers and to set a deterrent to others. It is an important part of our overall spread of options. The most serious we refer to the CPS for prosecution. For those that cross the boundary where there a case for prosecution but they are not the most serious prosecution cases, we would offer an administrative penalty as part of that. For the least serious ones, there might be a civil penalty that is part of that. Vikki, do you want to say a little bit about the balance?

As Peter has set out very clearly, we consider all of these on the basis of seriousness. The most serious cases we will refer to the CPS, and we do not prosecute within DWP. We make those referrals and we make that judgment. That judgment is made by counter-fraud experts who we invest in. We train them and make sure that we have specific authorised officers who make those decisions. We are constantly reviewing and reflecting on the guidance and updating them. For example, on administrative penalties, we have put in additional support. When we are going to offer an administrative penalty, it goes to an independent panel that has not been involved in that process, which is full of counter-fraud experts. They will make sure that we have the right elements there. To reinforce, this is voluntary to the individual. This is an alternative to us referring to the Crown Prosecution Service. The ultimate safeguard on this is that we now have HMICFRS coming in as part of the Act. It will oversee and do an independent report. I have written down that it has been around for 160 years, doing these sort of reviews of organisations. It is an important part of us.

In a nutshell, whether or not you decide to pursue criminal prosecutions with the CPS or a civil penalty, is DWP’s main motivation a cost-benefit calculation in terms of how much this could end up costing the public purse? You were maybe suggesting—correct me if I am wrong—that you are adopting more of a deterrence approach in terms of enforcement against tax fraud.

It is a deterrent. The amounts of money that we secure back are relatively small compared to the other things that we do through our counter-fraud and compliance teams, and through things such as targeted case review. It is important to bring justice and to provide that sense of deterrence.

It is really useful to see the different tones we have from the DWP compared to similar sessions we have with the chief of HMRC, when they look at tax fraud versus benefit fraud. It is helpful for this Committee to understand the similarities and the different ways of working in terms of pursuing enforcement. If I might move on to a different topic, permanent secretary, I would be keen to have a little more understanding of the work that DWP has done, and where you see this going in future, around reducing the levels of underpayments that are caused either directly by DWP or by other Whitehall Departments. We have a figure: it is estimated that it is around £1.2 billion for 2024-25. How do we continue to reduce that figure?

We report separately on underpayments and unfulfilled eligibility. In my answer to your question, I might bring the two together, although that is not appropriate in terms of how we report. They are reported very separately. It matters to me that we drive down underpayments and unfulfilled eligibility just as much as driving down overpayments, because it is important for me that our benefit recipients get the money they are entitled to. There are three elements to that. First, how do you remind people that we need them to tell us about a change of circumstance? That is part of the way that the system works. You will have seen in the statistics that one of the main causes of unfulfilled eligibility in the benefits system is people who are on one of the health benefits whose condition sadly has worsened, but they have not let us know. There is no particular way that we would know without them letting us know. That is the first thing: to remind people that that is what we need them to do to avoid the problem. Secondly, we need to make it easier for people to let us know about a change of circumstance. We have tried lots of different things. One of the things—I talked about this earlier—that is rolling out in the course of the next month is a campaign about telling DWP about a change of circumstance if something has happened in your life. I have seen the mock-up of the advert. It is, “Tell your best friend, but also tell DWP”. I do not know whether that feels like it is going to get traction.

There will be lots of testing.

There will be lots of testing for that. Please tell us about that. Alongside that, for some aspects of universal credit, we will be signposting people to a website where they can upload and get a sense of what sorts of changes of circumstance matter. Living together is one of the things that we have set out. What does that mean? What do you need to do? There is then a link to a universal credit journal portal so they can let us know. We will look to roll that out more broadly to other benefits as we develop that further. The third thing is really important. I have talked to this Committee about it before, so forgive me. It is that, when people come forward to talk to DWP, they trust that we will treat them fairly. I have spent time with customers, particularly customers on disability benefits, who have been worried about coming forward to DWP because they are already on the lower level or the standard level of PIP. Their condition has worsened. Their doctor has then said, “Your condition has worsened. You might be entitled to the higher level of PIP,” but they have been worried about letting us know because they rely on the PIP. They are worried about an assessment for some reason. They might feel a bit uncertain about how the process works and be worried that they might lose the money that they have as opposed to getting the additional money that they are entitled to. We have done a lot of work on that, particularly through the reforms we are making in the health transformation programme, which is our big effort to improve the way that the PIP process works. I have talked to the Committee about this once before. We are trying out something where, when you apply to PIP, a caseworker will phone you up and say, “I’m from DWP. I am going to help you on your journey and your PIP application. It is a complex process. This is what the benefit is for. This is the evidence that you will need to let us know about and this is why.” At the end of the process, when the person gets a decision, particularly when it is a decision that someone was not hoping for, that person will explain it to them. They will explain why that is the case. If they were not eligible for PIP, they might be eligible for some other form of support. That is a more expensive thing to add to the process, but it could be worth it if, as a result, more people understand the nature of a decision and we do not get further appeals. There is an accrued cost thing there. The key thing to test is trust. Do people, as a result of this, trust DWP more? I am accompanying that with a further piece of research. I think we have published this. If not, we can publish it. There is some research that we have done into sentiment about DWP. How do people feel about DWP? What is interesting is the difference between the wider population and people who are customers of DWP and have a regular interface with DWP. Those who have regular interface with DWP have a more positive view about DWP than the wider population, which is a positive thing. It reflects the professionalism and care of my frontline colleagues each and every day. If we continue to follow up that research time after time, we can see whether these initiatives that I have described make a further difference and improve trust further to the levels that we need it to be. All that will come together to make an improvement in the numbers on underpayments and unfulfilled eligibility.

Just for clarity, your view is that the two main reasons for underpayment are either deteriorating health or poor public trust in DWP in terms of making sure that people get the correct amount of support.

Unfulfilled eligibility is a bigger part of this than underpayments. The two reasons are the under-declaration of a deteriorating health condition in PIP and the under-declaration of deterioration of health condition in disability living allowance. I am looking at causes of underpayments as well as causes of overpayments. This is a cause of underpayments. What can I do to reduce the level of underpayment that comes into the system in that way? That is what we need to do. Trust is part of that. For me, it is not that trust is low. Customer satisfaction is generally high. We publish the numbers—they are in the 80s in percentage terms—in our annual report and accounts. We need to go further. That is what I would do. These are the actions that we are taking.

To follow on from that, I was going to ask when you expect levels of unfulfilled eligibility to fall, but I am not going to ask you that because I am presuming that you are relying on this “Tell DWP” campaign to make that happen, at least in part. What measurements do you have in place post or during that campaign to see how successful it is?

There are a number of elements to that. One is traction and take-up. Are people using this service and taking action as a result? You can then look at the overall declarations of changes of circumstance more broadly, because you do not want a situation where it goes up through the new channel, but that is just all the same people who would have told us anyway. We need to know that. The crucial part of our overall controls on overpayments and underpayments is our annual statistics on fraud and error. They tell us the headline numbers and the numbers that you see in the report, but I brought with me—I am not going to share because you do not need to see it—the full heat map, which tells me, benefit by benefit, underpayments and overpayments, where the main causes of loss are and how that has changed over the last year. This is the sort of material that I interrogate to make sure not only that we are doing the things that we said we would do in the plan, but that the plan remains up to date and relevant, so that we know there are not new things that we need to take action on in other places.

Finally, you have explained very well—it is very encouraging—that you are setting up this caseworker partnership with your clients to improve people telling you about changes of circumstance. In the case of people who have disabilities, do they notify you online sometimes? The caseworker thing might not click in.

There is a variety of ways of letting us know. To be honest, it is less about the caseworker. If I am thinking about how the process would go, the caseworker would be involved in the application process. They would be a voice of DWP—I hope a trusted voice of DWP. My hope is that, when people get to the end of the assessment process and get their decision on PIP, if they go on to PIP, they have a positive view about the experience. Let us say that, sadly, there is a deterioration in their condition. Some of the people I have talked to have gone to talk to their doctor and their doctor has said, “Your condition has deteriorated.” That has been the prompt. Our connection with GPs and other medical professionals more broadly is a very important part of our work across DWP. They would probably phone DWP to say that they want a reassessment, and then we would prioritise them for health assessment, get them through the system and get a decision on whether there is an enhanced amount as a result of that.

On a previous visit to this Committee, I raised a particular concern about significant potential fraud in a particular part of my constituency, which I notified the Department about at the end of last year. I wrote to you about it afterwards. You acknowledged it as a Department and said that it was being looked at, but you could not comment on the details because you do not comment on live investigations. I completely accept that. In terms of taking the public with you on these issues, at some point, when investigations are complete, it will be helpful for the Department to notify those who, in the public interest, have raised concerns with you about what has happened when you have a final position to tell them.

Yes, there is something in that. We are doing that more internally. Quite a lot of our referrals come from our colleagues within DWP. When a colleague raises a fraud referral within DWP, we ask them whether they are willing to put their email address, which means we can let the colleague know that what the result of that fraud investigation was. We should think about something similar. I would not want to compromise the investigation, but you are not saying that. There is something in terms of encouraging people to feel like they can engage with us, that it is worth doing and that it is not just putting their evidence into a black box that they never see again. We are working on that with our own colleagues, because so many of our referrals come from frontline colleagues. That is a good challenge to take away.

It would be helpful to have a response when you have had a further look at this. It was exactly that. This came from constituents, who gave quite a lot of evidence to me to pass on. They were happy to do that, but there is a bit of a feeling out there of, “We give it to them and they never do anything about it.” That may not be the case, but people do not know what you have done about it.

It is a very fair challenge. Vikki, is there anything that your team—

I know about this case, but I do not think I should say anything.

No, I am not asking you to comment on it.

You should not comment on individual cases.

As Sir Peter said, we should take away how we feed back ourselves.

The point that Clive was making was not really about an individual case. Particularly when we raise a case—this has happened many times—you say, “Yes, we’ll look at it,” and we never hear anything more. It would be quite useful for our constituents to know that you have looked at it and resolved it one way or another.

Let me give you an example that is in the public domain to illustrate some of the timescales. You will have heard of Operation Volcanic. This was the £54 million fraud that a Bulgarian organised crime gang did.

Yes.

Mr Lowe has heard of it—good. The original arrests were in May 2021. The investigation had been going on a number of years before then. It came to court last year. Sentencing was last year. Right now, today, the confiscation hearings are going on. These cases have a long life to them where they are a series of organised frauds. But I get the spirit in which the Committee is asking us.

That is the point.

There was a Bulgarian whistleblower, according to the media.

Yes—I would not believe everything that is in the media.

That is how they reported it.

Yes, indeed. He certainly did say that he did that. We were on the case a bit longer before then.

We must not get too drawn into one case. Just to finish up, I have two or three chasing points. One concerns Lloyd’s question on the fraud, error and recovery Bill. That Bill is giving you some pretty significant powers: powers of entry and powers to take money from people’s bank accounts. Going back to your point about trust in answer to Clive’s question, Sir Peter, trust is really important. You want people to co-operate with you. With these serious powers, you do not have to get too many wrong and reported, as Mr Lowe says, and it will soon erode that trust quite carefully. I know you have HMICFRS looking at these cases, but are you going to put sanctions in your own organisation to make sure that your use of these quite draconian powers is properly evidence based?

There is the backstop that comes from the oversight of HMICFRS. Within the Department, we will make sure that we have case conferencing and a collective view before any decision is made. As accounting officer, I need to be reassured that the powers are being properly used and used only in the circumstances in which they need to be used. We will make sure of that through governance, case conferencing and coming together. That is how we work in terms of investigations more broadly. I have seen some of that in action. Vikki, do you want to talk any more about the specifics?

There are two or three things that are worth saying. We are putting out the codes of practice over the coming weeks, and we will be consulting on those over the 12-week period. Those will give more detail about how we will ensure the external safeguards and internally how we will operate these. We have the external stakeholders, but it is also absolutely right, as you say, that we have the right checks and balances in place. During the passage of the Bill, we added further safeguards internally. For example, when we have the data push from banks, that data alone will not trigger an investigation. It has to be supported by other evidence. There are a number of other safeguards that we have put in place. If I think about the debt powers that you refer to, that is about fairness and making sure that we treat people who choose not to pay those debts back in the same way that we treat people who are on benefits and on PAYE. We will make sure that those people have the financial resilience to be able to recover those. Ultimately, there are some external further safeguards. We have to go to court if we get into the unfortunate position of the removal of a driver’s licence. That has to be done by the courts. When we do search and seizure, again we have to go for a warrant through the court. That requires us to come up with that level of evidence.

I have one further operational point just to reassure the Committee. While these are new powers with new safeguards attached to them, as Vikki and Sir Peter have been setting out, my investigators already operate under the Regulation of Investigatory Powers Act, for example.

That is a good point.

They are scrutinised every year on our use of those powers. We are scrutinised on other pseudo-police powers that we had already before this Act. That is all done externally. In one of the recent reports, the person reviewing us said, “You are about the size of a medium-sized police force, and I have not found any problems with how you are using these powers.” We take this very seriously because we recognise that we have to work on the basis of trust. When we come to Parliament to ask for more powers, as occasionally we do, showing that we have used the powers that we have in a proportionate and legitimate way is extraordinarily important.

As a final question from me, how are you getting on with your One Customer systems? It seems to me that part of this whole fraud thing is to do with what a customer might be doing in one benefit or another benefit, and you can link the two together with your systems. How are you progressing with that?

It is progressing well. There are two related systems. One is customer account, which enables a customer to see all their different interactions together. That links to the conversation that we were having with Mr Hatton about how we can find a single way for someone to let us know about a change of circumstance that might have implications for a wide range of benefits that they might be on. We then have One Customer View, which is the name of our colleague system. That will enable us to display in one place all the different elements of a customer’s claim. We are rolling that out gradually, bit by bit. It requires quite a lot of interaction with other systems, as you will appreciate, but it is part of our service modernisation programme. It is part of helping us to link information that we have from different sources in order to provide people with a really good service and to crack down on overpayments and underpayments, where they are identified.

We have come to the end of our session. We have covered a very wide field this morning and we have had some very interesting information. There are some serious matters we need to follow up on, particularly this whole business about whether we can get to a situation where your accounts are not qualified. We had a lot of useful discussions there. Thank you very much for attending. As you know, an uncorrected transcript of this hearing will be published on the Committee’s website in the coming days. The Committee will carefully consider all the evidence that you have provided today and produce a report in due course, with recommendations. I thank you today for giving some very clear evidence. Thank you very much indeed. [1] Correction: Universal Credit fraud and error rate is 9.7% [2] Correction: extending the targeted case review to other areas, particularly pension credit, where we are looking to spend between £60 million and £70 million a year, bringing out between £30 million and £190 million of annual savings from there. [3] Correction: The targeted case review was initially expected to deliver a 1:7 return.