This transcript was produced using AI and has been adjusted for clarity.
James Benford: Surveys and economic statistics is moving to a new phase from recovery to improvement. The Labour Force Survey (LFS) is now back on track after the issues affecting telephone interviews in May and June, supported by a larger interview field community. Achieved responses for our digitised business surveys are reaching historic highs. Improved methods for the national accounts and labour market statistics are delivering new insights about the UK economy. This is giving us a better understanding of the service sector, public sector balance sheets and productivity. It is helping our statistics keep pace with the growing importance of AI and digital infrastructure.
Our key performance indicators point to further encouraging signs on quality and delivery. There have been no higher impact corrections since January this year. The 6th month rolling correction rate is around half its level at the start of the year, and all tier one economic statistics have been delivered to schedule. We've maintained that performance despite some systems and data supply issues recently and have actions in trying to learn from those issues and strengthen processes further. Luke Ashton, our new Director General for Digital and Data, is rapidly coalescing the office around a new targeted technology architecture and a plan for legacy. We'll have more to say about that and what it means for our statistics in the new year. We continue to work with the Office for Statistics Regulation (OSR) towards re-accreditation of producer price indices and UK trade statistics. Recent feedback from the Public Accounts, Public Administration and Constitutional Affairs Committee and the OSR has recognised the progress being made, emphasising the importance of transparency, quality and sustained delivery. The OSR has welcomed the clearer and more disciplined approach to managing surveys, income and statistics, and has moved from quarterly to annual progress assessments.
We have prioritised recovery of the LFS following the operational issue in May, which temporarily reduced the number of achieved telephone interviews. Performance has since recovered to levels seen during the equivalent period of last year, and we've set out actions that we're implementing from our lessons learned review, which are strengthening our operational controls broadly. Learning transparently when things go wrong is really important to improving surveys and economic statistics and trust in them.
Our social surveys field community has reached its minimum target for the first time since this improvement plan began and is set to increase further in the coming months. This will continue to support the LFS and the Transformed Labour Force Survey (TLFS), while providing additional capacity for the Living Cost and Food Survey and the Wealth and Assets Survey later this year.
The performance of the TLFS has continued to improve following the changes introduced in April, with a further increase in the achieved sample size from July, as more survey cohorts experienced the changes that we introduced. We continue to progress work to test our innovative approach to determining occupation industry codes and to integrate plans for the Transformed Labour Force surveys across Great Britain and Northern Ireland. We will publish our next update on the TLFS in November.
The Annual Survey of Hours and Earnings (ASHE) has achieved its largest sample on record following the move to the digital by default collection, strengthening one of our most important sources of information on earnings and pay. We will publish the results of that survey on the 22nd of October, so they are available to inform decisions on the minimum wage in the upcoming autumn budget.
Exploratory research using HMRC Pay As You Earn (PAYE) data has also provided new insights into labour market trends, helping to explain periods of divergence between survey and administrative measures. It showed a stronger recovery in employment after the pandemic and a smaller and less prolonged rise in economic inactivity than suggested by the published LFS estimates alone. This research also suggests that recovery in the LFS during 2024 and 2025 has reduced non-response bias as using the linked payroll data made only a small difference to the level of employment and inactivity estimates in 2025. The next phase will examine how we can incorporate admin data on self-employment.
Much covered, new component approach to labour market, labour productivity combines multiple data sources and international best practise and has helped to explain a significant part of the so-called productivity puzzle. While the data still points to a slowdown in UK productivity following the financial crisis, the new methods change our understanding of the slowdown in output per worker growth after the crisis with a larger role for falling average hours. The new data make more sense. We'll incorporate the new methods into headline productivity estimates in November and bring in regional and industrial breakdowns early next year.
We have recently published new GDP data in line with the 2026 Blue Book. Rich information from the Annual Survey of Goods and Services has improved our measurement of the service sector, revealing a larger service sector with more secondary production of goods, increasing the estimated level of nominal GDP in 2024 by half a percent. Updated estimates of public sector pension liabilities strengthen measurement of Public Sector Net Financial liabilities, giving users a more complete picture of the public sector balance sheet.
We've also published new analysis of data centres and digital infrastructure investment, which reveal on the improved measure, strong investment during 2024 and 2025. We set out a framework for a UK AI thematic account, supported by an analysis of more than 5,800 UK companies specialising in AI. Together, this work is helping ensure our statistics keep pace with an economy increasingly shaped by AI and digital infrastructure. In December, we will publish our first estimates of UK investment in data as an asset, a foundational part of the growing AI ecosystem.
Over the next quarter, we will continue to deliver improvements for our users. I've already highlighted some forthcoming work on the labour market, on productivity and the national accounts. In addition, in November, we expect to publish the first experimental statistics from the Linked Employer Employee Dataset (LEED). These really exciting statistics will show how workers move between and within firms over time, how these patterns vary by workers age and by the size and industry of their employers, providing new insights into how the UK labour market is changing.
We are working to deliver new measures of the underlying movements in GDP and of monthly consumer price inflation, drawing on best practise at other national statistics institutes. The Statistical Business Register (SBR) remains on track to deliver its first transition state by the end of this year, reducing reliance on legacy systems and providing a platform to explore the composition of the economy using new SIC codes that we finalised earlier this year.
Of course, significant challenges remain. Running the LFS and TLFS in parallel in 2027 will increase costs next financial year. We are managing these pressures actively by slowing recruitment outside of our priority field interview community over the coming six months and launching new productivity and savings initiatives. To protect improvements to critical statistics, we will need to be disciplined in taking on new work, until we have made the evidence led transition to the TLFS. We remain on track to deliver that transition in 2027 in line with the plans that we set out in August. We're engaging closely and transparently with our users and the UKSA board to ensure the transition works for them.
To conclude, we are building a statistical system that is reliable, capable of keeping pace with a changing economy and is responsive to user needs. User feedback remains central to our work. If you have comments, particularly if there's anything you'd like us to do differently, please let me know. And I'd also like to say that we have a consultation in the field at the moment running to the end of October on release times for market sensitive economic statistics and on whether the current arrangements continue to serve the public good. Please do feed in your views and I'll be very happy to take any questions now.
Q: Hi guys, thanks as always James for the update. The only quick question I'd like to ask is, considering the time you've been on it now, the best part of the year, are you happy with where you are or do you think you have found bigger challenges in getting to this stage of the review?
James Benford: I'm happy with where we are and I've been really impressed since I've joined with the expertise, dedication, the passion of ONS staff and that has really helped to get to where we are. I know there's further to go and I know that the biggest, in some ways, a litmus test for many of our users will be the transition to the TLFS. We're pursuing that in an open way. We're making good progress and have made further progress and can see it in the data since the last update. I'm happy with where we are.
Q: Hi James. One quick question on the Annual Survey of Hours and Earnings, which as you say is highly budget relevant, it was just to check whether there are any new changes to the methodology this year or whether that was settled in the last couple of years. And then secondly, you've flagged up the impact of the double running of the survey on recruitment. Would you be able to say what the additional cost of that double running is for next year and a little bit more about how it's changed recruitment. We spotted last week that you were renewing the contract for agency worker interviewers.
James Benford: Okay, happy to answer both of those. So, the ASHE survey, we're going to publish that in the coming week. It will have a blog with it, taking you through the latest data and what the increased sample size means, it does increase coverage for the survey, including of small businesses and high earners. And so that will give us a more accurate read on hours and earnings, but that better coverage, it does mean we'll have to think about the annual comparisons, and we'll set that all out in the blog. So, there's that change in coverage that is important to work through. But I think it's best to put the detail that sets that out in the blog, so you've got the figures alongside that.
So, on the double running into next year, broadly, the Labour Force Survey costs the ONS around 30 million pounds a year to run. The LFS and also its successor, the TLFS survey, are the largest household surveys in the country. And yeah, we've seen this year, the complexities of running the two in parallel. It's important to say that that cost isn't just a financial cost, it's an opportunity cost. And I've been on record before as saying that the quality of the LFS and the TLFS won't reach their potential whilst we're running in parallel because of the practical limits that that gives us. And it's limiting to our ability to improve other wider surveys, economic statistics, particularly the wider suite of household surveys.
In terms of how we're adjusting our recruitment plans, we're not changing our plans for the field force. You know, whilst we're double running, getting the size of the field force as big as possible is crucial, but we are adjusting our plans more broadly. Broadly, what we're doing there is, you know, we're set in terms of our recruitment into the new year into January, but rather than continuing to increase headcount in surveys and economic statistics from January, we're going to broadly hold headcount flat. That's a difference of around, by the end of the current financial year, around 200 people. The impact of that will be primarily next year. You don't expect new hires to have an impact on delivery immediately and it will mean that, as we've said for a while now, during that period of double running next year, which is extended as we target that transition to the TLFS in November, we'll be constrained in how much additional work that we can take on. And that will be an important topic through the coming business planning round.
Q: Just one very quick question. There seems to be a constraint in taking on interviewers on civil service contracts at the moment. I just wondered, is that a civil service wide constraint?
James Benford: I mean, for a while, we've been basically pursuing multiple channels and to increase the size of the field force. And actually, for a while, we've been going through third parties to take on contractors too. And that came actually with a cost, there's higher turnover when you take on through third parties. So, we are working to convert some of those who we kind of take on through a third party into either fixed term or longer ONS contracts to reduce turnover. That's having an effect. But we've been working through many channels for a while now to increase the size of the field force. I think there's issues across the survey sector in terms of the maintaining the size of the field community and the onus is no different there and we're working through a number of channels to do that and having some success. It's the first time, as I mentioned, that we've gone through the minimum target and expect to increase the size of the field community to over 900 in the coming month.
Q: Hiya, thanks for your time. Just a quick question on the incorporation of administrative data on self-employment. I was wondering what you guys were hoping this would show and if you had any concerns about the accuracy of current self-employment data.
James Benford: There's a few things there. One thing is in terms of the availability of current administrative data on self-employment, it basically comes through the self-assessment system, which runs with a lag of at least 18 months, sometimes longer than that in terms of submission of that data and processing of it. So, it's tricky to use admin data in real time. In time, the HMRC's initiative on making tax digital could give us a more regular feed, though that's dependent on putting the right systems in place, which is something we're working with HMRC on. And yeah, it's an area where so far, obviously, when you match the payroll data to the surveys, that gives us another way to calibrate employees, but it doesn't give us an additional data point on the self-employed. And that means we have got less ability to triangulate there. We can do so in the past and I'm hoping by bringing in the self-assessment data into the calibration, that might produce more insights. I mean, in terms of the question on concerns with the accuracy of that data, one feature with the research that we've done is that issues with non-response bias in the Labour Force Survey basically peaked in 2023 and then came down and went away in 2025. So, if that pattern is the same, in terms of the self-employed, then it would be something affecting historic data, but not necessarily current data.
Q: Hi James. I'm sure I'm sure you've specified this before, but it‘s escaped my recollection right now. I was just looking at the improvement in the response rates of the TLFS core, etc. Given that and also the costs that you talk about with the double running, I mean, is there an option of moving to the TLFS earlier than late 2027, and if so, would you say the case for that is strengthening? And just separately on the productivity data, I'm sure it probably was in the release that you put out last month on this, but are you basically saying that you're going to move to that updated methodology that you outlined last month kind of for the foreseeable, you know, basically we're not going to get productivity figures based solely on the LFS going forward? And are there any other indicators that are kind of relying on the LFS currently that you're looking at doing a similar thing?
James Benford: So, let me just start with the productivity one. So, yeah, we said that we're going to publish in November with the next set of quarterly productivity estimates the figures under the headline component approach that we published the methods article on that over the past month and that will show the evolution of productivity under the component-based approach up until basically the middle of the year, 2026 Q2. It's going to take us a bit more time to take that component level approach through to the regional level and to the industry level that will follow in the new year. We'll still publish the figures under the LFS approach, the so-called direct approach, but that won't be the headline.
In terms of other areas, one thing we need to work through - and I'm sure we'll have more to say on this over time - we're kind of moving iteratively here, is basically the measure of labour costs in the national accounts, which at the moment, the most timely data that for the last year or so is based primarily on the LFS and there's a question whether moving to a more component based approach could help there. That's something we discussed with the one of our advisory panels on national accounts and there's a few details in the minutes that you can you can see there. I should say that that's just an issue in terms of labour costs in the national accounts for the most recent data. As time goes on and we go through the so-called balancing process in the national accounts, we do bring in actually the data we talked about, the admin data on self-assessments and of random data. But yes, it was a question about whether we could anticipate some of that more quickly if we use the component-based approach.
On the TLFS, we set out in August a quite deliberate quarter by quarter plan and that leads up to a progress check in January, which we will publish the results on later in that quarter, and then a decision in July about whether we're basically on a path to decommission the LFS and move to the TLFS in November next year. And that's kind of our final decision on that. It's really important as we kind of work through all that, we take our users with us and take the board with us. And it's really important as well, a bit like we did for the scanner data that we prepare the ground for what's coming with publication of, you know, the TLFS data and explanations of where that sits relative to TLFS. So, we have got quite a bit to do to kind of clear through the final pieces of the plan, including working through how we integrate data across Northern Ireland and Great Britain. We're just responsible for the Great Britain part of the data production, but we need to incorporate data for Northern Ireland who have their own TLFS and finalising how we're going to do coding of industry and occupation codes. And we need to take users with us, so I, I don't see a case for going faster than the plan that we set out.
Q: Hi James, thanks for that. I actually wanted to ask a more broad question. Just in terms of how you've adapted your processes or measuring to assess how AI is changing the economy and what's going to come up in, or what are you going to reveal potentially in that first estimate of data as an asset in December.
James Benford: Well, I won't preview figures, I'll never do that, but this is a great topic and one I think we'll spend more and more time on as time goes on. And there are quite a few dimensions to it. And there's a lot here that you know, where the methods aren't fully set and need to be worked through, including internationally.
I don't think I'm saying anything too controversial by saying that the new standards for national accounts agreed at the UN last year didn't fully anticipate the AI that we're seeing and how it's been adopted. How could they have? Those standards were developed over 10 or more years. And so, we will need to think closely about the AI that we're seeing as we implement those standards. Let's start with the data as an asset and I'll go to into AI more broadly.
So, data is the foundation for AI. That's true at the economy level. It is true for individual companies and what you're able to do with AI is dependent on the data that you put in it. It's interesting that when you look at companies' accounts, they don't list data as an asset or investment in data as an asset. It's not there. The kind of corporate accounting standards haven't got there yet. So when we're trying to do that for the national level, it's a tricky task and the guidelines that have come with the new standards and all statistics agencies are following those, basically guide one towards looking at people's time and professions that are closely associated with collecting data, curating it, better improving the quality of it, categorising it and so on, and using those professions and how they spend their time to estimate investments and data. That's the method that we've worked through. We've done it in lots of different ways. Actually, the Annual Survey of Hours and Earnings is a very good way of doing it. The vacancy data is a good way of cross-checking what we're doing and so that will produce good estimates in line with the standards. There's one caveat here which all statistics agencies are thinking through a bit, which is, you know, these days, and it's not something that's so true historically, but the way that data is collected, curated, processed, it's not just people's time, but it's the AI and machines time as well. So, I think as we iterate on those estimates and they are very much a, you know, a first go at it and everyone's doing the same thing here. We'll have to think about how we incorporate the changing way that the data has been collected and processed. So that’s one piece.
Lots of other areas, a very old area, and many of the problems here aren't very new, is the so-called production boundary, which is what's included in GDP and what's not. And you know, the historic convention still true is that the production of goods and service in the household sector is not in GDP. And and also freely provided digital business services and not in GDP. And there's a, you know, digitalization kind of puts that in sharp relief. A very good old example is basically travel agents, right? And one used to go to them on the high street to book things and then now he could just do it online and that caused the shift of activity and towards the household sector. I think AI will create lots of those types of effects and we'll have to think through how we're capturing what's being consumed in the household sector and maybe how households are using AI to do things for themselves.
That's one important area. I'll bring out two others and then I'll shut up. Sorry, it's quite a long answer for your question. One is, is basically how we measure the capital stock and economy, intangible assets, and AI is giving us lots of new forms of intangible assets. And it's not just, you know, the big models that are being produced, and rented out by the large AI companies. It's the applications that companies are building to use those. That's an asset for that company. And it's really tricky to measure those investments. And the core problem here is similar to the one I mentioned on data as an asset. Right. So typically, we measure investments in software. AI is basically a different form of software by looking at time spent by software developers and we can measure that. But now we've got like software that can be written by everyone. You do vibe coding. We've got software can write its own software. So how do we measure? That's a big, big question to work through.
The 3rd general area, which I'll highlight, is basically deflation and price adjustments. So, the longstanding issue, but it's there in spades for AI, is quality improvements. Your $25 for ChatGPT gives you a lot more than it did last year, a lot more than it did the year before that. So, we need to think through how we account for quality improvements too. There's a lot to work through. The thematic account we published was very much a first step, and we're going to work through things iteratively because this is an area where we need to learn as we go and learn from each other and we'll do that transparently.
Q: My only follow up might be, you're suggesting that how you measure this in terms of GDP might be different from other countries. Might that make international comparisons difficult?
James Benford: No, no, no. So, we're working to implement these new international standards by 2030. Everyone's doing that. But the way that we're all doing that is to introduce experimental releases as we go. And so, I think what you'll see is different NSIs will do their experimental releases, seek feedback, compare across the guidelines. This will evolve and we definitely want to preserve international comparability. That's one reason why it's quite helpful to have this long implementation time for the new standards.
Q: A very quick one. You mentioned the consultation and the close end of October. Have we got a post-close timeline - when the next stages are, any announcements?
James Benford: Yeah, I mean, on that, what I've said to the OSR, the regulator, is that we'd get back to them by the end of this year. So, we're closed in October and we'll try and reach the - or at least have a firm update - by the end of this year.