The Office for National Statistics published its latest Labour Force Survey quality update in July, and the top line is genuinely good news: responses are back. In January to March 2026 the LFS drew 80,078 responses, up 35,840 from the low point in July to September 2023. Anyone who has commissioned survey work in the last five years should be interested in how that happened.
The recovery was bought, not won
The ONS is unusually direct about the cause. Response levels have improved since 2023, it says, "because of interventions, including increases to survey incentives and interviewer numbers, to improve quality."
Read that again with a procurement hat on. The national statistical institute — an organisation with a legal mandate, a trusted brand, and no commercial motive that anyone could object to — could not restore its response volumes through goodwill. It restored them by paying respondents more and putting more interviewers on the phones. If that is what it takes for the ONS, it is worth being sober about what it takes for a housing association contacting tenants, or an agency running a customer feedback wave.
And the recovery is not complete. The 80,078 figure sits 3,984 below the 84,062 achieved in October to December 2019.
Levels are not rates, and the difference is the whole story
Here is the distinction that matters, and it is the one most easily lost in a headline. Response levels are how many completed responses you get. Response rates are what proportion of the people you approached agreed to take part.
The ONS recovered its levels substantially. Its response rates remain more moderate and below pre-pandemic figures — the volume came back in significant part through a larger issued sample rather than through a higher share of people saying yes.
That is the honest picture of fieldwork in 2026. Willingness to take part in surveys has not recovered. What has recovered is the willingness to spend enough to compensate for it. You can buy back volume. You cannot buy back enthusiasm, and every extra name you have to approach to hit the same completed count is cost that lands on someone's budget.
Why this should shape how you commission research
Three practical consequences.
The first is that "we got our target number of responses" is a weaker claim than it sounds. Ask how many people had to be approached to get there, and what the profile of the non-responders looked like. A sample that hits its count by over-issuing to the easiest-to-reach groups is not the same sample.
The second is that incentives are now a line item, not a nice-to-have. The ONS found they moved the needle. Budgets built on 2019 assumptions about who will answer for free are budgets built on a world that no longer exists.
The third is about channel. If reaching people is the binding constraint and the cost per contact attempt is what governs the budget, then the economics of any method that lowers the cost of an attempt change materially — which is the genuine, unglamorous argument for automating the repetitive tier of outbound calling. Not that a machine interviews better than a person. It doesn't, and anyone claiming otherwise is selling. Only that when the constraint is the sheer number of attempts needed, the cost of an attempt is the number that decides whether the study happens at all.
Related: AI vs human interviewers: what actually differs · Can an AI run a survey people actually finish? · What conversational AI actually does for a business