A headline count that cannot carry a contract decision
NHS England’s uptake and benefits page says 137 trusts are reporting benefits from the Federated Data Platform (FDP), yet the largest count it gives for any single patient-care product is 40. The independent evaluation meant to test those benefits delivers its final report in April 2029, more than two years after the contract’s February 2027 break clause. The Guardian’s report on 29 trusts that say they are not using the platform is the visible symptom of a wider gap: the state has a number for take-up, a before-and-after number for performance, and no result yet from the independent test of what the platform has caused.
Strategic Insight: The useful question is not whether 29 trusts or NHS England are telling the truth. Both accounts can hold at once, because NHS England’s headline count includes back-office tools. That makes the count a poor guide to whether hospitals are getting value from the platform.
Why is this a measurement problem and not only a Palantir row?
Benefit figures are already doing political work. When an MP asked in May whether the government would use the break clause, the written answer opened with the platform’s results before saying that the contract would be reviewed “and a decision made on its extension”. Medact, the campaigning charity that obtained the freedom of information replies, now asks whether ministers relying on “flawed claims” are “being misled”. That is its framing, and the charity wants Palantir out of the NHS, but the question only has force because the numbers sit so close to the decision.
One phrase, several thresholds
Put NHS England’s own figures side by side and they describe very different things.
| Measure | Value | What it counts |
|---|---|---|
| Trusts signed up | 170 | Signed a memorandum of understanding, end of August 2026 |
| Trusts live | 142 | Live on the platform, end of August 2026 |
| Trusts reporting benefits | 137 | Includes data analysis and submission tools for back-office work |
| Acute trusts “actively utilising” products | 129 | The department’s major projects return, April 2026 |
| Largest single product count | 40 | Trusts realising benefits from the inpatient theatres module, end of June 2026 |
| Smallest single product count | 5 | Trusts realising benefits from Cancer 360, end of June 2026 |
The product counts come with a demanding definition. NHS England’s methods page counts a trust for discharge tool OPTICA, for example, only where it is “actively being used by operational teams”. The methods page has no matching entry for the 137, and the phrase “reporting benefits” does not appear on it. On the main page, a footnote explains the 137: some products support patient care while others “relate to data analysis and submission tools”.
So the same language covers two different conditions. The list of live trusts marks 137 of its 142 trusts in bold as “realising benefits”. When we checked on 4 November, all six trusts the Guardian quotes by name were in bold, including Torbay and South Devon, which told the paper that it is “not currently operationally using the FDP”. NHS England’s reason for listing it as live, the paper says, is infrastructure “to support data submission”.
NHS England does not publish how many of the 137 use at least one of the five patient-care products reported on the page. Adding the product counts together would not answer that, because each product is counted separately, so one trust can appear under several. That missing figure would do more to settle this dispute than any other single number.
Critical Context: Twenty-nine is about a fifth of 137, but it is not a measured rate. The Guardian describes over 100 information requests, obtained by a charity campaigning against the contract, and we have seen only the replies it quotes. The Guardian also says 137 trusts are “live”; NHS England’s page says 142 are live and 137 are reporting benefits.
What does the evidence show beyond the count?
Palantir’s defence rests on three figures, given to the Guardian as “117,000 additional operations, a 14.3% reduction in discharge delays for long stay patients and a 5.6% improvement in the number of people finding out whether or not they have cancer within 28 days”. They most closely match three different products on NHS England’s page: inpatient theatres (40 trusts), OPTICA (30 trusts) and Cancer 360 (5 trusts). The last figure therefore describes a small group of trusts, not the platform’s reach. The operations figure is described there as cumulative to the end of June 2026, not as the single-year total the Guardian’s wording suggests.
All three are before-and-after comparisons. NHS England now says of such observations that it “cannot draw conclusions about cause and effect as other variables have not been controlled for”. The Office for Statistics Regulation (OSR) records that a caveat sentence was added to the methods page on 6 June 2026. Ten days later the written answer told Parliament “there has been a 14% decrease” in discharge delays for patients staying more than seven days, without that qualification. The OSR said in July that it expects any figures cited by ministers to be “communicated in a clear, accurate manner”. Palantir’s statement to the Guardian, as quoted, carries no such caveat either.
The seven-day figure has also moved. It was 14% in that June answer and 13.58% when the Health Foundation quoted it in a letter dated 22 July. It now stands at 12.50%. NHS England’s methods page discloses a change to its baselines, but we cannot say how much of the fall that explains.
The comparison that exists is disputed
The Health Foundation has released a comparison with trusts that did not adopt the tool. In a letter to the OSR, its chief executive, Jennifer Dixon, reported “no meaningful difference in delayed discharge performance” between trusts that adopted OPTICA and those that had not. She added that this was “not a criticism of OPTICA”. The foundation’s analysis calls itself descriptive, and lists an inferential or causal evaluation among the things it does not attempt.
NHS England disputes the method. According to an OSR freedom of information response, it told the regulator that the foundation calculated from day 0, while its own figures look at stays of 7, 14 and 21 days or more, in line with how the tool is meant to be used. The OSR did not complete its own quantitative analysis. City AM reported an unpublished NHS England document, seen by the paper, showing falls of 14.1% to 14.9% at OPTICA trusts against rises of 2.9% to 3.5% elsewhere, while cautioning that the figures do not show OPTICA was the only cause. Tom Bartlett, a former NHS England deputy director, told the paper there are “problems with both of the methodologies”, though he also called the “no noticeable improvement” finding “absolutely wrong”.
Reality Check: Two comparisons of the same tool point in opposite directions. The one that goes against the platform has been released by the Health Foundation. The one in its favour had not been published when City AM reported it in August, so readers could not check it.
Cost and benefit have a measurement gap too
The £330m headline is the Palantir contract, not the programme. In the latest government major projects data, the department puts whole-life cost at £1,084.31m and forecast benefits at £808.28m, in 2024/25 real prices. It says cost has increased and forecast benefits have decreased, and that it expects lifetime benefits “to continue to grow”. The same entry reports a green delivery confidence rating.
City AM reports Palantir saying the latest published ratio remains £4.92 for every £1. A ratio of 4.92 appears in a project assessment reported by PublicTechnology, dated May 2024, and applies once annualised benefits are in full effect. We have not been able to reconcile it with the newer totals, and we would not treat either as evidence of realised value: both are forecasts.
Strategic Reality: A green delivery rating and a falling benefits forecast can sit in the same government record. Delivery confidence answers a different question from whether the platform has caused better care.
Who carries the consequences of a weak number?
| Group | What changes for them | What is still unknown |
|---|---|---|
| Ministers, including health secretary Yvette Cooper | AFP reports a decision expected by December, with benefit claims now openly contested | What weight the decision gives to benefits, cost or timing |
| Trusts on the bold list | Their names support a national claim; 29 replied that they do not use the system or its apps | Which use each trust is counted for |
| NHS data leaders | Marc Farr, who chairs their network, wants benefits reported so they “can be properly verified” | Whether the December update changes the headline |
| Palantir and any challengers | Palantir’s case relies on figures NHS England says cannot show cause and effect | Whether any supplier could take over in time |
A reader who sees that a hospital is “realising benefits” might assume staff there use the product. On NHS England’s own footnote, that need not follow.
What this means for public bodies reporting AI benefits
The OSR’s July statement sets out a usable standard that reaches beyond this contract. Before-and-after metrics should be labelled as such, with clear information about causality. Statistical teams should work more closely with media teams and ministerial private offices. Case studies should say who wrote them. Other public bodies reporting benefits from an AI system may face similar questions, since the OSR considered part of this case under its Standards for the Public Use of Statistics, Data and Wider Analysis. The lesson we would draw is simple: decide what counts as use before the first count is published, and report connection and active use as separate numbers.
The open challenges
The evidence arrives after the decision
Imperial College’s strategy says the final report will be shared “at the end of the project, in April 2029”. Its timeline section gives no earlier publication date, and the word “interim” does not appear in the document. Quantitative work starts with theatres and OPTICA, and the plan assumes data access for three more products by January 2027. Unless earlier results are shared, the benefits evidence in front of ministers before the break clause will be observational figures and contested comparisons. Even the final report will not end the argument: the authors say their methods “cannot fully eliminate residual confounding or establish direct causality”.
Timing may already outweigh evidence
Health innovation minister James Frith told the Commons health committee on 21 August that, if a successor contract is required, it is “unlikely that a new contract could be in place by March”, with the assessment due to finish in the autumn and no decision yet reached. On that account, the benefits dispute may matter less to the 2027 decision than the calendar does. Layla Moran, who chairs the committee, had argued in July for preparing an alternative “in time for spring 2027”, while Bartlett told Computer Weekly that “there is no alternative to the current supplier” in that window.
A clean comparison is hard to build
Fair comparison is difficult on this platform. The OSR notes “there is no common adoption date for FDP at trust-level”. Bartlett told City AM: “If you’ve got a hospital that’s got 10 wards and two of them are using Optica, you should be doing an analysis of those two wards, not of the 10.” A national percentage hides that variation. Trust-level data would at least show the spread.
What to watch
NHS England says the next update to its page is due in December 2026, and AFP reports that ministers are expected to decide by the same month. Four signposts will show whether measurement is improving.
- The December update. If the 137 is split into patient-care use and back-office use, the “realising benefits” label becomes testable. If it stays a single figure, the dispute continues.
- The trust-level schedule. NHS England committed, in the OSR’s July statement, to prepare a publishing schedule for trust-level benefits data. The main page and methods page showed none on 4 November. Publishing it would let outsiders run the comparisons now being argued over.
- The OSR’s new casework. The regulator told the Guardian it is “taking this forward under our casework function”. A finding on the label itself would go further than July’s statement, which dealt with causality and presentation.
- The reasons given for the decision. A statement that rests on commercial or timing grounds would be defensible on its own terms. One that leans on the operations figure would sit badly with NHS England’s own caveat.
Strategic Insight: Any benefits figure makes one of three claims: trusts connected, trusts using a product, or a controlled effect on patients. Only the third speaks to whether the platform caused better care.
The contract may well be renewed or ended for reasons unrelated to any of this: cost, supply, sovereignty or politics. That would be a legitimate choice, as long as it is described as one. What the public record cannot yet support is a claim that the decision rests on proven benefit, in either direction. It would take little to improve on that before February: one split number and one published comparison.
Source: Palantir AI tech unused or little used by 29 NHS trusts said to be benefiting from it (The Guardian, 2026), by Robert Booth. Figures checked against NHS England’s uptake, methods and live-trust pages and the NISTA major projects data; ministerial decision timing from AFP via France 24 (October 2026). We covered the pressure on the contract in October.
This strategic analysis was written by Resultsense, a UK-focused AI news and analysis publication. We will be watching whether December’s figures separate connection from use, and what ministers cite when they decide. Read more analysis at Insights, or get in touch.