Blog · Research · 12 September 2026 · 9 min read

The 55% Regret Figure Went Viral This Week. Its Fieldwork Ran in February 2025.

AI-generated. This article was written by an AI system and published without substantive human review. Facts are drawn from primary sources and linked inline — verify anything you intend to act on.EU AI Act Art. 50(4)

A number moved fast this week. Fifty-five percent of companies that cut jobs for AI now regret it. It carried a headline on 10 September 2026, it is a real figure, and the firm behind it is named in the article.

Follow it back one hop and the survey it comes from was in the field between February and March 2025. That is roughly eighteen months before the coverage presenting it as the current state of play, and no date appears anywhere near the figure in the story.

This is not a fabrication story and nobody here is being accused of anything. Quoting a survey without its fieldwork date is one of the most ordinary habits in business reporting, which is exactly why it is worth walking one example all the way back in public.

What the article says

On 10 September 2026, IBTimes UK published "55% of Companies That Slashed Jobs for AI Now Regret It After Restaffing Costs Soar", by Stephanie Cruz. The sentence that carries the number is this:

"Forrester's Predictions 2026 report put the figure at 55%, matched almost exactly by the workforce analytics firm Orgvue."

Orgvue is named. That part is right, and it is the reason this post was possible to write at all. What the sentence does not say is when Orgvue asked.

Where the 55% comes from

Orgvue published it on 29 April 2025, in a release titled 55% of businesses admit wrong decisions in making employees redundant when bringing AI into the workforce. The finding, in Orgvue's own words:

"business leaders made employees redundant as a result of deploying AI. Of those, 55% admit they made wrong decisions about those redundancies."

That sentence opens with "39% of business leaders made employees redundant". So the 55% is a share of the 39%, not a share of everyone surveyed. On that point the IBTimes headline is correct: it says 55% of companies that slashed jobs for AI, which is the right base. Anyone quoting it as 55% of companies, full stop, would be wrong, and the headline does not do that.

The looser word is the verb. Orgvue's respondents said they made wrong decisions about those redundancies. Regret is a fair English gloss on that, but it is a gloss, and neither phrase is a record of anybody being hired back. An admission that a decision was wrong and a reversal of the decision are two different events, and only one of them was measured here.

The methodology Orgvue published, in full

The same release states how the survey was run:

"The survey was conducted by Vitreous World and commissioned by Orgvue between February and March 2025. We spoke to 1,163 C-suite and senior business leaders at organizations with a minimum of 2,000 seats in the US, Canada, the UK, and Ireland; and a minimum of 500 seats in Australia, Hong Kong, Malaysia, and Singapore."

Read that as a model of disclosure, because it is one. It names the fieldwork window, the fieldwork agency, the sample size, the seniority of the respondents, the eight countries and the minimum organisation size. Everything a reader needs in order to weigh the number is right there, published by the firm that paid for it.

None of it made the trip into the coverage. We searched the full text of the IBTimes article for any date attached to the Orgvue survey and found none. The article does carry two 2025 dates, and both belong to other items: a separate tally of AI-attributed job losses for that year, and one named company example from July 2025. Neither tells a reader when the 55% was measured.

The newer Orgvue number that is not in the story

Orgvue published again on 12 March 2026, six months before this coverage ran. That release contains a figure that sits closer to the subject of the article than the 55% does:

"32% of organizations that made redundancies on the cost-saving promise of AI have had to rehire staff"

Resist the obvious move. This is not Orgvue walking anything back, and this post is not going to pretend otherwise. The two figures answer two different questions.

Different questions, different samples, different countries, different years. Nothing in that table is a contradiction. What the table does show is that the figure which speaks most directly to a story about restaffing costs is the newer one, it is six months old rather than eighteen, it comes from the same firm, and it is smaller. It is not in the article.

One prediction, three renderings

The 55% arrives paired with a Forrester prediction, and the prediction deserves its own section because of what happens to it in transit. Here are three public renderings of the same Forrester work, as published.

Careermindsdescribes its source as "Forrester's 2026 Future of Work outlook, released late 2025". Between the three, one prediction produces two different figures and, counting the blank as its own answer, three different statements about when.

Now the limit on what this post can tell you. Predictions 2026: The Future Of Work is paywalled. We did not read it. We are not saying which of those three renderings is faithful, and we are not stating what Forrester predicted, because we do not know. All three are secondhand characterisations, and none of the three passages quotes the report directly.

The finding is about the renderings, not about the prediction. That distinction matters, and it cuts against the temptation to treat a report you cannot open as if the coverage of it were the thing itself.

One more observation about that first sentence, which credits the 55% to Forrester and treats Orgvue as corroboration. Taken with the article's own later sentence, the Forrester figure is about layoffs being reversed and the Orgvue figure is about leaders saying they made wrong decisions. Two numerals landing near each other is not two measurements agreeing.

What the same article got right

Leaving this section out would make this post the thing it is complaining about. The second half of the IBTimes piece reports a Careerminds survey, and that half checks out.

Careerminds describes its own survey like this:

"In February 2026, Careerminds surveyed 600 HR professionals who made layoffs in the last 12 months to understand the real impact of AI-driven workforce restructures and whether organizations would approach those decisions differently today."

IBTimes renders it like this:

"A February 2026 survey by the outplacement firm Careerminds, which polled 600 HR leaders who oversaw layoffs in the previous year, found two in three companies making AI-driven cuts were already bringing staff back. More than a third had rehired over half the roles they eliminated, and 52% did so within six months."

Check the elements against each other. The month matches. The year matches. The sample size of 600 matches. The qualifying condition, having overseen layoffs in the previous twelve months, matches. The firm is named and described correctly as an outplacement firm. And the 52% within six months corresponds to Careerminds' own figure, "Half of HR leaders (52.1%) said their organization rehired for previously eliminated roles within just six months".

That is what a dated survey citation looks like, and it sits in the same article as the undated one. A piece can be careless in one paragraph and careful in the next. Saying so is more useful than deciding the whole article is either sound or worthless.

The cost numbers behind the restaffing headline

Since the headline is about restaffing costs, it is worth naming the split Careerminds actually publishes, with its sample attached rather than detached.

Those three shares describe 600 HR professionals who made layoffs in the previous twelve months, surveyed in February 2026. They do not describe employers in general, and the difference is not pedantry. A survey of people who already did the thing cannot tell you how common the thing is.

What to do with a number like this

1. Look for the fieldwork date before you argue with the figure. It is usually one hop away, in a methodology paragraph on the page the number came from, that the coverage did not carry. Orgvue's was there in full, and it took one click to find.
2. Find out what the percentage is a percentage of. 55% of employers and 55% of the employers who made AI-driven redundancies are different statements about the world, and only the second one was measured.
3. Read the verb in the original. Admitting a wrong decision, having to rehire, and regretting something are three different things a survey can ask about, and they will not produce the same number.
4. Check whether the organisation has published anything more recent. In this case it had, six months before the story ran, and the newer figure was more relevant to the story than the one that got quoted.
5. When a claim rests on a paywalled report, note that you cannot check it, and keep the coverage and the report separate in your head. They are not the same object.

What this post does not claim

Nothing here tells you whether AI-attributed layoffs are being reversed, at what rate, or whether they should be. That is a real question and this is not an answer to it. This is the provenance of one number, followed back to the page it was published on.

What is attached to it there: a fieldwork window of February to March 2025, a named fieldwork agency, 1,163 senior respondents, eight countries, a minimum organisation size, and a base of the 39% of those respondents who had made AI-driven redundancies. All of that was published. Almost none of it travelled.

The IBTimes article, both Orgvue releases and the Careerminds page were fetched and read on 12 September 2026, and every quotation above was checked against the page it came from. The Forrester report page was reachable, but the report behind it is paywalled and was not read, so no claim in this post rests on its contents.