Four-day week adoption across three sectors
Written for a journalist reader
Revenue in the large 2022 UK pilot did hold steady, and for two publicly filing participants that is corroborated by audited accounts. [unbacked] The stronger claim — that output per worker is unaffected — does not survive: the researcher conceded that it rests on revenue as a proxy no source validates. [unbacked] A survivorship question about firms that dropped out remains open and unquantified. [unbacked]
Claims that survived the critique.
Firms in the 2022 UK pilot reported revenue broadly unchanged over the trial period, and for two publicly filing participants audited accounts match those self-reports. [unbacked]
Held up because: The critic argued the figures were unverified self-reports from motivated volunteers; the researcher produced statutory filings for two participants that match. That corroborates part of the sample, not all of it.
C1C4Most published four-day-week evaluations are single-arm before-and-after designs with no control group. [unbacked]
Held up because: The critic examined this claim and did not challenge it; it comes from a peer-reviewed review that states it directly.
C3
Both positions, and what evidence would settle each one.
- C2criticalO2
Four-day-week adoption does not reduce output per worker in knowledge-work settings.
Software firms Trial Reader
Conceded. The inference ran from flat revenue at constant headcount to unchanged output per worker, and no source measures output directly. [unbacked] The researcher withdrew it as a claim about output and lowered its confidence to low.
Devil's Advocate
Revenue is not a validated proxy for output over a months-long window in firms with long contract cycles, so the step from one to the other is unsupported. [unbacked]
What would settle this: A direct output measure — billable hours delivered, tickets closed, units shipped — reported alongside revenue for the same period.
- C1majorO3
Revenue held steady across the pilot as a whole (as distinct from the two verified firms).
Software firms Trial Reader
Could not establish dropout figures; the pilot report does not publish enrolled-versus-completed counts and no follow-up supplies them. [unbacked]
Devil's Advocate
If firms that abandoned the trial did so because it was going badly, the published revenue figures describe survivors and overstate the result by an unknown margin. [unbacked]
What would settle this: Filed statutory accounts for the same firms covering the trial window, compared against a matched set of non-participating firms. Also: The count and characteristics of firms that enrolled and did not complete, with their outcome data.
How much weight the evidence base as a whole can carry.
- low confidence
The evidence base is small and skewed toward firms that volunteered for a trial they publicly supported. [unbacked] Selection into these pilots is not random and no study located corrects for it. [unbacked]
- medium confidence
Corroboration by audited accounts covers two firms out of sixty-one. [unbacked] Treat it as a spot check, not as validation of the sample. [unbacked]
- low confidence
The researcher searched for and did not find any study measuring individual output with an objective pre-registered metric, or any data on firms that adopted and reverted.
Each tied to a specific open question.
- 01
Ask the pilot organisers directly for enrolled-versus-completed counts and the reasons given by any firm that withdrew.
Addresses: The unresolved survivorship objection (O3) on C1.
- 02
Pull filed accounts for every publicly filing participant, not just the two already checked, and compare against the self-reported figures.
Addresses: Whether the corroboration generalises beyond the two spot-checked firms.
- 03
Before publishing any "no productivity loss" line, get a firm to supply a direct output metric for the trial window — otherwise the claim is about revenue, not productivity.
Addresses: The conceded proxy problem in C2.
Comparison
Every cell names the claims behind it. Where nothing was established the cell says so rather than sitting blank — the pattern across a row is what you are about to draw a conclusion from, so a guessed cell would corrupt it invisibly.
2 contested cells4 not established
| Entity | Revenue over the trial | Direct output measure available | Dropout rate published |
|---|---|---|---|
| Software firms | Broadly unchanged C1 | No — revenue used as a proxy The researcher conceded revenue is not a validated proxy for output. C2 | Not established The pilot report does not state enrolment versus completion. |
| Professional services | Broadly unchanged, audited for two firms C1C4 | No Same proxy problem; nothing measures billable output directly. C3 | Not published C3 |
| Manufacturing | Not established No manufacturing participants appear in the pilot record found. | Not established Nothing located; units-shipped data was searched for and not found. | Not established No participants, so no dropout figure to publish. |
The same word, meaning different things
Differences that are definitional rather than substantive. A comparison that flattens these is wrong in a way the table alone cannot show.
- "Output" means billable hours in professional services and units shipped in manufacturing. The pilots that report an output measure are not reporting the same measure.