What Your Office Costs

Situation

The global office stock comprises about 776 million square metres across 66 markets. JLL’s 2024 study “Opportunity through Obsolescence” put 322 to 425 million of those square metres as requiring substantial retrofit. 78 percent of projected obsolescence, meaning the retrofit-relevant office product, sits in the US and Europe, 44 percent in the US and 34 percent in Europe. The same regions account for 83 percent of the necessary capex. The absolute investment need runs between 933 billion and 1.2 trillion US dollars.

In parallel, the EU EPBD recast (Directive (EU) 2024/1275) came into force on 28 May 2024. The transposition deadline into national law ends on 29 May 2026. The recast brings indoor environmental quality (IEQ) into the EPBD framework for the first time. For new buildings, it requires measuring and control devices for indoor environmental quality, where technically and economically feasible. The zero-emission standard for new buildings applies to public buildings from 1 January 2028 and to all new buildings from 1 January 2030.

In this situation, the field measures what is easy to measure. Occupancy rates and engagement scores, counted in hours per desk per week, dominate the conversation. Leesman publishes annual indices on the material, and Gensler and Worktech work adjacent niches with similar measures. Almost no one speaks about the cognitive performance of the people working inside those 776 million square metres, even though it is measurable.

Findings

Four design variables have a measurable effect on the people who work in an office.

Air Quality

Allen ran the COGfx-1 study in 2016 with 24 office workers across three indoor air conditions (Allen et al., 2016, Environmental Health Perspectives). In a lab chamber at the Total Indoor Environmental Quality Lab in Syracuse, CO2 levels and VOC exposure were controlled. On the Green+ day, with doubled ventilation rate and low CO2 (around 550 ppm), SMS decision-making scores averaged across nine cognitive domains were 101 percent higher than under the Conventional condition (around 1,400 ppm CO2). This is a laboratory finding with a small sample, and likely an upper bound.

The multi-country follow-up came in more soberly. Cedeño Laurent measured 302 office workers across six countries prospectively over twelve months in 2021 (Cedeño Laurent et al., 2021, Environmental Research Letters). Per interquartile rise in PM2.5, Stroop interference worsened by 6.18 percent, and per interquartile rise in CO2 by 7.88 percent. Significant effects appeared on eight of ten test metrics. The magnitudes were substantially smaller than in the lab, but they held across countries inside real offices.

MacNaughton put numbers on the economics in 2015 (MacNaughton et al., 2015, IJERPH). Doubling the ASHRAE minimum rate from 20 to 40 cfm per person costs the building owner 7.83 to 39.87 US dollars per person per year depending on climate zone. The modelled productivity gain for the employer is around 6,500 US dollars per person per year. The calculation inherits the effect assumption from CogFX-1 and converts the SMS improvement into a wage differential. It is the most-cited number in the field. It stands with a caveat, because it rests on a modelling assumption about how a lab effect transfers to the real office.

Independent of any manufacturer, Satish isolated CO2 as a variable in 2012 (Satish et al., 2012, Environmental Health Perspectives). She tested 22 students across three CO2 levels at a constant ventilation rate at the Lawrence Berkeley National Laboratory under EPA funding. At 1,000 ppm, six of nine SMS scales worsened significantly, and at 2,500 ppm seven of nine. The study is small, but it separates CO2 from ventilation rate as a variable, which few other studies manage cleanly.

Wargocki and Wyon set out the conservative range in 2017 (Wargocki and Wyon, 2017, Building and Environment). Their synthesis of the lab and field evidence since the 1970s puts performance loss at 5 to 10 percent for adults under conditions today considered acceptable. Labour costs, they write, exceed energy costs by two orders of magnitude.

Acoustics in Open-Plan

In the open-plan office, speech is the dominant disruptor. Banbury and Berry showed in three 1998 lab experiments that office noise with speech disrupts memory for read prose and mental arithmetic, while office noise without speech disrupts only mental arithmetic (Banbury and Berry, 1998, British Journal of Psychology). The semantic content of the speech does not matter. The phonological loop in working memory is disrupted by speech of any meaning.

Haapakangas went into 21 open-plan offices with 883 respondents in 2017 and measured which acoustic variable best explains perceived noise disturbance (Haapakangas et al., 2017, Journal of the Acoustical Society of America). The answer is distraction distance, the range at which speech remains intelligible. Background levels and sound absorption do not act independently. Office noise burden tracks linearly with speech intelligibility over distance.

Pejtersen measured sick days in a representative Danish working-age sample of 2,403 employees in 2011 (Pejtersen et al., 2011, Scandinavian Journal of Work, Environment and Health). People in open-plan offices with more than six seats report a rate ratio of 1.62 for sick days against single offices, adjusted for age, sex, socioeconomic status, BMI, alcohol, smoking and physical activity. Acoustics is not the only mechanism, but one of the most important.

Jahncke measured restoration in a 2011 lab study (Jahncke et al., 2011, Journal of Environmental Psychology). She tested 47 participants with office noise at 51 dB versus 39 dB. The higher noise exposure worsened memory for read text and reduced motivation. In the subsequent break, silence and natural sounds restored concentration and subjective fatigue measurably better than continued office noise.

Daylight and Circadian Regulation

Figueiro measured at five GSA federal office buildings in the US in 2017 (Figueiro et al., 2017, Sleep Health). She tracked 109 workers across seven days of Daysimeter measurement at the eye, once in winter and once in summer. Higher circadian-effective light exposure in the morning was associated with reduced sleep onset latency, particularly in winter, and with stronger phasor magnitude as a measure of circadian alignment. Self-reported sleep quality improved at the same exposures, and higher overall daytime exposure also lowered depressive symptoms.

Boubekri compared 49 office workers in a 2014 pilot study (Boubekri et al., 2014, Journal of Clinical Sleep Medicine). Workers with window access slept 46 minutes longer on workday nights than workers in windowless workstations. The actigraphy subset was 21 people. The authors describe it as a pilot study and the sample is small, but the instruments are validated (PSQI, SF-36) and the measurement happens inside an active office.

Brown laid out quantified thresholds in a broad 2022 consensus recommendation (Brown et al., 2022, PLOS Biology, with Cajochen, Lucas, Czeisler, Lockley, Münch and others). The daytime threshold is at least 250 lx melanopic EDI at the eye. In the three hours before sleep the threshold drops to under 10 lx, and during sleep itself to under 1 lx. The recommendation does not separate daylight from artificial light. Both act biologically through the same melanopic pathway. Architecturally, they are two variables with different levers.

View of Nature

The office-specific evidence for the nature factor is thinner than for the other three. It also sits closer to the plant industry than air, acoustics or daylight does.

Nieuwenhuis conducted three field experiments in 2014 in large offices in the UK and the Netherlands, two of them longitudinal over weeks and months, with more than 300 participants in total (Nieuwenhuis et al., 2014, Journal of Experimental Psychology: Applied). On subjective measures including perceived air quality and self-rated concentration, as well as workplace satisfaction, the greening produced consistent positive effects. On objective measures, the finding is not robust. In the call centre of a Dutch insurance company in Zwolle (Study 2), an objective productivity measure showed no significant effect of the office-design variable. The only significant objective task effect came in a single lab-like session in a London consultancy (Study 3, vigilance task). The +15 percent figure circulating in the industry is a general-discussion statement that arithmetically corresponds to the non-significant information-processing task. It is not used here as a result. The study was funded by the Dutch horticulture marketing board Productschap Tuinbouw and by the company Ki Plants.

Thatcher sharpened the picture in 2020 (Thatcher et al., 2020, Journal of Environmental Psychology). In a laboratory study, the positive plant effect on performance was reproduced. In two call-centre field studies with six and fourteen weeks of exposure respectively, no effect appeared on performance or well-being. The authors conclude that the role of plants in call-centre environments may be overstated. The plants came from Execuflora, disclosed in the acknowledgments. Because the conclusion runs against the sponsor’s economic interest, the study reads as credibly independent.

de Vries worked in nine Dutch organisations in 2023 with plant interventions against control workspaces, taking post-measurements up to four months after introduction (de Vries et al., 2023, Frontiers in Psychology). Significant effects appeared only on proximal variables: fewer complaints about dry air, higher perceived privacy, higher attractiveness of the workspace. The larger two-measurement analysis added greater workspace satisfaction and fewer health complaints. On the distal variables (mood, concentration, stress, mental health, sick leave, productivity, job satisfaction), no direct effect could be shown. The study was funded by the Dutch Topsector Horticulture and Starting Materials.

Aries examined view quality across ten Dutch office buildings with 333 workers in 2010 (Aries, Veitch and Newsham, 2010, Journal of Environmental Psychology). Attractive view quality, including a nature view, reduces physical and psychological discomfort on net. The counter-intuitive detail in the path analysis concerns window proximity without control for heat and glare, not the view itself.

Bringslimark surveyed 385 Norwegian office workers in 2007 and found a weak but statistically reliable association between the number of plants in sight and fewer sick days (Bringslimark et al., 2007, HortScience). The design is cross-sectional, so causation is not established.

Ulrich measured recovery from gallbladder surgery in 1984 (Ulrich, 1984, Science). That study gets pulled into this office discussion a lot, and it has little to do with plants or window views at a desk.

Split Incentive

The split incentive sits above all of this, a term from Allen. The person in the developer’s office who decides on the building specification is not the person who carries the effect at their own desk. The developer bears the investment cost of the specification. The productivity gain from better conditions flows instead to the employer, while the cognitive loss and the worsened sleep land on the employee. The healthcare costs end up at the insurer.

Structurally, this is the same constellation as with tobacco and lead paint. Schroeder examined the economics of tobacco control after the American Master Settlement Agreement in 2004 (Schroeder, 2004, New England Journal of Medicine). Gould calculated in 2009 that every dollar invested in lead-paint hazard control returns between 17 and 221 dollars in socio-economic benefit (Gould, 2009, Environmental Health Perspectives). In both cases the evidence was there long before regulation. Regulation came when the externality was no longer politically or legally tenable.

Position

The office field measures occupancy rates and engagement scores because that is cheap to measure. Cognitive performance per person per year stays outside the model, even though it is quantifiable. The direction of the effect is supported, through manufacturer-independent studies (Satish, Wargocki and Wyon, Pejtersen, Haapakangas, Figueiro) and through mechanism research. The large dollar magnitudes are sponsor-adjacent and stand with a caveat.

Research Context

The air line is the best-evidenced of the four factors, and at the same time the most heavily industry-funded. Four of the central studies (Allen 2016, MacNaughton 2015, MacNaughton 2017, Cedeño Laurent 2021) sit with the same sponsor, United Technologies / Carrier, an HVAC manufacturer. The economic headline number of around 6,500 US dollars in productivity gain comes from this funded material as well. That is not an accusation against the individual authors, but an epistemic situation that runs alongside the finding.

Independent confirmation exists. Satish 2012 is EPA-funded at the Lawrence Berkeley National Laboratory. Wargocki and Wyon work at the Technical University of Denmark. Their 2017 synthesis provides the conservative 5-to-10 percent range. These two studies carry the direction of the effect, but not the +101 percent from the lab or the 6,500 dollars from the model.

A named counter-voice to the main thesis exists. Rodeheffer ran the SMS decision-making tests with 36 submariners at three CO2 levels up to 15,000 ppm in 2018 (Rodeheffer et al., 2018, Aerospace Medicine and Human Performance). He reported no significant differences on any of the nine SMS measures. His verbatim quote is explicit: “unable to replicate this effect in a submariner population, even with acute CO2 exposures more than an order of magnitude greater than those used in previous studies”. The submariner sample is selected. Rodeheffer names this limitation himself. It is also the only explanatory anchor with which Allen’s camp can account for the null in replication.

Scully also found no clear dose-response relationship in 2019 with 22 astronaut-like participants in a double-blind crossover design at four CO2 levels (Scully et al., 2019, npj Microgravity). Du presented a critical review of 37 studies in 2020 and concludes that findings on the cognitive impact of CO2 are inconsistent (Du et al., 2020, Indoor Air). Du names heterogeneity across cognitive tests, designs, samples and exposure measurement as the four main sources of inconsistency.

The direction of the effect is supported by independent mechanism research, by independent real-office studies (Pejtersen, Haapakangas, Figueiro) and by Allen’s lab logic. The exact magnitudes remain contested, and they rest to a large extent on industry-funded research. Anyone building a CFO calculation on this evidence today should work with the conservative Wargocki range, not with the lab maxima.

The other three factors show a similar pattern in attenuated form. For daylight, Figueiro 2017 is funded as GSA contract research by the US federal government, and the Brown 2022 consensus recommendation has a panel with substantial entanglements with the lighting and glazing industries (Lutron, Velux, Signify, View Inc. and others). For acoustics, the data base is largely independently funded. For nature, Nieuwenhuis 2014 is funded by the Dutch horticulture board Productschap Tuinbouw together with Ki Plants, and de Vries 2023 by the Dutch Topsector Horticulture. Thatcher 2020 disclosed plant sponsoring by Execuflora but reaches a conclusion against the sponsor’s interest. Bringslimark and Aries are academically funded.

A second sponsor line comes from the glazing industry. Boubekri and colleagues published a real-office crossover study on electrochromic glass and sleep in 2020 (Boubekri et al., 2020, IJERPH). The study was funded by View Inc., the manufacturer of the electrochromic glass installed in the intervention condition, with View Inc. staff among the co-authors. It is not used here as evidence, but cited as an example of the situation: product manufacturers supply a large share of the published real-office studies on the very design variables their products address.

Consequences

For the employer. The cognitive price of poor office design is quantifiable per design variable and per person. The defensible range runs from 5 to 10 percent performance loss after Wargocki and Wyon (conservative), through the real-office effects in Cedeño Laurent 2021 (in the low single percent per interquartile rise in pollutant or CO2), to MacNaughton’s model calculation of around 6,500 US dollars per person per year on a doubling of ventilation rate. The conservative range belongs in the personnel and location model. The higher magnitudes can be used in rent negotiation, but they have to be marked with the sponsor caveat.

For the developer and landlord. The EPBD recast shifts the frame, with an IEQ measurement requirement for new buildings under technical and economic feasibility, and a zero-emission transition for all new buildings by 2030 at the latest. The added cost of the better specification (7.83 to 39.87 US dollars per person per year for the ventilation doubling per MacNaughton 2015), if it can be passed through as a premium, returns through the rent. Sadikin, Turan and Chegut measured a 4.4 to 7.7 percent effective rent premium for health-certified (Fitwel- or WELL-certified) Class-A buildings in a 2021 SSRN working paper, on CompStak data from ten US cities, in a hedonic model controlling for the usual building and market variables (Sadikin, Turan and Chegut, 2021, SSRN Working Paper 3784779). It is not peer-reviewed but methodically documented.

For asset owners and family offices. The structural difference between certified and non-certified stock grows as the EPBD deadline approaches and more buyers price the specification into acquisition. Anyone modelling a position on the spread today can work with Sadikin/Turan/Chegut for the rent side and with JLL’s obsolescence range for the retrofit probability. Class-B buildings without certification and without a retrofit plan stand under pressure from 2026 onward, and vacancy risk climbs from there.

The magnitudes are to be read with the sponsor caveat from the research context. Independent replication will sharpen them in the years ahead. Until then it pays to work with ranges rather than point estimates, and to treat the independent anchors as defensible.