The Presence Premium: Office Mandates Need a 10% Productivity Miracle to Break Even. The Measured Number Is Zero.

RTO mandates carry an estimated cost of $6.3B a year at Amazon (7.9% of operating income), $2.3B at JPMorgan (4.1% of net income), and $1.0B at Goldman Sachs (5.9% of earnings)

Mostly from attrition, not rent — a break-even model built on FY2025 financials, with Monte Carlo simulation and runnable code

Here is the conclusion up front: a five-day office mandate costs a knowledge-work firm roughly $15,000–$18,000 more per employee per year than a hybrid or remote-first operating model — and most of that cost isn’t rent. It’s attrition. To justify that premium, in-person work would need to make every employee 10–12% more productive, every year, forever. The best randomized controlled trial evidence ever run on this question, published in Nature, measured the actual productivity difference at zero.

That gap — between the productivity edge the office must deliver and the edge it does deliver — is what I call the Presence Premium, and the rest of this article is the receipts: a full break-even model built from published data, 100,000 Monte Carlo scenarios, and runnable code so you can stress-test it with your own firm’s numbers. Then we mark the model to market against three flagship mandate firms’ FY2025 financials, where the modeled drag comes to $6.3B a year at Amazon (7.9% of operating income), $2.3B at JPMorgan (4.1% of net income), and $1.0B at Goldman Sachs (5.9% of earnings).

Why build a model at all? Because every return-to-office debate collapses into the same two unfalsifiable claims. Executives say the office produces collaboration, culture, and serendipity. Employees say they’re just as productive at home. Both sides trade anecdotes, nobody trades models, and the argument resets every quarter.

The break-even framing ends the stalemate by converting it into a falsifiable question: an office-centric operating model carries measurable costs — real estate, elevated attrition, a compensating wage differential, slower hiring. So how large would the office’s productivity advantage need to be, per employee per year, just to pay for itself? If in-person work clears that bar, office-centric operations are rational. If it doesn’t, the mandate is a subsidy paid to a real estate decision. What follows shows it doesn’t come close.

The Cost Stack: What an Operating Model Is Made Of

We compare three operating models for a knowledge-work firm:

Office-centric — five days on-site, dedicated seats. Hybrid — three office days, two remote, shared seating. Remote-first — distributed by default, with home-office stipends, on-demand coworking, and deliberate in-person offsites.

Each model generates costs across six channels. Every parameter below comes from published benchmarks or peer-reviewed research.

1. Real estate and occupancy

Total occupancy cost — rent, operating expenses, utilities, facilities, furniture amortization — runs between $4,000 and $15,000 per employee annually depending on market and work model; a mid-tier market hybrid operation lands around $7,000–$9,000, while full-time office space in Manhattan or San Francisco pushes past $12,000. Density is compressing too: JLL’s 2025 occupancy planning benchmarks show organizations moving from roughly 165 rentable square feet per person toward a target of about 132 — but a five-day model still requires a seat for everyone, while hybrid environments commonly run at 40–50% average occupancy, meaning the effective cost per occupied desk is double the headline figure.

Base case: $10,000/head office-centric, $6,500 hybrid, $3,000 remote-first (stipend + coworking + offsites).

2. Attrition

This is the channel executives systematically underprice, and it’s where the causal evidence is strongest.

The cleanest experiment ever run on this question is Nicholas Bloom’s randomized controlled trial at Trip.com, published in Nature in 2024. 1,612 employees in the IT and airfare divisions were randomized to either five days in office or a hybrid schedule; over the intervention period, quit rates were 33% lower in the hybrid group, with the largest improvements among non-managers, women, and employees with long commutes. Critically, hybrid work had zero effect on productivity or career advancement while dramatically boosting retention.

The mirror image: the University of Pittsburgh study of S&P 500 return-to-office mandates found a 14% increase in departures following mandate implementation — with turnover rising 20% for women, 18% for skilled workers, and nearly 19% for senior managers. RTO doesn’t just raise attrition; it selects for losing exactly the people with outside options. Mid- and top-level managers and high-skilled employees show greater attrition than junior staff, suggesting mandates disproportionately drive away experienced, skilled employees with bargaining power.

Replacement isn’t cheap: SHRM estimates replacing an employee costs six to nine months of their salary. We use 75% of salary as the base case, 50–125% as the range.

3. The compensating wage differential

Flexibility is compensation. Workers price it that way even when firms don’t. Survey evidence puts the average valuation of remote flexibility around 5–8% of pay, and in technical labor markets it’s dramatic: tech workers report they’d trade up to 25% of total compensation to avoid a five-day commute — $60,000 on a $240K senior engineering package — and roughly 71% of workers say they’d accept a straight pay cut for fully remote work.

The mechanism is arithmetic, not preference. LendingTree estimates the average American commute (26.8 minutes each way) translates to $37.73 a day in lost time — $9,470 a year, and a typical in-office day now runs roughly $15 for the commute, $9 for parking, $13 for breakfast or coffee, and $18 for lunch. A five-day mandate transfers those costs from the firm’s P&L to the employee’s household budget — and the labor market claws it back through wages, hiring difficulty, or quits. We model the office-centric wage premium conservatively at 5% of salary (range 2–10%); hybrid at 1%.

4. Vacancy drag

Empty seats have a run-rate. The Baylor analysis of RTO firms found job vacancy duration increases by 23%, from 51 to 63 days on average, and hire rates decline 17% even after adjusting for national trends. In our model, vacancy drag = quit rate × extra vacancy days × daily value of a fully loaded employee.

5. Distributed-operations investment

Honesty requires charging remote-first for what it takes to run well: documentation culture, async tooling, deliberate offsites, structured mentorship. We budget $2,500/head/year for remote-first and $1,000 for hybrid. This is the line item most remote advocates conveniently omit, so we won’t.

6. Productivity differential

The parameter everyone fights about — so we let the evidence set it. For hybrid, the Nature RCT is unambiguous: employees randomized into three office days and two remote days showed no measurable productivity difference from fully in-office peers. We center hybrid at 0% (±2%). For fully remote, the literature genuinely ranges: the original Ctrip experiment found a 13% performance increase from home working, while poorly managed full-remote studies find declines. We center remote at 0% with a wide ±8% band — the uncertainty is the point, and the Monte Carlo carries it.

The Model

Fully loaded cost per knowledge worker: $120,000 salary × 1.30 load factor = $156,000. Baseline voluntary quit rate: 13%. Firm size: 500. All uncertain parameters get triangular distributions over their published ranges.

import numpy as np
import pandas as pd

rng = np.random.default_rng(42)

SALARY = 120_000
FULLY_LOADED = SALARY * 1.30

PARAMS = {
    # Real estate per head (base, low, high) — JLL/Gable occupancy benchmarks
    "office": {"re_per_head": (10_000, 6_500, 14_000)},
    "hybrid": {"re_per_head": (6_500, 4_500, 9_000)},
    "remote": {"re_per_head": (3_000, 1_800, 4_500)},

    "base_quit": (0.13, 0.10, 0.16),

    # Attrition multipliers: Ma et al. (RTO +14%), Bloom et al. (hybrid -33%)
    "office_quit_mult": (1.14, 1.05, 1.25),
    "hybrid_quit_mult": (0.67, 0.60, 0.80),
    "remote_quit_mult": (0.72, 0.60, 0.90),

    # SHRM: replacement costs 6-9 months of salary
    "replace_cost_frac": (0.75, 0.50, 1.25),

    # Compensating differential for mandated presence
    "onsite_comp_premium": (0.05, 0.02, 0.10),
    "hybrid_comp_premium": (0.01, 0.00, 0.03),

    # Baylor: on-site vacancies 51 -> 63 days
    "vacancy_days": {"office": (63, 55, 75),
                     "hybrid": (54, 48, 60),
                     "remote": (51, 45, 58)},

    # Productivity deltas vs office baseline
    "hybrid_prod_delta": (0.00, -0.02, 0.02),
    "remote_prod_delta": (0.00, -0.08, 0.08),

    # Cost of running distributed work well
    "remote_coord_cost": (2_500, 1_500, 4_500),
    "hybrid_coord_cost": (1_000, 500, 2_000),
}

N = 100_000

def tri(key):
    base, lo, hi = PARAMS[key]
    return rng.triangular(lo, base, hi, N)

base_quit = tri("base_quit")
replace_frac = tri("replace_cost_frac")

sims = {}
for m in ["office", "hybrid", "remote"]:
    b, lo, hi = PARAMS[m]["re_per_head"]
    re_cost = rng.triangular(lo, b, hi, N)
    quit = base_quit * tri(f"{m}_quit_mult")
    attrition = quit * replace_frac * SALARY
    comp = tri("onsite_comp_premium") * SALARY if m == "office" \
        else tri("hybrid_comp_premium") * SALARY if m == "hybrid" \
        else np.zeros(N)
    vb, vlo, vhi = PARAMS["vacancy_days"][m]
    vacancy = quit * (rng.triangular(vlo, vb, vhi, N) / 260) * FULLY_LOADED
    coord = tri("remote_coord_cost") if m == "remote" \
        else tri("hybrid_coord_cost") if m == "hybrid" else np.zeros(N)
    prod = -tri("remote_prod_delta") * FULLY_LOADED if m == "remote" \
        else -tri("hybrid_prod_delta") * FULLY_LOADED if m == "hybrid" \
        else np.zeros(N)
    sims[m] = re_cost + attrition + comp + vacancy + coord + prod

mc = pd.DataFrame(sims)
gap_hybrid = mc["office"] - mc["hybrid"]
gap_remote = mc["office"] - mc["remote"]

print(mc.describe(percentiles=[0.05, 0.5, 0.95]).round(0))
print(f"Presence Premium vs hybrid: median "
      f"{np.median(gap_hybrid)/FULLY_LOADED:+.1%}")
print(f"Presence Premium vs remote: median "
      f"{np.median(gap_remote)/FULLY_LOADED:+.1%}")

Results

The deterministic base case

Annual model-attributable cost per employee:

Notice what the table does to the conventional framing. Real estate — the line item RTO decisions obsess over — is not the biggest cost of the office-centric model. Attrition is. The single largest expense of mandating presence is replacing the people who leave because you mandated presence. The second-largest is the wage premium the labor market charges for the mandate. The building is third.

For a 500-person firm, the deterministic gap is $7.8M per year vs. hybrid and $9.1M vs. remote-first. Over a five-year planning horizon, that’s $39–45M — roughly the cost of an entire product line, spent to have people badge in.

The Monte Carlo

Point estimates can be gamed, so we ran 100,000 scenarios drawing every parameter from its published range simultaneously.

The office and hybrid distributions do not meaningfully overlap. Even the luckiest office-centric scenario (5th percentile, cheap market, low attrition) costs more than the median hybrid scenario. Remote-first has the widest distribution — its productivity uncertainty is real — but its 95th-percentile worst case still beats the office median.

The Presence Premium

Now the payoff. Dividing the cost gap by fully loaded compensation gives the productivity premium the office must deliver, per employee, to break even:

vs. hybrid: median +10.4% (90% interval: +6.8% to +14.3%)
vs. remote-first: median +12.2% (90% interval: +5.9% to +18.6%)

That is the claim every RTO mandate is implicitly making: our people are at least 10% more productive in the building. Not “collaboration feels better.” Not “culture matters.” Ten-plus percent of measurable annual output, every year, for every employee, forever.

Against that requirement, the measured evidence: the strongest RCT we have — graduate engineers, marketing, and finance professionals doing creative team work, randomized, tracked for two years, published in Nature — found the productivity difference between hybrid and full-time office attendance was zero. Managers predicted remote work would hurt productivity, then changed their minds by the time the experiment ended.

The sensitivity curve makes the burden of proof explicit:

Even granting the office a +8% productivity edge — four times anything the hybrid RCT literature can detect — office-centric operations win in about one scenario in seven. The office doesn’t need to be somewhat better to be rational. It needs to be miraculously better.

Marking the Model to Market: JPMorgan, Amazon, Goldman Sachs

A stylized 500-person firm is a clean laboratory, but the mandate wars are being fought by named companies with published P&Ls. So let’s mark the model to market. We take three flagship five-day mandate firms, estimate their mandate-exposed headcount — the knowledge workers actually moved from hybrid to full-time presence, not the branch tellers and warehouse associates who were never remote — and run the same Monte Carlo gap through their reported FY2025 financials.

The setup for each firm:

JPMorgan Chase announced its five-day mandate in January 2025, covering its 300,000+ global workforce; reporting at the time indicated roughly 60% of staff (branches, trading floors) were already in five days, so the mandate primarily moved the hybrid back-office and technology population. FY2025 results: $182.4B net revenue, $57.0B net income, ~318K employees. We model 90K–160K mandate-exposed employees (base 125K) at an average $140K salary.

Amazon implemented five-day RTO for its corporate workforce in January 2025 — roughly 350K corporate employees of its 1.58M total. FY2025: $716.9B net sales, $80.0B operating income. We model 250K–350K exposed (base 300K) at $160K average.

Goldman Sachs never left: five days since 2021, essentially the entire ~46K workforce. FY2025: $58.3B net revenues, $17.2B net earnings. We model 34K–44K exposed at a $200K average salary — conservative for Goldman.

For each firm we draw exposed headcount, salary, and every Part I parameter simultaneously across 100,000 scenarios, computing the annual cost of running five-day instead of hybrid.

Read those numbers the way an activist investor would.

Amazon’s five-day mandate carries a median run-rate of $6.3 billion a year — 7.9% of the operating income of one of the most operationally disciplined companies on Earth, a company that will restructure an entire fulfillment network to save basis points on shipping. Over five years, the median cumulative drag is $31.6B: more than Amazon’s entire 2023 operating income. And the model may be conservative for Amazon specifically, because the observed reaction was extreme: a Strategic Organizing Center survey found 48% of mandate-impacted employees had applied to other jobs and 68% said they were likely to leave within a year, while a Glassdoor poll showed roughly three-quarters of Amazon professionals reconsidering their future at the company. Our stress case: if even an extra 5% of exposed employees quit in year one beyond the model’s baseline uplift, that’s 15,000 departures and another $2.6B in replacement and vacancy costs — an extra 3.2 points of operating income. At 10%, it’s $5.1B and 6.4 points.

JPMorgan’s mandate costs a median $2.3B annually — about 4.1% of net income. For context, that approaches the run-rate of entire business investments the bank trumpets in its shareholder letter. The bank’s own filings note rising occupancy expense; the model says occupancy is the smaller half of the bill. The larger half is churning the back-office and technology talent the mandate binds — the same technologists JPMorgan competes for against remote-flexible fintechs.

Goldman is the most instructive case because it’s had five-day the longest and pays the most. At Goldman salaries, the per-employee gap rises to ~$25,700 because attrition, vacancy drag, and the compensating differential all scale with pay. The mandate costs a median $1.0B a year against $17.2B in earnings — call it a permanent ~6% earnings handicap the firm has chosen to carry since 2021. Goldman can afford it; its franchise economics are extraordinary. But “we can afford it” is not the same as “it pays.”

The implied claim, restated in each CEO’s language

Divide each firm’s gap by fully loaded compensation and the model returns essentially the same number for all three: the mandate breaks even only if presence makes each exposed employee ~10% more productive (JPM 10.2%, Amazon 10.0%, Goldman 9.8% — the consistency is because most cost channels scale with salary).

For Amazon, 10% of 300K corporate employees is the output of 30,000 people — the mandate must conjure a phantom workforce the size of Goldman’s entire firm, every year, or it’s destroying value. No RCT in the literature has ever detected an effect close to that; the Nature trial measured it at zero, and Amazon’s own workforce predicted the opposite direction — 87% of surveyed employees expected their productivity to decline under the policy.

The natural experiment on the other side

The contrast firms are running the counterfactual in public. Atlassian — roughly 12,000 employees under its Team Anywhere policy — reports that its workforce tripled under distributed work, job applications more than doubled, more than half of recent hires live over two hours from an office, and 92% of employees say the model lets them do their best work. Airbnb, whose ~7,000 employees can live and work anywhere, reported revenue up 40% following the policy with departures dropping below 10%. These are self-reported and selection-laden numbers, not RCTs — treat them accordingly. But note the asymmetry: the flexible firms publish retention and hiring metrics; the mandate firms publish convictions. One side is showing its data.

None of this means JPMorgan or Amazon will visibly stumble. That’s precisely what makes the Presence Premium politically survivable: a 4–8% profit drag at a franchise this dominant disappears into the noise of a quarterly print. The cost is real, but it’s paid in counterfactual dollars — the technologist who took the fintech offer, the vacancy that ran twelve extra days, the wage premium buried in comp expense. No line item on the income statement says “mandate.” That’s not evidence the cost is small. It’s evidence the accounting system wasn’t designed to catch it.

Steelmanning the Office

A model is only as honest as its treatment of the other side, and there is one genuinely strong piece of evidence for presence.

Emanuel, Harrington, and Pallais studied engineers at a Fortune 500 firm and found that engineers working in the same building as teammates received 23% more online feedback on their code than engineers with distant teammates — an advantage that collapsed when offices closed. The effect concentrates where it matters most: new hires and young engineers receive more feedback when co-located and lose more mentorship when remote, and female engineers are more affected by proximity. Proximity, in their framing, trades productivity today for human capital tomorrow — sitting near coworkers leads to fewer early pay raises but larger raises in the long run.

This is a real cost of distribution and the model’s coordination line item exists because of it. But read carefully, it’s an argument for engineered proximity — onboarding cohorts, structured mentorship, deliberate co-location for juniors — not for 260 badge-ins a year across the entire workforce. The mentoring channel affects a subset of employees during a subset of their tenure. The Presence Premium is charged to everyone, forever. Paying a firm-wide 10% tax to deliver a targeted benefit that could be purchased with offsites and apprenticeship design is not a strategy; it’s a rounding error in the wrong direction.

There’s also a selection asymmetry the steelman can’t escape: the people RTO mandates actually remove are not the juniors who need mentorship — they’re the senior managers (+19% turnover) and skilled workers (+18%) who provide it. A policy justified by mentorship that preferentially expels mentors is self-defeating on its own terms.

What the Model Doesn’t Say

Three honest limitations — four, now that named firms are involved. First, this is a knowledge-work model; none of it applies to labs, manufacturing, healthcare, or client-facing retail. Second, the remote-first productivity band (±8%) is wide because the causal evidence on fully remote creative teamwork is genuinely thinner than on hybrid — the Nature RCT tested 3/2, not zero days. A firm choosing remote-first is accepting more variance in exchange for a lower expected cost, and the Monte Carlo shows exactly how much. Third, the model prices operating models, not transitions; unwinding leases and rebuilding management practice around outcomes both cost real money on the way to either destination. Fourth, the firm-level figures are model estimates applied to public financials, not audited disclosures — mandate-exposed headcounts and average salaries are documented assumptions with explicit ranges, and none of these companies reports “cost of the mandate” as a line item. That’s rather the point: nobody’s accounting system does. The 90% intervals in the table are the honest width of what public data supports, and the code ships so anyone — including the firms themselves — can rerun it with better internal numbers.

But the burden of proof has been running in the wrong direction for four years. Executives demand that remote work prove itself against an office baseline assumed to be free. The office is not free. It is the single most expensive operating-model choice on the menu, and its costs compound through the two channels — attrition and compensating differentials — that never appear on the real estate budget.

The Bottom Line

Run the break-even math before the mandate, not after:

The office’s biggest cost isn’t the building — it’s the churn. Attrition and vacancy drag exceed real estate in every scenario we simulated.
The Presence Premium is ~10%. That’s the productivity miracle a five-day mandate implicitly promises shareholders — and the model returns essentially the same required premium whether the workforce costs $120K (our stylized firm), $140K (JPMorgan’s back office), or $200K (Goldman).
Marked to market, the mandates are billion-dollar policies: a modeled $6.3B/year at Amazon (7.9% of operating income), $2.3B at JPMorgan (4.1% of net income), $1.0B at Goldman (5.9% of earnings) — paid in counterfactual dollars that no income statement line ever names.
The measured premium, in the best causal evidence available, is zero — with a 33% retention dividend going to the flexible arm, and the flexible-model firms (Atlassian, Airbnb) publishing hiring and retention data while the mandate firms publish convictions.
Presence has real value — for juniors, in mentorship — and it’s purchasable in targeted doses for a fraction of the firm-wide tax.

The organizations that get this right won’t be the ones that “won” the RTO standoff. They’ll be the ones that priced presence like any other input: bought where it pays, declined where it doesn’t, and never confused the comfort of a full parking lot with a return on capital.

The full model — Part I break-even simulation and Part II firm-level analysis — ships as two Python scripts with every parameter documented inline with its source, so you can rerun it with your own firm’s numbers.

References

Bloom, N., Han, R., & Liang, J. (2024). Hybrid working from home improves retention without damaging performance. Nature. (RCT of 1,612 Trip.com employees; NBER Working Paper 30292.)
Bloom, N., Liang, J., Roberts, J., & Ying, Z. J. (2015). Does Working from Home Work? Evidence from a Chinese Experiment. Quarterly Journal of Economics, 130(1).
Ma, M., et al. (2024). Return-to-Office Mandates. University of Pittsburgh, Katz Graduate School of Business. (S&P 500 turnover analysis.)
Van Dijcke, D., Gunsilius, F., et al. (2025). RTO mandates, employer attractiveness, and brain drain. Baylor University / University of Michigan analysis of vacancy duration and hire rates.
Emanuel, N., Harrington, E., & Pallais, A. (2023). The Power of Proximity to Coworkers. NBER Working Paper 31880.
JLL (2025). Occupancy Planning Benchmarking Report. (Density trends: ~165 → ~132 sq ft/person.)
Gable (2026). Workplace Spend Benchmarks. (Total occupancy cost: $4,000–$15,000/employee/year.)
SHRM. Employee replacement cost estimates (6–9 months of salary).
LendingTree (2025). The Cost of Commuting analysis of U.S. Census ACS data.
Barrero, J. M., Bloom, N., & Davis, S. J. (2021–2023). WFH Research / Survey of Working Arrangements and Attitudes. (Commute time savings; valuation of remote flexibility.)
Aksoy, C. G., et al. (2023). Commute time savings when working from home. CEPR/VoxEU. (~72 minutes saved per WFH day across 27 countries.)
JPMorgan Chase & Co. (2026). Form 10-K FY2025 and 4Q25 earnings release. (Net revenue $182.4B; net income $57.0B; ~318K employees.)
Amazon.com, Inc. (2026). FY2025 Annual Report / consolidated statements of operations. (Net sales $716.9B; operating income $80.0B; 1,576,000 employees; ~350K corporate.)
The Goldman Sachs Group, Inc. (2026). FY2025 full-year earnings release. (Net revenues $58.28B; net earnings $17.2B.)
Bloomberg (2025). JPMorgan five-day return-to-office reporting. (~60% of staff already on-site five days pre-mandate.)
Strategic Organizing Center (2024). Survey of Amazon employees impacted by the 2025 RTO mandate. (48% applied to other jobs; 68% likely to leave within a year; 87% expected productivity to decline.)
Fortune (2025). Atlassian Team Anywhere profile. (~12,000 employees; workforce tripled under distributed work; applications more than doubled; 92% report doing their best work.)
Airbnb, Inc. — Live and Work Anywhere policy reporting (Fortune, 2023). (Revenue up 40% post-policy; departures below 10%.)

About the Author

Brian James Curry is a Kansas City–based data scientist and applied AI researcher working at the intersection of economic modeling, causal inference, and machine learning — with a standing preference for models where the economics is load-bearing. He is the founder of Vector1 Research, an independent lab exploring marketing economics, cognitive AI architecture, and AI-driven enterprise transformation, with research contributions including Papilon, PyCausalSim, and Memory-Node Encapsulation. His twenty-plus years of enterprise work span Koch Industries, Tractor Supply, Vail Resorts, Hallmark, Garmin, and AT&T.