Fee Compression Is Permanent: The Operating Leverage Case for Institutional Real Estate Managers

A fund’s draft terms reach an allocator’s consultant. She does not read the management fee clause first. She opens a workbook in which every manager she covers is a row, the fee is a column and the column is sorted. The new fund lands in the row the data says it should, and the negotiation, if there is one, concerns whether the fee steps down to invested capital after the investment period and on which date. The level itself is barely discussed, because the table has settled it.

The table exists because several bodies now publish it. Cambridge Associates, analysing the funds it underwrote between 2018 and 2022, reports the investment-period management fee for real estate at 1.5% at the first quartile, 1.5% at the median and 1.5% at the third quartile in its 2022 sample. An interquartile range of zero is an unusual thing to find in any market for a service, and it says something plain: the middle half of the market charges the same price. Preqin, working from its own fund-terms dataset, reported in October 2024 that the mean real estate management fee stood at 1.31%, a twenty-year low, in an asset class that has historically carried the lowest fees in private capital. The two measurements rest on different samples and different bases and should not be read as one figure, but they point the same way.

The vehicle-level measure has converged too. INREV’s fees and terms study for 2023, released in September 2024, put the Total Global Expense Ratio at 0.94% of gross asset value and 1.72% of net asset value across 91 European non-listed funds, and observed that management fees were more consistent across vehicles than the other costs a vehicle carries. In July 2025 ANREV, INREV and NCREIF published a single Total Global Expense Ratio for the Global ODCE index: 0.82% of gross asset value, unchanged on the year, and 1.21% of net asset value, across 47 of the index’s 49 funds and USD 336 billion of gross assets. INREV’s 2026 study, covering 2025, reports core funds outside the ODCE set at 0.72% of GAV, ODCE funds at 0.91% and multi-country, multi-sector vehicles at 1.11%.

What these series share matters more than their levels. Each is published by a body the allocator trusts, each is refreshed annually, and each lets a consultant tell a manager that its fee sits so many basis points above or below a named peer group. That is the condition under which a term stops being negotiated and starts being benchmarked. A benchmarked number does not move with the manager’s cost of doing the work, because the benchmark contains no information about that cost. The expense ratio measures the level of what a vehicle charges; it says nothing about the accuracy of any figure the manager produced or the hours that went into producing it. No published series does.

The Smart Bricks Research Lab has been working through what follows from that asymmetry, drawing on the fee studies above, the industry’s own productivity survey and the public record of how allocators test managers. The argument is short. A business cannot hold a margin when its price is standardised and its cost is not, so the variable a manager controls is the cost of producing a verified analytical output, and that variable is currently being moved by workload rather than by design.

• 5.1% vs 5.4%

  Annual growth in asset management revenues versus costs, 2010 to 2025, with margins flat near 30%

  BCG, Global Asset Management Report 2026

• 1.31%

  Mean management fee for private real estate funds, a twenty-year low

  Preqin, 2024 Private Capital Fund Terms Advisor

• 0.82% of GAV

  Total Global Expense Ratio across 47 Global ODCE funds holding USD 336 billion, end-2024

  ANREV, INREV and NCREIF, July 2025

• 42%

  Salary and benefit expenses as a share of revenue at the median real estate investment manager

  NAREIM and Ferguson Partners, 2026 Global Management Survey

• 59% vs 46%

  Share of managers reporting higher expenses versus share reporting higher revenue in 2025, with median headcount flat

  NAREIM and Ferguson Partners, 2026

• Just over half

  Share of 2024 committed capital going to the 100 largest funds, up from more than a third in 2021

  Preqin, State of Private Capital Fundraising in 2025

Revenue per unit of AUM is falling against a cost base that is mostly people. The figures come from the industry's own benchmarking bodies and surveys; none is a vendor estimate.

Source: BCG (April 2026); Preqin (October 2024, 2025); ANREV, INREV and NCREIF Global ODCE Management Fees Study 2025; NAREIM and Ferguson Partners 2026 Global Management Survey

The cost side has a survey of its own. NAREIM and Ferguson Partners have polled real estate investment managers on their operations for eighteen years; the 2026 edition covers 79 firms and reports on 2025. Its released figures describe a year in which the median firm’s book grew and its people did not.

Median net assets under management grew 5%. Median headcount was flat: 48% of firms added staff, 11% held steady and 41% reduced. Salary and benefit expenses ran at 42% of revenue at the median firm. More firms reported higher expenses (59%) than reported higher revenue (46%), and a majority reported lower profitability for the year. The survey states the profitability finding as a majority rather than a percentage, and it is reported here on that basis.

The load figures are the ones to sit with. The median asset manager was responsible for 16 properties in 2025, against 13 in 2024, and oversaw USD 40 million of net operating income, against USD 28 million a year earlier. Our own arithmetic on those medians gives 23% more properties and 43% more income per head in twelve months. Nothing in the two organisations’ releases describes a change in how the work of asset management was done over that period. They describe a change in how much of it each person did.

Read together, the figures say something the fee studies cannot. Revenue is a benchmarked percentage of a book that grew 5%. Cost is 42% salaries at a median firm that did not grow its headcount. Profitability still fell for most respondents, because the fee compressed faster than the load could be spread and because the other costs a firm carries kept rising. The survey also records where firms say they will go next: 67% expect to add headcount in 2026, and the share naming underwriting and modelling as a priority area for AI rose from 8% to 29% in a year. Both responses answer the same pressure. One repeats the pattern the survey describes; the other has not yet been measured by anyone.

There is a limit to what a survey of 79 firms can carry, and it should be stated. The respondents are members of one trade body and clients of one search firm, the full dataset is available to members only, and the figures used here are those the two organisations released publicly. What the survey does establish, on its own terms, is the direction of the load, and the fact that it moved without a corresponding move in people or in profit.

Signal	Stretch	Leverage	What a stranger can check
Properties per asset manager rising	StretchSame people, more assets, longer hours	LeverageSame people, more assets, fewer hours per asset	What a stranger can checkHours per recurring output over four quarters, not assets per head
Headcount flat while expenses rise	StretchContractors, overtime and rework absorbing the load	LeverageProduction cost per output falling as volume rises	What a stranger can checkCost per LP pack, per fee invoice, per memo, quarter on quarter
Reporting clock met	StretchMet in the final week, from re-keyed PDFs	LeverageMet as a slice of an always-reconciled record	What a stranger can checkDays from quarter-end to a pack every figure of which traces to source
Memo numbers cited	StretchCitations composed alongside the figure	LeverageCitations recorded as a by-product of the arithmetic	What a stranger can checkChange one source input, re-run, see whether the citation moves
Exception handling	StretchDiscrepancies averaged or fixed by whoever notices	LeverageA named owner clears a visible queue against a written tolerance	What a stranger can checkThe tolerance document and the age of the oldest open exception

Both raise output per head. Only one changes the slope of the cost line, and only one survives an LP validator. The last column is what a committee or an allocator can check without trusting the manager's description.

Source: Smart Bricks Research Lab, drawing on NAREIM and Ferguson Partners 2026 and ILPA Quarterly Reporting Standards v1.1

Operating leverage, in the plain accounting sense, is the rate at which profit changes when revenue changes, and a firm has more of it when its costs are fixed rather than variable. A management business whose cost is people has very little, because each increment of book has historically required an increment of staff. The NAREIM figures show firms acquiring leverage anyway, by declining the increment of staff and letting the ratio of assets to people rise. Call that loading.

Loading works, up to a point, because most of what an asset manager does in a quarter sits well inside anyone’s judgment. Rent rolls are refreshed, budgets reforecast, lender reports assembled, valuations circulated, the model rolled forward a period. A capable person can carry sixteen of those cycles rather than thirteen. What gives way under load is the set of steps around the tasks that make a number defensible afterwards: the reconciliation of the rent roll to the executed lease, the as-at date recorded beside the figure rather than assumed, the note of what was adjusted between the source and the report. The earlier note on the IC memo standard set out the test a cited number has to meet, and that test is exactly the set of steps a person under load stops doing first, because none of them changes the number and all of them take time.

Loading, then, is leverage that consumes traceability. It appears in a productivity survey as a higher ratio, and it appears in the firm’s own records as nothing at all until a deadline is missed or a figure is restated. It also collides with clocks the manager does not control. ILPA’s quarterly reporting standard expects quarterly financials within 60 days of quarter end and targets 45, with audited annual accounts allowed a further 30. A manager loading its asset managers cannot move that clock, so the time comes out of the checking.

The second kind of leverage changes the process rather than the ratio. Re-engineering, for want of a better word, lowers the cost of producing a number that meets the standard: the source, location, date, path and definition are captured while the figure is produced rather than assembled afterwards, so that carrying sixteen assets costs less in checking per asset than carrying thirteen did, rather than more. Its unit of measure is the cost of one traceable number rather than assets per head, and the rest of this paper measures against that unit.

The distinction matters to a GP because the two kinds of leverage look identical in a fee negotiation and entirely different in diligence. An allocator’s consultant can read the ratio of assets to staff off a questionnaire. She cannot yet see whether the ratio was reached by skipping the reconciliation or by making the reconciliation cheap. The instruments that will let her see that are the subject of Section V.

Process	External clock or benchmark	Who set it	Where leverage comes from
LP quarterly reporting	External clock or benchmark60 days expected, 45 targeted; v2.0 template from Q1 2026	Who set itILPA	Where leverage comes fromPack assembled from a reconciled record, reviewed rather than built
Fee calculation and billing	External clock or benchmarkFee and expense controls failed in over 50% of early SEC exams; two-thirds of LPs run no carry validation	Who set itSEC (2014); bfinance (2025)	Where leverage comes fromInvoice computed from extracted LPA clauses and reconciled before issue
Expense classification and TGER	External clock or benchmark0.82% of GAV across Global ODCE; standard definitions since 2018 consultation	Who set itANREV, INREV, NCREIF	Where leverage comes fromClassification at posting against the standard, not a year-end rebuild
DDQ response	External clock or benchmark20 sections plus appendices, including a schedule of every technology tool	Who set itILPA DDQ 2.0 (2021)	Where leverage comes fromAnswers drawn from an indexed document library and reviewed, not redrafted
IC memo	External clock or benchmark97% of institutions run a formal process; about two months to write and review	Who set itAddepar and Stanford (2022)	Where leverage comes fromEvery material number traced outside the memo at the moment it is produced

Each process has a published clock or benchmark set by someone other than the manager, and each scales with assets and investors rather than with judgment. Leverage is won or lost one row at a time.

Source: ILPA Quarterly Reporting Standards v1.1 and Reporting Template v2.0 guidance; SEC, Spreading Sunshine in Private Equity (2014); bfinance 2025 snap poll; ANREV, INREV and NCREIF 2025; ILPA DDQ 2.0; Addepar and Stanford Long-Term Investing (2022)

A manager’s cost base divides into work that scales with the number of assets and work that scales with the number of decisions. The second is small and expensive: deciding to buy, to sell, to refinance, to hold through a covenant breach. The first is large and repetitive, and it is where the load in the NAREIM figures has landed. Fund administration and the calculation of fees and waterfalls against the LPA; the assembly of the quarterly LP pack; responses to due diligence questionnaires; the review and abstraction of leases and their amendments; the periodic refresh of the underwriting model against actuals. Each repeats per asset, per fund or per quarter, and each produces figures that an allocator, an auditor or an examiner may later ask to have retraced.

Standardisation on the allocator side has already made some of this repeatable. ILPA’s Reporting Template, by the association’s own estimate, is used by roughly half the funds in the industry, and its January 2025 revision was the first since 2016. A template used by half the market turns LP-pack assembly from a bespoke document into a fixed mapping from the ledger to a known set of lines. The Global ODCE expense study does the same for the expense ratio, which every constituent fund now computes the same way. Where the output is standardised, its production can be engineered rather than staffed, and the engineering pays back every subsequent quarter.

The economics of that engineering are being set by people outside real estate. Bessemer Venture Partners, in its published playbook for vertical AI businesses, frames the opportunity as competition for labour budgets rather than IT budgets, and sizes it accordingly: business and professional services at 13% of US GDP, roughly ten times the software market. Across its own vertical AI portfolio the firm reports gross margins of roughly 56% on an average burn ratio of 1.6x, and it describes an emerging pricing pattern that pairs a base subscription with charges tied to output. Those are one investor’s numbers about its own companies and are used here for direction only. The direction is clear enough. Software sold into a manager’s process will be priced against the analyst-hours it displaces, at margins that leave the vendor most of the saving unless the manager can show, from its own records, what an hour of its analysis produces.

That is the second reason the unit of measure has to be cost per traceable number rather than assets per head. A vendor pricing against labour saved needs the manager’s throughput data to set a price, and a manager without that data negotiates blind. ILPA’s 2021 fee-terms survey shows who has been paying for the tools so far: a quarter of the LPs it polled reported administrative expenses, including computer software, being shifted outside the management fee and charged to the fund. An allocator reading that line beside the manager’s fee will ask what the software produced, and a headcount ratio is no answer.

A fee benchmarked to the basis point cannot be defended by a cost base that is measured in headcount.

Smart Bricks Research Lab

Allocator sophistication about fees arrived through examination before it arrived through negotiation. In May 2014 the director of the SEC’s examinations office reported that in more than 150 examinations of private equity advisers, staff had found violations of law or material weaknesses in controls over fees and expenses more than half the time. Those were private fund advisers generally rather than real estate managers, and the finding should be read at that width. The Division’s January 2022 risk alert, by which time the agency counted more than 5,000 advisers managing about USD 18 trillion in private fund assets, up 70% in five years, named the recurring defects with some precision: management fee bases left unreduced after an asset was sold or written down; LPA terms such as “permanently impaired” left undefined and then applied inconsistently; recycling provisions that produced management fees in excess of what the agreement allowed. Each of those is a calculation an LP could have reproduced from the LPA and the fund’s statements, and each had gone unreproduced for years.

The templates followed the examinations. ILPA’s 2021 fee-terms survey found that 59% of LPs received the association’s fee template at least half the time, that headline management fee rates had held at 1.5% to 2.0% while covering a smaller share of total expenses, and that a quarter of respondents were seeing administrative costs move outside the fee. The Reporting Template’s 2025 revision and the Global ODCE expense study are the same instrument at greater scale: a fixed line structure that lets an allocator recompute what the manager charged, line by line, against what the manager said it would charge.

What allocators worry about and what they check remain two different things. bfinance’s 2025 snap poll of private markets investors, whose sample size the firm does not state, found that 90% ranked hidden or opaque costs among their two largest concerns, while nearly two-thirds carried out no validation at all of the carried interest and waterfall calculations their managers presented, only half checked waterfall arithmetic in periodic reviews, and 42% scrutinised compliance with the LPA. The gap between the concern and the check is wide, and it is closing from the check side, because every template, validator and published deficiency list lowers the cost of checking.

Where that closing leads is the point. Every instrument named above tests the fee: its basis, its offsets, its arithmetic against the agreement. None of them yet tests the process the fee pays for, which is the production of the analysis behind the assets on which the fee is charged. But the allocators who built fee validators out of the SEC’s deficiency list will not stop at the fee line. The due diligence questionnaire already asks how many people a manager employs per billion of assets and which tools they use. Its next revision will ask what those tools produced, how often the output was restated and how long a memo took to reach committee. The manager who can answer with a record rather than a narrative is the one who clears that diligence, and that record is the manager-selection variable this paper has been circling.

1. 1

   Output per head

   How many recurring analytical outputs (packs, invoices, memos, reconciliations) did each investment or finance professional produce last quarter, and how has that moved over four quarters?

2. 2

   Hours per output

   For each output, are hours per unit falling as assets rise, or are assets per head rising while hours per head rise with them?

3. 3

   Traceability share

   What share of material numbers in the last IC memo and LP pack trace to a source outside the document, and who checked?

4. 4

   Clock margin

   How many days before the ILPA 60-day expectation was the last pack delivered fully reconciled, and did any figure change after delivery?

5. 5

   Tolerance and owner

   Is there a written tolerance for each reconciliation, and who is named to clear the exception queue each cycle?

6. 6

   Recalculation

   Could the firm reproduce its last fee invoice and distribution from the LPA's own clauses, on demand, before an LP validator does?

Three an LP can add to a DDQ; three a GP should be able to answer before the LP asks. All six are ratios the firm already holds the numerator and denominator for.

Source: Smart Bricks Research Lab, drawing on ILPA DDQ 2.0 sections 1, 12 and 18 and Appendix D

The measures below are proposed rather than standard; no industry body publishes any of them. Each can be computed from records a manager already keeps, and each separates leverage taken from process from leverage taken from people. A GP can run them on its own firm in a quarter. An LP can ask for them in a questionnaire.

Traceable numbers per analyst-hour. Take the last committee memo and the last quarterly report, count the material figures that meet the five-part test, and divide by the hours the team logged producing them. A loading firm’s ratio falls as assets per head rise. A re-engineering firm’s ratio rises, because source, date and path are captured in production rather than added afterwards.

Memo cycle time against the diligence window. The days from data room access to a committee-ready memo, set against the exclusivity period the seller granted. A firm under load either uses the whole window or trims the memo. A firm with the process engineered uses less of it and can show what it did with the remainder.

Reporting delivery against the ILPA clock. The share of quarters delivered within the 45-day target rather than the 60-day expectation, and the trend. This is the one measure with an external benchmark already written down.

Fee-calculation reproducibility. Whether an outside party can recompute the management fee, the offsets and the waterfall from the LPA and the fund statements alone and arrive at the figure the manager charged. The SEC’s 2022 deficiency list is the test set: fee base after write-downs, definitions of impairment, recycling. A manager that runs the test on itself finds what an examiner would.

Restatement rate. The share of reported figures corrected in a later period, by cause. Under load, restatements rise and their causes cluster in the skipped steps: a rent roll not reconciled, an amendment missed, a date assumed.

Set beside NAREIM’s own metrics, properties and income per asset manager and employees per billion of assets, these supply the denominator the survey lacks. Sixteen properties per head is a fact about volume. Sixteen properties per head at a stable restatement rate and a falling cost per traceable number is a fact about leverage. The industry publishes the first, and nobody yet publishes the second.

The leverage described here is built before it is bought: a data layer, a set of written tolerances and a review architecture in which whatever produces a figure is not what checks it, with software attached only once those exist. The order matters because software attached to an unreconciled rent roll and an undefined fee base reproduces the NAREIM pattern faster and with a worse error profile: more figures per head, produced sooner, none of them retraceable. The survey already shows adoption running ahead of governance, with 29% of firms naming underwriting and modelling as an AI priority while 42% have a formal technology committee. Tools without controls are loading by another means.

The software will be priced against the labour it displaces. If Bessemer’s portfolio figures are any guide, vendors will hold gross margins in the region of 56% and will move towards pricing tied to output, which makes the manager’s own throughput record the basis of every renewal negotiation. A firm that cannot state its cost per traceable number will pay a price set by the vendor’s estimate of its analyst-hours instead.

Nor will the leverage show up in the fee line. INREV’s 2023 finding was that management fees are consistent across vehicles while the other costs a vehicle carries disperse widely, so a manager’s cost discipline appears, if anywhere, in its expense ratio relative to peers and in the margin it retains on a benchmarked fee. The fee itself will not rise because the process improved. That is what permanent compression means: the price is set by the table, and the manager’s only claim on the difference lies on the cost side.

Judgment does not scale, and this paper should say so plainly. The decisions that scale with decisions rather than assets, to buy, to sell, to refinance, to accept a covenant waiver, remain the work of a small number of people whose hours are the firm’s scarcest input. Re-engineering the asset-scaling work does not multiply them; it returns their hours from checking to deciding. A firm that expects a data layer to make its investment committee wiser will be disappointed. One that expects it to stop the committee spending its time asking where a number came from will get what the layer can give.

The evidence is thinner than the argument, and the thin places should be named. No standards body, whether INREV, NCREIF, RICS or ILPA, benchmarks analytical output per head or the reproducibility of a manager’s process, so the unit of measure proposed in Section VI has no published distribution to be compared against. The load figures rest on 79 firms that chose to answer a survey. The examination findings concern private fund advisers at large. The software economics are one investor’s account of its own companies. What can be said with confidence is narrower: the fee has converged and is published; the load per person has risen without a change in process or a rise in profit; and allocators have already built the instruments to recompute the first of those and are within one template revision of asking about the second.