Workspace Flexibility Without Losing Control of Spend: The Risk Is Idle Seats, Not Power Users

The objection to workspace flexibility is always financial. If employees can book a desk wherever they are, whenever they want, what stops the bill from climbing? So companies reach for the instruments they trust: a flat stipend, an approval step, a per-seat plan with a fixed headcount. Each one is designed to contain a spender who turns out not to exist.

Across Croissant's company accounts we can see the whole chain, from who is eligible to book, to who actually books, to what each booking costs. That covers 71,395 company-attributed bookings by 3,207 employees at 287 companies. Two things in that data contradict the way most workspace budgets are built, and the second one is where the money actually goes.

How concentrated is workspace usage among employees?

Very. Among companies with at least ten active bookers, a median of 15.4% of active employees account for half of all bookings. The same measure applied to spend gives virtually the same answer.

Bookings are measured across 34 companies with ten or more active bookers, spend across the 25 of those with per-booking pricing. Loosen the threshold to five or more active bookers and the population grows to 72 companies with a median of 20%. The Gini coefficient on bookings is 0.6, which is the kind of inequality you would expect in income distribution rather than in an employee benefit.

Look at it per employee-year and the tail is just as clear. Of 5,901 employee-years of usage, more than half involve five bookings or fewer.

The heaviest 12.9% of employee-years generate 55.5% of bookings. The lightest 52.5% generate 10.3%. Any budget built on an average per employee is describing a population that is barely represented in its own data.

Does a workspace budget per employee mean anything?

Not as a planning number. Among employees who booked at all in a month, spend in US dollars distributes like this:

In both currencies the mean sits well above the median, which is what a long right tail does to an average. Inside a single company the spread is just as wide: across companies with at least twenty priced employee-months, the 90th percentile employee spends a median of 2.8 times the median employee. A budget set at the average overfunds most people and still rations the few who use workspace as a working pattern rather than an occasional convenience.

Why spend follows frequency almost exactly

Here is the finding that dissolves the runaway-spend fear. A workspace booking costs roughly the same amount whoever makes it. Heavy users are not choosing more expensive space, they are simply booking more often.

Unit cost is flat, and if anything it falls slightly with volume. The euro figures agree: €25.53, €26.63 and €22.24 per booking across the same three segments. The rank correlation between an employee's bookings and their spend is 0.82 in dollars and 0.88 in euros, which is about as tight as behavioural data gets.

That has a direct governance consequence. If unit cost is flat and spend tracks frequency, then controlling frequency controls spend, and you do not need to police where people go or what they book. A cap expressed in days per month is a budget. It is also a rule an employee can understand and act on without asking anyone, which is precisely what an approval step fails to deliver.

The bigger leak: eligible seats that never book

If concentration were the whole story, the fix would be to fund the heavy users properly and move on. But the larger financial hole is on the other side of the ledger, and it only becomes visible when you compare who is eligible to book against who books.

Across 6,138 company-months at 115 companies with at least five seats on the account, 68.1% of company-months saw no bookings at all. In the trailing twelve months that rises to 75.1%. This is not an onboarding problem: a median 92.3% of seats were fully onboarded. People had access. They did not use it.

Read the last two columns together, because they pull in opposite directions. Small accounts go dark for months at a time but engage a decent share of their people when they are live. Large accounts never go dark, yet in a typical month only 16.1% of their seats book anything. At that end, roughly one seat in six turns into a booking in a given month, and the other five are carried.

Every one of those idle seats is free under usage-based pricing and expensive under any per-seat arrangement. That is the actual failure mode, and it is the mirror image of the one companies guard against. The risk is not an employee spending too much. It is an organisation paying for standing capacity that a predictable majority of its people will not touch this month. It is the same mechanism that makes a city office look reasonable on headcount and indefensible on usage.

What buyers say about paying for space nobody uses

The behavioural data says idle capacity is the leak. Workplace and finance leaders describe the same thing in their own words, without prompting. These are illustrative quotes from recorded buyer conversations, not a measured sample, and they are included because they explain the pattern rather than prove it:

• "A lot of times we come way under because people just don't use it, but we plan on it."
• "My other concern is we spend and they don't use it."
• "We were paying for something that was not being used."
• "We put limits on it because it's more like a benefit." Paired, in the same conversation, with "we want to control how many times consultants go to the coworking space."
• "If I can't see usage by team and city, finance won't touch this."

Note what is absent. Across the conversations we reviewed, nobody described an employee overspending. The anxiety is about committed money going unused, and about not being able to see usage clearly enough to defend the program internally.

Do employees in secondary markets behave differently?

Yes, and not in the direction most expansion plans assume. Ranking each company's markets by lifetime bookings, usage thins out fast beyond the primary market, and it gets less habitual rather than more evenly spread.

Outside the primary market, usage is also more concentrated, not less: 6.1% of employees there account for half the bookings, against 7.5% in the primary market. So a company opening its fifth or tenth market should expect a handful of committed users and a long list of people who try it once, which argues for access everywhere and commitment almost nowhere. This is the employee-level counterpart to the way demand fragments as companies spread across cities.

The operating model this data supports

Company sets policy and budget, employee chooses inside it, usage is captured centrally, policy is adjusted against observed behaviour. The data tells you what each of those steps should contain.

Make eligibility broad and cheap. Since unit cost is flat and most people book rarely, restricting who is allowed to book buys you almost no savings and costs you the occasional traveller who genuinely needs a desk that week. Give access widely and pay for what gets used. Restricting eligibility is only a real lever when eligibility itself carries a per-seat fee, which is an argument for changing the pricing model rather than the access list.

Cap frequency, not choice. A limit of, say, eight booking days a month per person is a hard budget, because spend and frequency move together at 0.82 to 0.88 correlation. It leaves the choice of city, venue and day entirely with the employee, which is the part of flexibility they actually value. Policy rules that cap hours, restrict regions, or require approval only past a threshold put that ceiling in place before spend happens instead of discovering it in an expense report.

Set the cap above the median and below the tail. The median active employee books 2 days a month and the 90th percentile books 6. A cap set at the median will bind on your most productive users for no financial gain. Set it where it only catches genuine outliers, then review the outliers individually, because at 25 or more bookings a year a dedicated seat may honestly be cheaper for that person.

Stop reimbursing individually. Reimbursement produces receipts, not usage data, and it pushes the administrative load onto whoever approves it. It is also the mechanism least able to answer the question finance keeps asking, which is usage by team and by market.

How to tell whether the policy is working

Four measures, in the order they will tell you something:

1. Participation rate, the share of eligible seats that book in a month. Given that 68.1% of company-months in our data have none, this is the first number to watch and the one most programs never compute. Compare against 20% to 28% in an active month for accounts of a similar size.
2. Days per active employee per month. The median is 2 and the 90th percentile is 6. If your distribution is tighter than that, your cap is probably binding.
3. Concentration. Track the share of employees producing half your bookings. Near 15% is normal. Well below it suggests your program has become a small group's habit rather than a company capability.
4. Cost per booked day, by market. Because unit cost is flat, a rising cost per day is a supply or policy problem, not an employee behaviour problem. Usage and spend reporting by person, team, and market is what turns each of these into something you can put in front of finance.

Participation and frequency answer different questions and need different responses. Low participation is a demand or awareness problem and is fixed by communication, coverage, or by removing a per-seat fee. High frequency in a small group is not a problem at all until it exceeds what a dedicated seat would cost. Companies that forecast workspace as a variable people-cost rather than a fixed property line tend to separate the two by default.

Methodology and limits

This analysis covers company-attributed usage on Croissant: 71,395 bookings by 3,207 employees across 287 companies, from the first booking on record through 1 September 2026. A booking is one coworking day visit, guest visit, meeting room booking, or private office booking. Cancelled bookings and sessions longer than 24 hours are excluded. Units are stated explicitly and not mixed: company, employee, booking, employee-month, employee-year, and company-market.

Concentration is measured among active bookers, so it says nothing about employees who never booked. Those are handled separately through the eligibility figures, where seat counts come from weekly company account snapshots rolled up to the month by their maximum, and bookers are counted from bookings in that same month. Measuring participation per company-month rather than against a final snapshot matters: comparing a stale seat count with a twelve-month booker count produces participation rates above 100%, which is meaningless.

Spend covers only those bookings priced per booking, which is a minority of all bookings across the full history and a larger share in the trailing twelve months. Usage drawn from prepaid hour balances carries no per-booking price, so it appears in booking counts but not in spend. That gap is why cost per booking is reported on the priced subset with both numerator and denominator restricted to the same bookings. An earlier cut that compared each segment's share of spend against its share of bookings suggested occasional users were more expensive; that was an artefact of power users being priced less often through prepaid balances, and the unit-cost comparison above replaces it.

Spend is never summed across currencies. Concentration figures are shares within a single company and so need no conversion; absolute figures are reported separately in US dollars and euros. Market tiers are assigned from lifetime bookings, so a market that grew late in the window can rank lower than its current usage implies. Cohorts at the distributed end are small, and the trailing-twelve-month concentration cut rests on too few companies to carry an argument on its own, so the all-time figure across 34 companies is the one reported throughout.

Finally, all of this describes companies using Croissant, not companies in general. These are organisations that have already decided to buy flexible workspace centrally, which is a selected population. Employees who also have a company office available may be using it on the days they do not appear here, so these figures describe demand for flexible workspace rather than total workplace attendance.