How to Scale Workspace Across Multiple Cities Without Opening Offices Everywhere

Most companies treat multi-city workspace as a sourcing problem: employees are in more places, so find space in more places. That framing is why workspace programs break at around the fourth or fifth market. The sourcing is the easy part. The hard part is that demand in each new market arrives small, arrives unevenly, and mostly stays small, while the number of vendors, invoices, and decisions grows every time someone asks for somewhere to work.

We went looking for the actual shape of that problem in our own data. Croissant has booked more than 4 million workspace hours since 2015. Within that, 287 company accounts have generated 71,395 bookings across 1,325 distinct workspaces in 126 markets and 50 countries, across 949 distinct company-market relationships. That last number is the interesting unit of analysis: not a company, not a city, but the relationship between a company and a market it has started using.

The three findings that should change how you plan

1. Every market you enter adds four to six buildings, not one. A company using a single market books at a median of 6 distinct workspaces. At 4–10 markets the median is 22; at 11 or more it is 71; above 20 markets it is 252. The ratio holds steady across the whole range, roughly four to six distinct workspaces per market, so the number of places to administer scales with markets rather than with demand.

2. The center of gravity holds far longer than people expect. From 3 markets all the way to 20, the median company still concentrates about 85% of its usage in its single largest market. Distribution does not redistribute demand; it adds a tail.

3. Most of that tail never matures. Of 883 company-market relationships old enough to observe for a full year, 52.9% were active in only one calendar month of that year. Only 17.7% reached six or more active months. Only 9.1% reached the equivalent of weekly usage by months seven to twelve.

How multi-city workspace actually becomes complicated

The complexity does not arrive as a strategic decision. It arrives as a queue of individually reasonable requests:

• "Can we get somewhere in Chicago?"
• "I need a meeting room in London next Thursday."
• "There are six of us in Austin now."
• "We're onboarding three people in Lisbon."
• "Can the team have an office two days a week?"

Each one is small enough to solve on its own, so each one gets solved on its own. A membership here, a direct operator relationship there, a day office for one team, an expensed meeting room, a reimbursement policy for everyone else. Nothing about any single decision is wrong. The accumulation is the problem.

Our data shows what that accumulation looks like from the inside. The median company operating in 4 to 10 markets is booking at 22 different workspaces. The median company in 11 or more markets is booking at 71. Half of all company-market relationships in the dataset involve exactly one person: 53.6% have never had more than a single employee using workspace in that market. A workspace program at that stage is not a real estate portfolio. It is a queue.

What 4 million booked hours say about geographic demand

We bucketed every company account by the number of distinct metro markets it has used, then measured how its demand is distributed. The table below covers the 198 company accounts with at least 10 bookings, so that percentages are not computed from a handful of visits.

Two things move together here, and one thing does not. The number of workspaces and the number of people booking climb steeply. Usage intensity per person does not: in a month when an employee books at all, the median is two to three days, in every cohort. Distributed companies do not consume workspace more intensively. They consume it in more places, through more people, at the same low frequency each.

That is the single most important planning fact in this dataset. A program that grows from 1 market to 11 is not absorbing more demand per person; it is absorbing more relationships per unit of demand.

Where the center of gravity finally breaks

Bucketing hides the shape of the curve, so we also measured market count one step at a time. The result was not the smooth decline we expected.

From 3 markets to 20, the largest market's share of usage sits in a flat band around 85%. It does not erode gradually. It holds, and then, past roughly 20 markets, it collapses to 44%. Only at that extreme do companies genuinely have no center of gravity.

For everyone below that threshold, and that is the overwhelming majority, the strategic picture is specific: you have one market that behaves like a headquarters and a long tail of markets that behave like nothing you have a process for. The mistake is applying headquarters logic to the tail.

The long tail is real, and it is thin

Across all 949 company-market relationships in the dataset:

• 37.5% of all company bookings happen outside the company's single largest market.
• 16.2% happen outside its top three markets.
• 53.7% of company-market relationships involve five bookings or fewer, ever. The median is five.
• 21.5% involve exactly one booking.
• 53.6% have never involved more than one person.

Read those together. A meaningful share of demand, more than a third of all bookings, sits outside the primary market. But it is spread across markets where the median relationship is five bookings by one person. There is no version of that pattern that a lease serves well. There is no version that a per-seat subscription serves well either, because a seat billed monthly for a person who needs three days a quarter is mostly a payment for nothing.

Not every city needs an office, and the data says which ones might

We took every company-market relationship whose first booking was at least twelve months before the end of the study window, 883 of them, and tracked what happened over the following year.

The median company-market relationship generated four bookings in its first year, concentrated in a single month. Just over half never came back after that month. Roughly a quarter were still alive at the end of the year, and fewer than one in ten had become anything a facilities team would recognize as recurring demand.

This is the finding that should govern multi-city workspace strategy. Most new-market workspace demand does not justify a lease, a dedicated office, or a local vendor relationship, because most of it does not survive its first quarter. The 9.1% that does survive is exactly the population worth committing real estate to, and it is identifiable in advance only if you are already capturing usage market by market.

There is a corollary worth stating plainly: the way to find out whether a city deserves an office is to serve the demand cheaply for two quarters and read the result. Committing first destroys the information you needed.

What workspace model fits each level of demand

Matching the model to the observed pattern, rather than to headcount or to ambition, is most of the work:

The meeting-room numbers support this progression rather than just asserting it. Only 8.5% of single-market companies have ever booked a meeting room. Among companies in 11 or more markets, 46.7% have. As teams disperse, the reason for gathering shifts from "somewhere to sit" to "somewhere to meet", which is a different product, in a different place, on a different cadence.

How to run this centrally

Once you accept that the problem is coordination rather than sourcing, the operating model follows. Six things need to be centralized, in this order:

1. Policy. What each employee or team is entitled to, by role and location, decided once, not per request.
2. Booking. Employees self-serve within policy. Every request routed through an admin is a request that will be asked again next month.
3. Sourcing. Inventory is available in the tail markets before demand appears, or the tail gets solved by expense reports.
4. Vendors. One relationship instead of one per market. This is where the 71-workspace median becomes tractable.
5. Billing. One invoice, coded by market, team, and person.
6. Reporting. Usage per market over time, the only input that tells you which of your markets is in the 9.1%.

Notice that the last one is not administrative hygiene. It is the decision input for every real estate commitment you will make in the next two years.

When to open an office

Based on what the maturation data shows, three conditions are worth requiring before committing space in a market:

• Six or more active months. Only 17.7% of company-market relationships reach this. It filters out the half that never return after month one.
• More than one person. 53.6% of relationships never exceed a single employee. A market driven by one person is a person's preference, not a market.
• Weekly-equivalent usage sustained through months 7–12. The 9.1%. This is the population where fixed space starts to beat on-demand space on cost per utilized hour.

Below all three, on-demand access is not a stopgap. It is the correct permanent answer for that market. Roughly nine out of ten of your markets will stay there.

Methodology and limits

Croissant has booked more than 4 million workspace hours since 2015: 611,226 booked sessions, each counted as a standard seven-hour workday. Logged attendance averaged 3.8 hours per session, so booked hours describe reserved capacity rather than time physically present. The company-level analysis covers Croissant company accounts from 2016 through August 2026: 287 accounts with any usage, 71,395 bookings, 309,402 logged hours, 949 company-market relationships. A market is a metro area, not a city boundary. Bookings include coworking day visits, guest visits, meeting room bookings, and private office bookings; cancelled bookings and sessions longer than 24 hours are excluded. Cohort tables are restricted to the 198 accounts with 10 or more bookings. Maturation covers the 883 relationships with at least 12 months of observable history. "Weekly-equivalent" means 24 or more bookings across months 7–12.

Three limits are worth stating. First, the distributed end of the sample is small: 15 accounts use 11 or more markets and 6 use more than 20, so those rows are directional rather than population-representative. Second, all figures in the tables are medians of companies, not usage-weighted averages, because usage-weighted figures in the 11+ cohort are dominated by its two largest accounts. Third, this measures workspace booked through Croissant; a company's leases and local arrangements outside the platform are not visible here, which means the fragmentation we can see is a floor, not a ceiling.

Individual, non-company member usage, which is the majority of the platform's 4 million booked hours, is excluded from every company-level figure above.